Handbook of Methods in Leadership Research-Edward Elgar Publishing (2017)

HANDBOOK OF METHODS IN LEADERSHIP
RESEARCH
SCHYNS_9781785367274_t.indd 1
10/11/2017 15:19
HANDBOOKS OF RESEARCH METHODS IN MANAGEMENT
Series Editor: Mark N.K. Saunders, University of Birmingham, UK
This major series will provide the starting point for new PhD students in business
and management and related social science disciplines. Each Handbook will give
definitive overviews of research methods appropriate for particular subjects within
management. The series aims to produce prestigious high-quality works of lasting
significance, shedding light on quantitative, qualitative and mixed research methods.
Each Handbook consists of original contributions by leading authorities, selected
by an editor who is a recognized international leader in the field. International in
scope, these Handbooks will be an invaluable guide to students embarking on a
research degree and to researchers moving into a new subject area.
Titles in the series include:
Handbook of Research Methods on Intuition
Edited by Marta Sinclair
Handbook of Research Methods on Human Resource Development
Edited by Mark N.K. Saunders and Paul Tosey
Handbook of Research Methods on Trust
Second Edition
Edited by Fergus Lyon, Guido Möllering and Mark N.K. Saunders
Handbook of Qualitative Research Methods on HRM
Innovative Techniques
Edited by Keith Townsend, Rebecca Loudoun and David Lewin
Handbook of Methods in Leadership Research
Edited by Birgit Schyns, Rosalie J. Hall and Pedro Neves
SCHYNS_9781785367274_t.indd 2
10/11/2017 15:19
Handbook of Methods in
Leadership Research
Edited by
Birgit Schyns
NEOMA Business School, France and Durham University
Business School, Durham University, UK
Rosalie J. Hall
Durham University Business School, Durham University, UK
Pedro Neves
Nova School of Business and Economics, Portugal
HANDBOOKS OF RESEARCH METHODS IN MANAGEMENT
Cheltenham, UK • Northampton, MA, USA
SCHYNS_9781785367274_t.indd 3
10/11/2017 15:19
© Birgit Schyns, Rosalie J. Hall and Pedro Neves 2017
All rights reserved. No part of this publication may be reproduced, stored
in a retrieval system or transmitted in any form or by any means, electronic,
mechanical or photocopying, recording, or otherwise without the prior
permission of the publisher.
Published by
Edward Elgar Publishing Limited
The Lypiatts
15 Lansdown Road
Cheltenham
Glos GL50 2JA
UK
Edward Elgar Publishing, Inc.
William Pratt House
9 Dewey Court
Northampton
Massachusetts 01060
USA
A catalogue record for this book
is available from the British Library
Library of Congress Control Number: 2017947087
This book is available electronically in the
Business subject collection
DOI 10.4337/9781785367281
ISBN 978 1 78536 727 4 (cased)
ISBN 978 1 78536 728 1 (eBook)
06
Typeset by Servis Filmsetting Ltd, Stockport, Cheshire
SCHYNS_9781785367274_t.indd 4
10/11/2017 15:19
Contents
List of contributorsvii
Acknowledgementsix
PART I
1
INTRODUCTION
Introduction and overview
Birgit Schyns, Pedro Neves and Rosalie J. Hall
PART II
2
3
MEASUREMENT AND DESIGN
Implicit measures for leadership research
SinHui Chong, Emilija Djurdjevic and Russell E. Johnson
13
3Puppet masters in the lab: experimental methods in leadership
research48
Eric F. Rietzschel, Barbara Wisse and Diana Rus
4
ssessing leadership behavior with observational and
A
sensor-based methods: a brief overview
Alexandra (Sasha) Cook and Bertolt Meyer
73
5
he contribution of sophisticated facial expression coding to
T
leadership research
Savvas Trichas
103
6
Behavioral genetics and leadership research
Wen-Dong Li, Remus Ilies and Wei Wang
7 Biosensor approaches to studying leadership
Aurora J. Dixon, Jessica M. Webb and Chu-Hsiang (Daisy)
Chang
PART III
8
127
146
UANTITATIVE METHODS AND ANALYTIC
Q
APPROACHES
ediation analysis in leadership studies: new developments
M
and perspectives
Rex B. Kline
173
v
SCHYNS_9781785367274_t.indd 5
10/11/2017 15:19
vi Handbook of methods in leadership research
9Person-oriented approaches to leadership: a roadmap forward
Roseanne J. Foti and Maureen E. McCusker
195
10
Multi-level issues and dyads in leadership research
Francis J. Yammarino and Janaki Gooty
229
11
A social network approach to examining leadership
Markku Jokisaari
256
12
Diary studies in leadership
Sandra Ohly and Viktoria Gochmann
296
13
odeling leadership-related change with a growth curve
M
approach317
Rosalie J. Hall
PART IV
14
UALITATIVE METHODS AND ANALYTIC
Q
APPROACHES
ualitative content analysis in leadership research: principles,
Q
process and application
Jan Schilling
15 Biographical methods in leadership research
Miguel Pina e Cunha, Marianne Lewis, Arménio Rego and
Wendy K. Smith
PART V
349
372
SUMMARY
16
Leadership in the future, and the future of leadership research
Robert G. Lord
403
17
Authors’ tips for doing top-quality research
430
Index439
SCHYNS_9781785367274_t.indd 6
10/11/2017 15:19
Contributors
Chu-Hsiang (Daisy) Chang, Michigan State University, USA
SinHui Chong, Michigan State University, USA
Alexandra (Sasha) Cook, Chemnitz University of Technology, Germany
Miguel Pina e Cunha, Nova School of Business and Economics,
Portugal
Aurora J. Dixon, Michigan State University, USA
Emilija Djurdjevic, University of Rhode Island, USA
Roseanne J. Foti, Virginia Tech, USA
Viktoria Gochmann, University of Kassel, Germany
Janaki Gooty, University of North Carolina at Charlotte, USA
Rosalie J. Hall, Durham University Business School, Durham University,
UK
Remus Ilies, National University of Singapore, Singapore
Russell E. Johnson, Michigan State University, USA
Markku Jokisaari, University of Turku, Finland, and Durham University
Business School, Durham University, UK
Rex B. Kline, Concordia University, Canada
Marianne Lewis, Cass Business School, City University of London, UK
Wen-Dong Li, Chinese University of Hong Kong, Hong Kong
Robert G. Lord, Durham University Business School, Durham University,
UK
Maureen E. McCusker, Virginia Tech, USA
Bertolt Meyer, Chemnitz University of Technology, Germany
Pedro Neves, Nova School of Business and Economics, Portugal
Sandra Ohly, University of Kassel, Germany
vii
SCHYNS_9781785367274_t.indd 7
10/11/2017 15:19
viii Handbook of methods in leadership research
Arménio Rego, Católica Porto Business School, Universidade Católica
Portuguesa, and Instituto Universitário de Lisboa (ISCTE-IUL), Business
Research Unit, Portugal
Eric F. Rietzschel, University of Groningen, the Netherlands
Diana Rus, Creative Peas Consultancy and University of Groningen, the
Netherlands
Jan Schilling, University of Applied Administrative Sciences, Hannover,
Germany
Birgit Schyns, NEOMA Business School, France, and Durham University
Business School, Durham University, UK
Wendy K. Smith, University of Delaware, USA
Savvas Trichas, Open University of Cyprus and CDA College, Cyprus
Wei Wang, University of Central Florida, USA
Jessica M. Webb, Michigan State University, USA
Barbara Wisse, University of Groningen, the Netherlands, and Durham
University Business School, Durham University, UK
Francis J. Yammarino, State University of New York at Binghamton,
USA
SCHYNS_9781785367274_t.indd 8
10/11/2017 15:19
Acknowledgements
We are indebted to the external reviewers for this book who gave their
time generously to improve the chapters of this book. Our thanks go to:
Talib AlHinai, Durham University Business School, Durham University,
UK
Susanna Chui, Durham University Business School, Durham University,
UK
Shahira Dahari, Durham University Business School, Durham University,
UK
Alexandra Hauser, Ludwig Maximilian University of Munich, Germany
Steve Lockey, Durham University Business School, Durham University,
UK
Ekaterina Pogrebtsova, University of Guelph, Canada
Micah Roediger, Virginia Tech, USA
Dean Rosenwald, University of Pittsburgh; SolarCell Design, LLC, USA
Thomas Sasso, University of Guelph, Canada
Maria Joao Velez, Nova School of Business and Economics, Portugal
Kenneth Wenk, University of Pittsburgh, USA
Xiatong (Janey) Zheng, Durham University Business School, Durham
University, UK
Yuyan (Cherry) Zheng, London School of Economics, UK
ix
SCHYNS_9781785367274_t.indd 9
10/11/2017 15:19
SCHYNS_9781785367274_t.indd 10
10/11/2017 15:19
PART I
INTRODUCTION
SCHYNS_9781785367274_t.indd 1
10/11/2017 15:19
SCHYNS_9781785367274_t.indd 2
10/11/2017 15:19
1.
Introduction and overview
Birgit Schyns, Pedro Neves and
Rosalie J. Hall
This volume provides an overview of a variety of established and newer
methods for leadership research. It is intended for any individuals
wanting to undertake research on leaders, whether they are academics or
practitioners, undergraduates, graduate students, or new or established
professionals. Our goal in this volume is to help leadership researchers
obtain a first insight into a specific method and its potential application
to leadership research, so that they can make a decision about whether
or not to delve deeper into the method and use it for their own research.
We particularly encourage academics who want to try a new method and
graduate students who are just starting their own research programs to
read this book. This book may also prove helpful to individuals who want
to better understand and assess the quality and implications of leadership
research undertaken by others.
It was interesting when collecting the chapters to see the great variety
of methods applied in leadership research, all of which contribute to a
more complete and nuanced understanding of the leadership process.
Recent editorials from, for example, the Academy of Management Journal
(Colquitt, 2013), stress the increasing breadth of methodological design
and analysis techniques used (e.g., from inductive/qualitative to experimental research). Chen (2015) similarly highlights the breadth of designs
that are covered in the Journal of Applied Psychology as well as the necessity for the method to clearly fit the research question (see also Edmondson
& McManus, 2007). As Rico (2013), former editor of the European Journal
of Work and Organizational Psychology, described it: “[A]fter reading a
Methods section, the reader needs to be able to understand what was done
and why this approach was selected” (pp. 2–3). We hope to add to that
increasing variety by making more techniques accessible to scholars and
by discussing when and how each method is more pertinent.
In addition, Green, Tonidandel, and Cortina (2016) analysed reviewers’ comments on submissions to the Journal of Business and Psychology
and found that a large number of reviewer comments referred to method/­
analysis issues such as mediation/moderation or issues with multi-level
models and regression analysis. These issues are commonly linked to
3
SCHYNS_9781785367274_t.indd 3
10/11/2017 15:19
4 Handbook of methods in leadership research
rejection (though to a lesser degree than measurement issues) but are also
involved in decisions to ask for “revise and resubmit.” Our volume could
be helpful in making decisions about the appropriateness of design and
methods for particular research questions in leadership.
There were many potential techniques to choose from, and we could
not cover all in a single volume. Thus, we have focused on methods,
techniques and analytic approaches that are either currently valuable
for understanding leadership/followership or that we believe would – if
adopted – provide useful tools in the future. In addition, we have worked
with our authors and reviewers to keep the presentation of the techniques
as straightforward as possible, providing a detailed enough overview to
get readers started with a technique, but avoiding overly technical details.
Indeed, we have been very fortunate to be able to get many of our contributions from authors who have expertise in, and have published on, the
topic of leadership in organizations. This means that the descriptions of
methodological techniques in this volume are often embedded with recent
and helpful illustrations that are particularly relevant to the study of leadership. However, we have also included a handful of other authors who
are not leadership researchers per se, but contribute with their methodological expertise and rigor in areas where leadership researchers “could do
better.” All authors have been asked to provide detailed guidance on the
use of their featured techniques, including in many cases access to online
aids and sample datasets. Our hope is that this volume may be particularly
helpful not only in helping with the “how to” of a given approach, but also
as a source of answering “why?”
In this introductory chapter, we provide a brief overview and structural
framework for the following chapters. Our ultimate goal is to motivate
researchers both to try new techniques and to fine-tune their use of more
common techniques. We are in a time when new methodologies are being
developed, and old ones improved and made easier to use. Finally, we
hope that instructors of leadership courses aimed at advanced undergraduate and graduate students will consider using all or parts of our book
alongside a more content-oriented text, in order to better demonstrate the
varied and creative options in the “how to” of leadership research.
This handbook is divided into three main areas: Part II: Measurement
and Design; Part III: Quantitative Methods and Analytic Approaches;
and Part IV: Qualitative Methods and Analytic Approaches. The book
finishes in Part V with a summary/conclusion chapter and a chapter on
tips for leadership researchers from each author.
Part II of the book focuses on measurement and design issues, and consists of six chapters. First, in Chapter 2, Chong, Djurdjevic, and Johnson
provide an introduction to implicit measures for leadership research. They
SCHYNS_9781785367274_t.indd 4
10/11/2017 15:19
Introduction and overview ­5
argue that to understand leadership, both explicit elaborate and implicit
automatic processes are relevant. Most leadership research has focused on
the former, though this will only capture part of what influences judgments
and behaviors. Chong et al. present a taxonomy of implicit measures and
outline what these are measuring. They review the extant leadership literature that has used implicit approaches and provide examples of the use of
implicit measures. This chapter will help leadership researchers to make a
better decision on whether or not their research would benefit from using
implicit measures, and, if so, how to decide which types to use.
In Chapter 3, Rietzschel, Wisse, and Rus introduce experimental
methods in leadership research. They argue that experimental research can
help researchers to further understand causal relationships when studying
leadership. The authors provide an overview of varieties of experimental
methods that are currently used in leadership research, including both the
benefits of using these methods and potential issues involved in conducting experiments. We believe that a reading of this chapter will convince
leadership researchers that experiments can help them draw the stronger
causal conclusions that (most) field studies do not allow, as well as provide
them with possible ideas about designs for experiments.
Chapter 4 by Cook and Meyer outlines how observational and sensorbased methods can be used in leadership research. Their argument builds
around the idea that leadership is a process and that observational
methods allow for a better understanding of how leaders and followers
influence each other with their respective behavior. Cook and Meyer
describe different observation study approaches, including how to capture
and record relevant behavior and how to analyse observational data.
Their chapter outlines the advantages as well as the problems related to
conducting observational studies. Readers should learn from this chapter
why observational methods are useful in capturing the leadership process
in its entirety, and what ways of capturing and analysing observational
data are available.
Following on from that, Trichas explains in Chapter 5 the benefits of
sophisticated facial expression coding to leadership research. So far, this
method has not been widely used in leadership research. However, Trichas
uses examples of studies that have used and others that have not used
(but examined similar research questions) sophisticated facial expression
coding to highlight how this approach can further our understanding of
the leadership process. After reading the chapter, leadership researchers
should understand the value of sophisticated facial expression coding, for
example, when studying emotions in leadership as well as knowing how to
use this method.
Li, Ilies, and Wang focus on behavioral genetics approaches in Chapter
SCHYNS_9781785367274_t.indd 5
10/11/2017 15:19
6 Handbook of methods in leadership research
6. They argue that genetic approaches can help leadership researchers to
disentangle the nature/nurture arguments prevalent in leadership research.
Introducing twin studies and a molecular genetic research approach as
useful lenses through which to study leadership, they explain the contribution of those approaches as well as review examples of their application in
leadership research so far. This chapter provides the reader with an overview of genetic approaches, sample studies, and application in leadership
research, and an understanding of how this type of research can advance
our knowledge of leadership nature versus nurture.
In Chapter 7, Dixon, Webb, and Chang introduce biosensor approaches
to studying leadership. They begin by describing four general categories of
biosensor methods (i.e., involving collection of bodily fluids, cardiovascular activity, brain activity, and genetics/evolutionary characteristics) and
their associated advantages and disadvantages for leadership research.
This is followed by a review of previous biosensor-related research that
contributes to our knowledge of leadership and followership, using Bass’s
(2008) three general approaches for understanding leadership (leadercentric, leadership as an effect, and leader–follower interactions) as an
organizing structure. Thus, this chapter gives readers a broad overview
of the possibilities for integrating biosensor approaches with more traditional research methods.
Part III of the book focuses on quantitative methods and analytic
approaches. Here, six chapters explain how to analyse different types
of quantitative data. First, Kline in Chapter 8 outlines methods for the
analysis of mediation in leadership research. He particularly focuses on
common misunderstandings around this topic, bringing into question the
value of the results obtained using mediation in some studies. Issues comprise, amongst others, theoretical assumptions about directionality and
design problems. After reading the chapter, leadership researchers should
understand these issues and be able to design better studies to test mediation and/or to acknowledge more clearly the limits of their m
­ ediation
testing.
In Chapter 9, Foti and McCusker introduce person-oriented approaches
to leadership. This alternative to more commonly used variable-oriented
approaches provides researchers with the possibility to focus on “types”
or “patterns” based on individuals. Foti and McCusker review different
methods within person-oriented approaches and illustrate them using
examples from leadership research. This chapter provides the reader
with sample research questions that require the use of person-oriented
approaches, as well as an understanding of the approaches available to
analyse data in this novel way.
Chapter 10 by Yammarino and Gooty focuses on multi-level issues
SCHYNS_9781785367274_t.indd 6
10/11/2017 15:19
Introduction and overview ­7
and dyads in leadership research. Very often in the study of leadership,
researchers deal with so-called nested data – that is, followers nested
in leaders or teams, or dyadic data such as leader–follower dyads. This
chapter provides an overview of the issues as well as methods relating to
dyadic data. Here, the authors discuss in depth three analytic approaches
that might be applied to dyadic leadership data. The reader of this chapter
will gain a better understanding of the logic of nested and dyadic data collected in leadership research and how this type of data can be meaningfully
analysed.
Following on from this, in Chapter 11 Jokisaari introduces a social
network approach to leadership research. He argues that when studying leadership, we often ignore that leadership happens in networks of
individuals. He outlines methods to address this issue and uses examples
to illustrate how this approach can be applied. This includes issues of
research design, sampling and data collection, methods to measure social
networks, and central measures of networks for use in data analysis.
Reading this chapter will provide the reader with a clear idea of how and
when to apply network methods, and for which types of research questions
they can be useful tools.
In Chapter 12, Ohly and Gochmann focus on diary studies in leadership. Here, they argue that leadership is often assessed at one point in time
but that diary studies provide the opportunity to better understand the
process involved in leadership. This also includes testing the notion that
leadership is a stable phenomenon. Ohly and Gochmann review existing
diary studies and outline approaches that can be used as well as issues
that researchers might face. In identifying relevant research questions,
the authors provide the reader with guidelines on when and how to apply
diary studies in leadership research.
Finally, in Chapter 13, Hall explains the value of applying latent growth
curve modeling to leadership research. She points out that in a substantial
amount of leadership research, dynamics and change are important and
latent growth curve modeling is an appropriate method to examine these
changes, allowing both the identification of average change and for variability in change across individuals. After reading this chapter, the reader
should be familiar with the technique, knowledgeable about issues and
alternative analytic choices, and be able to reflect on appropriate designs
as well as make a reasoned choice of software options.
Part IV of the book summarizes some qualitative methods and analytic
approaches to leadership research. The two chapters in this part outline
when and how to use qualitative approaches to leadership. We appreciate that this part of our volume is short in comparison to the previous
two, and acknowledge that it can only scratch the surface of qualitative
SCHYNS_9781785367274_t.indd 7
10/11/2017 15:19
8 Handbook of methods in leadership research
methods in leadership research. The reader is encouraged to consult additional overview books in this area.
First, in Chapter 14, Schilling provides an extensive overview of qualitative content analysis and how it can be applied to leadership research. He
argues that qualitative approaches to leadership are somewhat undervalued currently. His chapter gives a comprehensive overview on qualitative
content analysis as a systematic, rule-based process of analysing verbal
and textual data (e.g., interviews, group discussions, documents). As he
presents the process of qualitative content analysis step by step, and develops guidelines for researchers’ decision and action, his chapter provides
the reader with a “how to” guide for approaching qualitative data in the
leadership area.
Following on from that, in Chapter 15, Cunha, Lewis, Rego, and Smith
outline the use of biographical methods in leadership research. They
define biographical methods as an umbrella term comprising approaches
such as self-narratives, autobiographies, and historical biographies that
explore an individual’s life story to elucidate nuanced dynamics over time.
They describe the features of these methods, namely, narrative, holistic,
constructivist, context-sensitive, dynamic and temporally situated, relational, self-reflexive, and contradiction-sensitive. Their chapter also provides the reader with a clear idea of how to collect biographical data and
the method involved in sampling and analysing data. The chapter includes
a discussion of potential issues pertaining to biographical methods.
Finally, in Part V, Chapter 16, Robert Lord provides an outlook for
the future of leadership research. In his chapter, he aims to improve
leadership theory, methodology, and practice both in the short and the
long term, including an assumption of radical change in how leadership
research is approached. He addresses issues such as theory proliferation
and ­aggregation across entities and time. In addition to improving theory,
he describes how to improve the measurement of leadership and study
designs. He concludes with an outlook on the potential future of leadership research.
At the end of this handbook, in Chapter 17 the authors provide some
handy tips for leadership researchers based on material covered in their
chapters. The reader is encouraged to turn to this chapter for a summary
of the key points the authors consider a “must know.”
In conclusion, we hope that the readers of this book enjoy the chapters
as much as we enjoyed gathering, reading, and editing them. Hopefully,
this book will motivate you to understand and apply new methods in your
leadership research.
SCHYNS_9781785367274_t.indd 8
10/11/2017 15:19
Introduction and overview ­9
REFERENCES
Bass, B.M. (2008). The Bass handbook of leadership: Theory, research and managerial applications. New York: Simon and Schuster.
Chen, G. (2015). Editorial. Journal of Applied Psychology, 100(1), 1–4. Retrieved from http://
www.apa.org/pubs/journals/apl/
Colquitt, J.A. (2013). The last three years at AMJ – Celebrating the big purple tent. Academy
of Management Journal, 56(6), 1511–1515. doi: 10.5465/amj.2013.4006
Edmondson, A.C., & McManus, S.E. (2007). Methodological fit in management field research.
Academy of Management Review, 32(4), 1155–1179. doi: 10.5465/AMR.2007.26586086
Green, J.P., Tonidandel, S., & Cortina, J. (2016). Getting through the gate: Statistical and
methodological issues raised in the reviewing process. Organizational Research Methods,
19(3), 402–432. Retrieved from http://journals.sagepub.com/doi/abs/10.1177/10944281166
31417?journalCode5orma
Rico, R. (2013). Editorial letter: Publishing at EJWOP. European Journal of Work and
Organizational Psychology, 22(1), 1–3. Retrieved from http://dx.doi.org/10.1080/13594
32X.2013.752247
SCHYNS_9781785367274_t.indd 9
10/11/2017 15:19
SCHYNS_9781785367274_t.indd 10
10/11/2017 15:19
PART II
MEASUREMENT AND
DESIGN
SCHYNS_9781785367274_t.indd 11
10/11/2017 15:19
SCHYNS_9781785367274_t.indd 12
10/11/2017 15:19
2.
Implicit measures for leadership research
SinHui Chong, Emilija Djurdjevic and
Russell E. Johnson
INTRODUCTION
Attitudes influence judgments and behaviors in the workplace through
explicit and implicit processes (Bowling & Johnson, 2013; Uhlmann et al.,
2012). Explicit processing occurs when individuals engage in deliberative
and effortful analysis of the costs and benefits of a decision and behavior,
while implicit processing occurs automatically and outside of individuals’
awareness (Chaiken & Trope, 1999). Despite being useful and practical,
explicit measures that assess conscious work attitudes and behaviors
are frequently associated with response biases that may undermine the
validity of research findings (Johnson & Tan, 2009). These concerns
have prompted organizational researchers to turn to implicit measures
in the hope of more accurately capturing work attitudes and behaviors,
especially those driven by automatic processes that reside outside people’s awareness and control (Uhlmann et al., 2012). For example, Ziegert
and Hanges (2005) demonstrated that implicit racist attitudes and racist
climate interacted to positively predict employment discrimination against
racial minority applicants, and Stajkovic, Locke, and Blair (2006) showed
that the implicit activation of a do-your-best attitude in individuals led to
better goal performance than no activation. These findings highlight the
promise that implicit measures hold for organizational research.
In this chapter, we focus on the use of implicit measures in leadership
research. Leadership scholars are among the earliest to acknowledge the
importance and value of implicit processes. The most prominent is the
implicit theory of leader categorization, which argues that individuals rely
on their implicit expectations and prototypes of personality traits, instead
of veridical leadership behaviors in the workplace, to define and categorize others as leaders (Eden & Leviatan, 1975; Rush, Thomas, & Lord,
1977). Scholars also used implicit measures, such as sentence completion
tasks, to examine employees’ attitudes toward their supervisors (Burwen,
Campbell, & Kidd, 1956) and individuals’ motivation to lead (Stahl,
Grigsby, & Gulati, 1985). However, beyond these initial works, our review
revealed that the use of implicit measures in leadership research is quite
13
SCHYNS_9781785367274_t.indd 13
10/11/2017 15:19
14 Handbook of methods in leadership research
rare. This trend is baffling, especially in a time when implicit measures are
gaining increasing popularity and credibility in organizational research
(Uhlmann et al., 2012). The failure of leadership research to adopt implicit
measures creates an inadequacy to capture leadership-related phenomena
that operate at levels below individuals’ consciousness, and constitutes a
critical methodological concern.
We believe that leadership scholars’ inertia in adopting implicit measures is likely due in part to uncertainty about what implicit measures are,
when to use them, and how to administer them. Therefore, our chapter
aims to shed light on the utility of implicit measures for leadership
research, and to produce clear and actionable knowledge for scholars to
incorporate implicit measures in their works. We organize our chapter
into six parts to achieve these objectives. First, we review implicit theories
of leadership. Second, we highlight the rationales and possible advantages
of using implicit measures for leadership research. Third, we present an
established taxonomy of implicit measures, describe the assumptions they
are based on, and explain the types of variables and processes they are
designed to capture. We also use this taxonomy to organize and review
existing leadership studies that have utilized implicit measures. Fourth, we
propose a set of criteria to guide leadership scholars in their selection of
appropriate implicit measures, which depends on their research question
and focal constructs. Fifth, we present an empirical example of a leadership study that employed implicit measurement. Finally, we conclude by
offering ideas for using implicit measures in leadership research.
LEADERSHIP THEORIES THAT INVOLVE IMPLICIT
CONTENT OR PROCESSES
Existing research supports a dual process model in which explicit and
implicit processes operate in parallel to influence attitudes and behaviors
(Chaiken & Trope, 1999). In explicit processing, individuals engage in
conscious and effortful analysis in order to form judgments and decide
how to behave (Fazio & Olson, 2003). Explicit attitudes and behaviors are
assumed to be deliberative enough to be captured by self-report or otherreport methodologies (Uhlmann et al., 2012). For example, leadership
scholars often have followers self-report their exchange quality with their
leaders (Liden & Maslyn, 1998), and rate their leaders’ leader behaviors
(Liden, Wayne, Zhao, & Henderson, 2008; Tepper, 2000). In contrast,
implicit processes influence attitudes and behaviors in an automatic, unintentional, and sometimes even unconscious manner (Greenwald & Banaji,
1995; Wegner & Bargh, 1998). Leadership scholars generally assume
SCHYNS_9781785367274_t.indd 14
10/11/2017 15:19
Implicit measures for leadership research ­
15
that implicit processes reflect an experiential system that is slow learning
but fast acting (Lord, Diefendorff, Schmidt, & Hall, 2010; McClelland,
McNaughton, & O’Reilly, 1995). This means that individuals gradually
learn and develop cognitive schemas about the attributes that characterize ideal leaders through socialization and life experiences (Lord, Brown,
Harvey, & Hall, 2001). These schemas reside in individuals’ memory and
function as a cognitive bias that spontaneously influences how individuals
make sense of various aspects of leadership (Kenney, Schwartz-Kenney, &
Blascovich, 1996). Implicit theories prove to be useful in explaining automatic processes in leadership-related phenomena, and leadership scholars
have applied implicit theories to the following topics.
Leader Categorization
People make sense of the world by segmenting the environment into categories, where they classify stimuli perceived to be similar into the same
category (Mervis & Rosch, 1981). Categories have unique representative
symbols; for example, vehicles are mobile and primarily function to transport people or goods between places, while birds are feathery animals that
lay eggs to reproduce. Categories serve important information-processing
purposes by helping people simplify abundant and complex stimuli in
the environment (ibid.). Building on this notion, Lord, Foti, and De
Vader (1984) proposed a categorization theory to leadership. This theory
argues that individuals develop leader prototypes, defined as abstract
conceptions of the most representative members of the category of leaders
(Rosch, 1978; Rush & Russell, 1988), from their prior experiences and
social interactions with leaders. These prototypes are easily accessible
and rapidly activated, and they help individuals distinguish leaders from
non-leaders based on the perceived match between the target and their
prototypical attributes. Research in this area demonstrated individuals’
tendency to define individuals who are intelligent and honest (Rush &
Russell, 1988), white (Gündemir, Homan, De Dreu, & Van Vugt, 2014), or
male (Rudman & Kilianski, 2000), as leaders. Individuals also form different male and female leader prototypes, where they expect female leaders to
demonstrate sensitivity and strength and male leaders to demonstrate only
strength (Johnson, Murphy, Zewdie, & Reichard, 2008).
Leadership Motives
Scholars have long been interested in how basic motives such as the need
for power, need for affiliation, and need for achievement predict behavioral outcomes (Atkinson, 1958). Need for power refers to the desire to
SCHYNS_9781785367274_t.indd 15
10/11/2017 15:19
16 Handbook of methods in leadership research
impact and control others’ behaviors or emotions; need for affiliation
refers to the desire to build lasting relationships with others; and need for
achievement refers to the concern to compete and perform better than
others (ibid.). Motives such as the need for power and achievement positively predict leadership effectiveness (Locke, 1991). However, self-report
and implicit measures of motives rarely correlated significantly because
people are not always consciously aware of their own motives or may
conceal their real motives for social desirability purpose (McClelland,
Koestner, & Weinberger, 1989). According to McClelland and colleagues
(1989), implicit motives reflect a primitive motivation system that predicts
spontaneous and affect-charged responses while explicit motives reflect
thoughtful cognitive constructs and predict socially reinforced responses
in structured situations (McClelland, et al., 1989). These arguments have
prompted scholars to study the influences of both explicit and implicit
motives on leadership outcomes (Spangler, 1992). These studies generally
supported a positive relationship between an implicit power motive and
leadership outcomes (De Hoogh et al., 2005; Jacobs & McClelland, 1994).
Leader Evaluation
Implicit theories also serve as a framework for understanding how followers evaluate the effectiveness of leaders. Studies in this area showed
that individuals rely on performance cues (i.e., successful or unsuccessful)
to evaluate leaders’ effectiveness, where individuals exposed to successful
performance cues were more likely to rate a leader as being considerate and exhibiting more structuring types of behaviors than individuals
exposed to unsuccessful performance cues (Larson, 1982; Lord, Binning,
Rush, & Thomas, 1978). Scholars attributed this phenomenon to the
perceptual-memory bias, where implicit schemas function as perceptual
filters that guide individuals’ attention to schema-consistent behaviors as
they evaluate leaders’ effectiveness (Rush et al., 1977).
Implicit Self-concepts and Leadership
Implicit self-concepts have been studied in leadership research in two key
ways. First, scholars have used implicit theories to explain how the activation of a certain self-identity in individuals predicts their leadership selfconcepts. For example, exposure to non-traditional roles such as a female
engineer (i.e., successful women in a male-dominated career) reduces
women’s implicit beliefs about how well they can lead (Rudman & Phelan,
2010). This is consistent with the backlash effect, which argues that examples of successful atypical women (i.e., female engineer) provokes upward
SCHYNS_9781785367274_t.indd 16
10/11/2017 15:19
Implicit measures for leadership research ­
17
comparison and causes perceived threat of competence among average
women (Parks-Stamm, Heilman, & Hearns, 2008). In another program
of research, scholars have developed conceptual models to explain how
leaders influence the implicit values and self-concepts of followers (Lord
& Brown, 2001). These conceptual works are built on the assumption
that leaders can increase the salience of certain values and hence activate
certain self-identities (i.e., individualistic, collectivistic, or relational identity) among followers through task design, communication patterns, and
leader behaviors (Brewer & Gardner, 1996; Hanges, Lord, & Dickson,
2000; Johnson & Lord, 2010; Lord & Brown, 2001).
The above review illustrates some of the areas where implicit theories
have been applied in leadership research, and highlights the potential of
implicit theories and measures to expand leadership theory. In the next
section, we discuss the rationales and possible advantages of using implicit
measures in leadership research.
ADVANTAGES OF IMPLICIT MEASURES FOR
LEADERSHIP RESEARCH
The primary objective of implicit measures is to obviate conscious processing and to capture automatic, unintentional, and/or unconscious processes
underlying judgments and behaviors (De Houwer, Teige-Mocigemba,
Spruyt, & Moors, 2009; Fazio & Olson, 2003). This is achieved through
limiting individuals’ ability to control their responses and minimizing their
awareness of what is being assessed (Uhlmann et al., 2012). There tends
to be an explicit aspect to all responses, however, such that individuals
can possibly still control their automatic responses to implicit measures
(Conrey, Sherman, Gawronski, Hugenberg, & Groom, 2005). This suggests that responses to implicit measures are likely reflecting the joint
influences of both implicit and explicit processes, and it may not be possible to completely tease apart the two types of processes when assessing
cognition and behavior (ibid.). Nevertheless, implicit measures capture a
larger proportion of implicit processes as compared to explicit measures
(Uhlmann et al., 2012). Existing meta-analyses also found that implicit
measures and explicit measures only modestly correlate with each other
(Hofmann, Gawronski, Gschwendner, Le, & Schmitt, 2005; Krizan &
Suls, 2008), thus supporting the incremental utility of implicit measures.
Fazio and Olson (2003) contended that how much implicit and explicit
measures correlate really depends on individuals’ motivation, opportunity, and ability to deliberate on their responses. If individuals have
little motivation, opportunity, and ability to deliberate, their responses
SCHYNS_9781785367274_t.indd 17
10/11/2017 15:19
18 Handbook of methods in leadership research
to explicit measures will correlate higher with their responses to implicit
measures.
Building on the above notion, implicit measures are most useful for
studying phenomena that are theorized to operate partially or mostly
below individuals’ awareness or that are not easily observed or introspected by individuals themselves (Uhlmann et al., 2012). For example,
individuals are rarely aware of how their prejudices or stereotypes
influence their definition and categorization of leaders and non-leaders
(Chung-Herrera & Lankau, 2005; Rudman, Ashmore, & Gary, 2001). In
such cases, using implicit measures addresses the limitation posed by using
explicit measures to capture implicit processes.
The use of implicit measures also helps to overcome instances when
individuals are unwilling to report their attitudes truthfully due to social
desirability concerns (Uhlmann et al., 2012). For example, individuals
may not want to admit that their motivation to lead stems from their
desire for power, which may sometimes be perceived as a manipulative,
insincere, and undesirable motivation. With the help of implicit measures,
leadership scholars fare better at getting to such underlying motives that
individuals would otherwise conceal in their responses to explicit measure.
In addition, some implicit measures have been demonstrated to overcome
faking because they are designed to elicit spontaneous responses that are
difficult to consciously control (LeBreton, Barksdale, Robin, & James,
2007). This makes them especially useful for overcoming the influence of
evaluation apprehension on responses, such as when assessing followers’
satisfaction with their leader. Furthermore, individuals sometimes engage
in self-deception when responding to explicit measures (Uhlmann et al.,
2012). This may happen when individuals adjust their explicit attitudes
from their implicit attitudes in the face of cognitive dissonance in order
to protect their self-esteem (Taylor, 1983). For example, research demonstrated that exposing female participants to successful female leaders
in male-dominated careers threatened their self-esteem and resulted in
negative implicit evaluations of female leaders but had no effects on
explicit evaluations of female leaders (Rudman & Kilianski, 2000). When
conducting research in such sensitive domains, scholars would especially
benefit from the adoption of implicit measures.
In addition, implicit measures often predict incremental variance in
criteria above and beyond corresponding explicit measures (Greenwald
& Banaji, 1995; Johnson, Tolentino, Rodopman, & Cho, 2010). This is
beneficial when it is important to explain unique variance in the criterion
(Uhlmann et al., 2012). For example, Johnson and Saboe (2010) demonstrated that followers’ implicit self-concept positively predicted the quality
of leader–member exchanges and subsequent follower outcomes such as
SCHYNS_9781785367274_t.indd 18
10/11/2017 15:19
Implicit measures for leadership research ­
19
task performance and citizenship behavior above and beyond explicit
measures of self-concept. Such findings imply that scholars and organizations will benefit from incorporating implicit measures in their research
in order to more comprehensively uncover and understand factors that
influence leadership criteria.
Taken together, these strengths of implicit measures underpin their usefulness for leadership research. To provide a comprehensive view of how
implicit measures have been used in extant leadership studies, we present a
taxonomy of implicit measures and use it to organize and review existing
leadership research that have used implicit measures.
A TAXONOMY OF IMPLICIT MEASURES AND
THEIR USE IN EXISTING LEADERSHIP RESEARCH
Implicit measures can be organized in numerous ways (Fazio & Olson,
2003; Uhlmann et al., 2012). For example, implicit measures can be
distinguished based on the administration format (e.g., computer-based
reaction time measures vs pen-and-paper questionnaires). While such
methodological properties are important, we present a taxonomy based
predominantly on the conceptual differences between the measures. This
taxonomy, developed by Uhlmann and colleagues (2012), classifies implicit
measures into three categories based on the type of implicit content they
are designed to capture. Accessibility-based measures assess the degree to
which a target concept is spontaneously activated in individuals’ minds
(e.g., leadership behaviors may activate and make certain self-concepts
more salient in followers). Association-based measures determine whether
several targets are linked as part of a cognitive schema in individuals’ memories (e.g., associating intelligence to leadership effectiveness).
Interpretation-based measures evaluate the reactions to and inferences
that individuals draw from ambiguous stimuli to uncover inner attitudes
(e.g., assessing leadership motives through leaders’ responses on how they
would react in given situations). Although interpretation-based measures
are designed based on the concepts of accessibility and association, they
serve to capture deeper and more complex worldviews or motives that the
simpler interfaces of accessibility-based or association-based measures are
inadequate in eliciting (ibid.).
This taxonomy has the strength of clustering implicit measures that
have substantive conceptual similarities but are being administered and
delivered differently (ibid.). For example, both word completion tasks
and lexical decisions tasks are accessibility-based measures that assess
how salient a target concept is in individuals’ minds, but the former is a
SCHYNS_9781785367274_t.indd 19
10/11/2017 15:19
20 Handbook of methods in leadership research
pen-and-paper task while the latter is a reaction-time task. On the other
hand, although implicit association task and lexical decision task are
both computer-based reaction-time tasks, they tap into different types of
implicit information where the former assesses implicit linkages between
multiple target concepts while the latter assesses the ease of accessibility
of a single target concept. We contend that classifying measures based on
their conceptual purpose and properties instead of their administration
features offers a straightforward means for leadership scholars to determine which measures are more suitable for the purpose of their research.
Extant findings that demonstrated significant convergence between scores
from accessibility-based tasks (Johnson & Lord, 2010) further support the
basis of this classification approach.
Accessibility-based Implicit Measures
Accessibility-based measures assess whether a single target concept is
activated and accessible in individuals’ minds (Uhlmann et al., 2012). The
activation of concepts can be trait or state based (ibid.). Trait-based activation exists when a target concept enjoys a generally high level of accessibility over time and across situations (Robinson & Neighbors, 2006).
For example, the concept of “unpleasant” is more readily accessible and
will be recognized more quickly by individuals high in negative affectivity
(Johnson et al., 2010). Alternatively, state-based activation exists when
a target concept becomes more accessible for a limited period of time
(Robinson & Neighbors, 2006). For example, the concept of “service” may
be more accessible when employees are interacting with customers versus
leaders. Both trait- and state-based activations of a target concept influence how individuals make judgments and behave (ibid.). Accessibilitybased implicit measures include word or sentence completion tasks, lexical
decisions tasks, and Stroop tasks. Below, we describe these measures and
give examples of how they have been used in leadership research.
Word- or sentence-completion tasks
Participants are presented with word or sentence fragments that they
have to complete to form a complete word or sentence. Word fragments
are designed in a manner where they could be completed to form a word
that either reflects a target construct or neutral non-target words. For
example, the fragment “_OY” can be completed to form “JOY” (i.e., a
word that reflects positive affectivity as the target construct) or “SOY”
(i.e., a neutral non-target word). Sentence completion tasks operate in a
similar way, but instead of forming words, participants form sentences by
choosing one or more words from a set of options. Among the options are
SCHYNS_9781785367274_t.indd 20
10/11/2017 15:19
Implicit measures for leadership research ­
21
words that reflect the target concept and other non-target concepts, and
participants are instructed to select the option that best reflects the first
thought that comes to mind. Implicit scores, which reflect the proportion
of target words generated or selected, are used to infer how accessible the
target concept is in participants’ minds (Koopman, Howe, Johnson, Tan,
& Chang, 2013).
Sentence completion tasks appear to be the most commonly used
accessibility-based implicit measure among leadership researchers. Several
existing leadership studies have utilized the Miner Sentence Completion
Scale-H (MSCS-H; Miner, 1978) to assess motivation to lead (Ebrahimi,
1997; Miner & Smith, 1982; Miner, Chen, & Yu, 1991; Stahl et al., 1985).
The MSCS-H presents respondents with 35 incomplete sentences assessing
six dimensions of motivation to lead (e.g., the desires to compete, assert
oneself, exercise power, etc.). The options are worded to reflect high (+1),
neutral (0), or low (–1) motivation to lead, and respondents have to select
an option that best reflects their first thought to complete the sentence.
The scores are aggregated to derive individuals’ motivation to lead scores,
with higher scores representing a stronger motivation. Researchers have
also used sentence completion tasks to study attitudes toward supervisors. In an early study, Burwen and colleagues (1956) used a sentence
completion task to assess Air Force cadets’ attitudes toward their leaders
by asking them to complete sentences with their first thoughts on how
a cadet would feel or behave in different scenarios with his [sic] leader.
An example item is “Whenever he saw his superior coming, he (1) threw
up, (2) ducked, (3) saluted, (4) gave him a warm greeting, or (5) was very
happy.” Cadets’ attitudes toward their leaders were scored and computed
from their selected options to all the sentences. In another study, Konst,
Vonk, and Van der Vlist (1999) asked employees to read and complete
sentences describing performance-related behaviors of eight people (e.g.,
“J did not succeed in reaching a decision. . .”). When participants were
told that these people were leaders (as opposed to subordinates), they were
more likely to make inferences about the behavior’s consequences for the
environment when completing the sentences. This suggests that participants viewed leaders (vs non-leaders) as having greater influence over their
environments (Konst et al., 1999).
Lexical decision tasks
Lexical decision tasks require participants to use a computer keyboard
to quickly decide whether a string of letters is an actual word or a nonsensical non-word (Meyer & Schvaneveldt, 1971). Out of all the real
words presented, half of them reflect the target concept (e.g., “abuse” or
“reprimand” reflect the concept of abusive supervision), and half of them
SCHYNS_9781785367274_t.indd 21
10/11/2017 15:19
22 Handbook of methods in leadership research
are neutral, non-target words. Scholars have previously validated neutral
words based on the number of letters and frequency of use in English
with validated references such as the English Lexicon Project (Balota
et al., 2007) and the frequency dictionary of contemporary American
English (Davies & Gardner, 2013). Following up on the example, “brush”
and “anonymous” can serve as the neutral counterparts of “abuse” and
­“reprimand,” respectively, based on their number of letters and their frequency of use in English. After participants complete the decision task,
scholars compute the average within-person response time for correctly
identifying words that reflect the target concept to ascertain how accessible the target concept is to participants (faster times represent higher
accessibility). Researchers interested in using lexical decision tasks can
download software and modifiable scripts from developer websites.1
Our review identified only one leadership study that used a lexical
decisions task. In this study, Scott and Brown (2006) found that participants recognized communal trait words (e.g., kind, helpful) more readily
than agentic trait words (e.g., dominating, independent) after reading a
­sentence describing a female (vs male) leader.
Stroop tasks
A final example of an accessibility-based implicit measure is the Stroop
task. In traditional Stroop tasks, participants view color words (e.g.,
“BLUE,” “RED”) with congruent or incongruent font colors on the
screen, and are required to name the font color instead of reading the
word (Stroop, 1935). For example, participants presented with the word
“BLUE” in red font color should answer “red” instead of “blue.” The
Stroop task is designed to capture how much cognitive interference participants experience during the task. When interference is low, participants
find it easier to focus on the font color instead of the meaning of the word.
Over the past decades, Stroop tasks have been modified to assess interference that result from the meaning of non-color words too. The key idea is
that, if the presented word is highly accessible to participants at implicit
levels, participants will find it more difficult to ignore the word and focus
on the font color. Hence, longer response times reflect greater accessibility
of the target concept. For example, followers experiencing abusive supervision may take longer to name the font color of the word “ABUSE” as
compared to the font color of a neutral counterpart such as “BRUSH.”
This is because leaders’ abusive behaviors will be more accessible to them,
and the word “ABUSE” will interfere with their ability to focus and name
the font color. Researchers can download software and modifiable scripts
for Stroop tasks online.2
The use of the Stroop task for leadership research is not common, as
SCHYNS_9781785367274_t.indd 22
10/11/2017 15:19
Implicit measures for leadership research ­
23
only one published study has used it. In this early study, Megargee (1969)
investigated how well leaders with high versus low trait dominance could
convey information to subordinates under time pressure. They used the
original Stroop task where color words were presented in congruent or
incongruent font colors, and found that leaders with high trait dominance
experienced less interference and were able to more accurately convey the
correct font color to subordinates as compared to leaders with low trait
dominance (ibid.).
Association-based Implicit Measures
The second class of implicit measures is association-based measures,
which assess how multiple target concepts are spontaneously linked in
individuals’ memories (Uhlmann et al., 2012). They make the key assumption that activating a single target concept will trigger spreading activation to nearby related concepts in individuals’ semantic networks (Collins
& Loftus, 1975). Association-based measures are especially useful for
uncovering attributes that individuals associate with a given category, for
example, when individuals think about leadership, they associate with it
traits such as intelligence and honesty (Lord et al., 2001). Priming tasks
and implicit association tests are examples of association-based implicit
measures.
Priming
Priming measures attempt to activate target concepts by presenting participants with specific priming stimuli without their awareness (Fazio &
Olson, 2003). Priming typically involves embedding the priming stimulus
in filler tasks or in the physical environment, such that participants are not
consciously aware of the exposure (De Houwer, et al., 2009). It is assumed
that the target concept activated by the priming stimulus will influence
individuals’ subsequent judgments and behaviors. Interested researchers
can gain access to several types of priming tasks online.3
Priming measures appear to be a rather popular choice of implicit
measure among leadership scholars. For example, Larson (1982) primed
participants with information on whether a group had succeeded or
failed in its task before showing them a video of an interaction between
a leader and team members. Participants primed with the success information rated the leader’s effectiveness higher than those primed with the
failure information because success is associated with leader effectiveness
(ibid.). In another study, Baldwin, Carrell, and Lopez (1990) conducted
an experiment where they primed participants with the disapproving face
of their department chair or of another person, and demonstrated that
SCHYNS_9781785367274_t.indd 23
10/11/2017 15:19
24 Handbook of methods in leadership research
­ articipants who were primed with the disapproving face of their departp
ment chair rated their own research ideas lower as a result. Finally, Latu,
Mast, Lammers, and Bombari (2013) primed female participants with
photos of female leaders (e.g., Hillary Clinton, Angela Merkel) outside
their awareness, and found that these female participants experienced
greater empowerment and performed better on a subsequent leadership
task compared to female participants primed with photos of male leaders.
Scholars have also used priming measures to study stereotypical associations with female leaders (Davies, Spencer, & Steele, 2005; Rudman &
Phelan, 2010).
Implicit association tests
Implicit association tests (IATs) present participants with pairs of words,
where words representing a target concept (e.g., “leader”) are paired
with either a positive concept word (e.g., “good”) or a negative concept
word (e.g., “bad”). Participants have to use the computer keyboard to
respond to the word pairs, where they are supposed to press a particular
key when the target concept word appears with a positive concept word
(e.g., “leader” and “good” appear together on the screen), and another
key when the target concept word appears with a negative concept word
(e.g., “leader” and “bad” appear together on the screen). When participants respond quickly to a word pair, it suggests that the two words presented in the pair are more strongly associated in memory (Greenwald,
McGhee, & Schwartz, 1998). For example, if participants responded
more quickly to “leader + good” as compared to “leader + bad,” it
implies that participants have a positive attitude toward the leader.
Researchers interested in IATs can obtain information about creating
and using IATs online.4
The IAT is commonly used to study stereotypical traits or attributes that individuals associate with different categories of leaders such
as male versus female leaders (Dasgupta & Asgari, 2004; Rudman &
Kilianski, 2000; Rudman & Phelan, 2010), or white versus minority
leaders (Gündemir et al., 2014). For example, Rudman and Kilianski
(2000) demonstrated that participants responded more quickly to pairs of
words linking men with high authority roles and women with low authority roles. Nevertheless, Dasgupta and Asgari (2004) showed that letting
participants read biographical information about model female leaders
before they completed an IAT reduced their likeliness of associating
women with low authority roles. Schoel, Bluemke, Mueller, and Stahlberg
(2011) also used the IAT to study the attitudes of high or low self-esteem
individuals on autocratic leadership and democratic leadership, and found
that high self-esteem individuals associated democratic leadership with
SCHYNS_9781785367274_t.indd 24
10/11/2017 15:19
Implicit measures for leadership research ­
25
positive valence while low self-esteem individuals associated autocratic
leadership with positive valence.
Interpretation-based Implicit Measures
Interpretation-based implicit measures comprise the final category.
Interpretation-based implicit measures capture individuals’ responses
to ambiguous stimuli (Uhlmann et al., 2012). These measures assume
that individuals project their chronically accessible personality, values,
attitudes, or motives when they are trying to make sense of ambiguous
information (ibid.). Thus, individuals’ responses are expected to be indicative of their latent personality, values, or attitudes (Tomkins & Tomkins,
1947). For example, when interpreting how a leader should behave in a
team, individuals with a strong power motive are likely to find it reasonable for leaders to exert strong control over followers (James et al., 2012).
Examples of interpretation-based implicit measures include thematic
apperception tests and conditional reasoning tests.
Thematic apperception test
The thematic apperception test (TAT) requires participants to interpret
pictures depicting ambiguous situations and narrate a story about what
is happening in each picture (Morgan & Murray, 1935). Researchers then
code the responses to infer participants’ social motives such as the needs
for power, affiliation, and achievement. Previous research has found that
TAT measures of achievement motives predict work outcomes such as
career success and managerial success (McClelland & Boyatzis, 1982;
Spangler, 1992).
The TAT appears to be the only interpretation-based measure that has
been used for leadership research. Leadership studies typically used the
TAT to measure leadership motives such as the need for power, affiliation, and activity inhibition (House, Spangler, & Woycke, 1991; Jacobs
& McClelland, 1994; McClelland & Boyatzis, 1982). For example, Jacobs
and McClelland (1994) administered the TAT to entry-level staff and
derived their scores for power, affiliation, and activity inhibition. They
found that employees with a high need for power, a low need for affiliation, and a high activity inhibition were the most likely to reach senior
management positions 12 years later for both male and female managers. In addition to leadership motives, Winter (1991) developed a coding
scheme to code TAT responses for responsibility. In his study, he found
that TAT measures of the need for power and responsibility collectively
predicted leadership success 16 years later.
SCHYNS_9781785367274_t.indd 25
10/11/2017 15:19
26 Handbook of methods in leadership research
Conditional reasoning test
The TAT relies on coding of qualitative responses to assess a target construct, which raises concerns of validity and reliability (Lilienfeld, Wood,
& Garb, 2000). However, not all interpretation-based measures face this
problem. A couple of interpretation-based measures employ standardized
scoring. One of these is the conditional reasoning test (CRT), which captures the reasoning process that individuals undergo to justify behavioral
choices (James, 1998; James et al., 2012). The CRT is administered to
participants as a cognitive ability test that contains several descriptions
of situations that participants have to make logical inferences from. Each
situation contains two response options that make sense and seem plausible (out of four), and one of these two options is designed to appear
reasonable only to those with a specific motive (e.g., achievement motive).
Participants’ responses to the situations are then scored to represent the
target concept (e.g., achievement motive). The CRT has several strengths.
First, the objectively scored items overcome the reliability concerns that
plague the TAT (Uhlmann et al., 2012). Second, the CRT is typically
robust to faking attempts as long as the purpose of the measure is not
disclosed (LeBreton et al., 2007). In addition, well-validated CRTs are
available for several constructs related to leadership. For example, CRTs
measuring power or achievement motives (James et al., 2012) may be
useful for examining leadership motives, and CRTs assessing aggression
(James et al., 2005) may be appealing for scholars interested in abusive
supervision. Despite this, our search failed to find any leadership studies
that used the CRT.
As gleaned from our discussion above, some implicit measures are more
commonly used than others for leadership research. We contend that the
lag in the adoption of certain implicit measures is due in part to researchers’ uncertainty about how to modify and adapt them to assess variables
of interest. Therefore, in the next section, we offer some advice to guide
leadership researchers in their selection and use of implicit measures for
research.
CRITERIA TO GUIDE APPROPRIATE USE
OF IMPLICIT MEASURES FOR LEADERSHIP
RESEARCH
According to Uhlmann and colleagues (2012), researchers ought to consider the following eight questions when choosing an implicit measure for
their research:
SCHYNS_9781785367274_t.indd 26
10/11/2017 15:19
Implicit measures for leadership research ­
27
1. Do I need an implicit measure?
2. Which category of implicit measure should I use?
3.Can the implicit measure be flexibly adapted to assess my construct of
interest?
4. What do the scores represent?
5. Does the implicit measure predict my outcome of interest?
6. Is the implicit measure reliable?
7. Is the implicit measure adaptable across cultures?
8. What resources on implicit measures are available for my study?
These questions apply to leadership researchers too, and below we finetune these recommendations to specifically help leadership scholars decide
how to incorporate implicit measures in their research.
Do I Need an Implicit Measure?
Based on our discussion earlier, there are several reasons and advantages
for using implicit measures. Leadership scholars should use implicit measures if they are interested in (a) assessing a construct that operates outside
an individuals’ awareness, or (b) understanding the associations among
constructs in connectionist memory. They may also consider using implicit
measures for the purposes of reducing the influences of evaluation anxiety
or social desirability on responses, and predicting incremental variance
beyond explicit measures.
Which Category of Implicit Measure Should I Use?
After ascertaining the need for an implicit measure, leadership researchers
should select a category of implicit measures that match how the construct
is conceptualized in their research questions. If they are interested in
capturing what comes to individuals’ minds automatically in naturalistic
settings, they should use accessibility-based implicit measures like the
word/sentence-completion tasks, lexical decision tests, or Stroop tasks.
For example, Scott and Brown (2006) used a lexical decision task in their
study, and demonstrated that communal leadership traits (i.e., being kind)
came to mind more readily than agentic traits (i.e., dominating) when
­individuals were evaluating a female leader.
If researchers are primarily interested in assessing the connections
between multiple constructs at an implicit level, they should find association-based measures such as priming and the IAT more useful. For
example, Schoel and colleagues (2011) used an IAT to show that followers with lower self-esteem were more likely to associate autocratic
SCHYNS_9781785367274_t.indd 27
10/11/2017 15:19
28 Handbook of methods in leadership research
leadership with positive valence and democratic leadership with negative valence, while followers with higher self-esteem were more likely
to ­associate democratic leadership with positive valence and autocratic
leadership with negative valence. This class of measures is especially useful
for detecting individual differences in implicit attitudes and beliefs about
specific leaders or general leadership style. However, it should be noted
that stimuli representing concepts are often greatly simplified in priming
or IAT tasks for the ease of presenting these stimuli on the computer
interface (Uhlmann et al., 2012). For example, Schoel and colleagues
(2011) represented autocratic leadership and democratic leadership with
five words related to autocracy and five words related to democracy
respectively. They obtained these words from a pilot test. This means that
pilot tests have to be conducted to validate representative stimuli for the
targeted constructs, and that these measures may not be well suited for
capturing attitudes toward nuanced and complex targets. Also, the IAT is
based on the relative comparisons between multiple target constructs in a
2 × 2 fashion typically (e.g., autocratic vs democratic leadership, and positive vs negative valence). Yet, symmetrical comparisons that fit this matrix
may not always be available.
Finally, leadership scholars may consider interpretation-based measures
if they hope to assess complex social beliefs and motives. Most interpretation-based measures are partially structured, and allow complex thoughts
to be captured (Uhlmann et al., 2012). When administering open-ended
qualitative measures like the TAT, leadership scholars can adapt the
coding scheme to capture multiple constructs of interest from participants’
responses. For example, Jacobs and McClelland (1994) administered the
TAT to entry-level managers and coded their responses for their need for
power, need for achievement, need for affiliation, and activity inhibition.
They also developed a scheme to further code the power motives into
themes of reactive power and resourceful power, which illustrates the
­malleability of interpretation-based measures.
Can the Implicit Measure be Flexibly Adapted to Assess My Construct of
Interest?
Next, leadership scholars have to consider whether and how they are going
to adapt and administer the selected measure in their research. Implicit
measures may offer greater flexibility than explicit measures (Uhlmann
et al., 2012). Accessibility-based implicit measures (e.g., word-completion
tasks, lexical decision tasks) and association-based implicit measures (e.g.,
priming, IAT) are relatively easy to modify to measure other constructs.
This process is usually straightforward, and involves finding alternative
SCHYNS_9781785367274_t.indd 28
10/11/2017 15:19
Implicit measures for leadership research ­
29
stimuli, such as word fragments for word completion tasks or representative words for IATs, to replace the original stimuli (see Scott & Brown,
2006 for an example of a modified lexical decision task, and Schoel et al.,
2011 for an example of a modified IAT). Scripts for administering these
tasks in computer software are also readily available online.5 Nevertheless,
researchers should always conduct pilot tests to validate the new stimuli
and to make sure that the replacement of the stimuli does not compromise
on the psychometric properties of the task. In contrast, the modification
of interpretation-based implicit measures is less straightforward because
researchers have to create new scenarios or vignettes or develop new
coding schemes in order to measure a different construct. This process is
extremely laborious, and may take numerous pilot tests and several years
to complete (Uhlmann et al., 2012).
What Do the Scores Represent?
After leadership researchers modify the implicit measure to assess their
construct of interest, they should make sure they understand what the
scores of the focal measure represent. For example, in an IAT, quicker
reaction times represent a stronger association between two constructs,
but in a Stroop task, quicker reaction times indicate lower interference and
weaker accessibility of the targeted construct. A clear understanding of
what the scores represent will allow leadership scholars to draw accurate
conclusions from their research findings.
Does the Implicit Measure Predict My Outcome of Interest?
Next, leadership scholars should ask whether their selected implicit
measure predicts their outcome of interest. Existing studies demonstrated
mixed evidence regarding the predictive validities of implicit measures
(Lilienfeld et al., 2000; Uhlmann et al., 2012). In general, our review of
existing leadership studies that have used implicit measures illuminates
the promise that implicit measures hold for leadership research. For
example, priming task success or failure influences ratings of leadership effectiveness (Larson, 1982) and priming followers with approving
or disapproving leader faces also influences their evaluation of their
research ideas (Baldwin et al., 1990). In addition, IATs seem to exhibit
strong predictive validities on a wide array of leadership-related associations such as white ethnicity with high leadership potential and minority
ethnicity with low leadership potential (Gündemir et al., 2014), men with
high authority and women with low authority (Rudman & Kilianski,
2000), and autocratic leadership with positive valence among low-esteem
SCHYNS_9781785367274_t.indd 29
10/11/2017 15:19
30 Handbook of methods in leadership research
individuals and democratic leadership with positive valence among highesteem individuals (Schoel et al., 2011). Interpretation-based measures
such as the TAT also exhibit robust predictive validity on several behavioral outcomes (Spangler, 1992). However, it should also be noted that
some measures, specifically accessibility-based measures such as sentence
completion tasks, have typically been used as dependent measures of constructs such as motivation to lead (Miner et al., 1991), leader adjustment
(Fitzsimmons & Marcuse, 1961), and attitude toward supervisor (Burwen
et al., 1956; Konst et al., 1999) instead of as predictors. This again suggests the need for leadership researchers to select an implicit measure
based on the conceptualization of their construct of interest to ensure that
the measure is able to appropriately assess what the researcher intends to
capture.
Is the Implicit Measure Reliable?
After addressing the questions above, researchers should consider whether
their selected implicit measure is reliable internally and temporally because
this ultimately has implications for the accuracy of research findings.
For implicit measures involving quantitative scoring such as the CRT,
researchers can assess internal and temporal reliability by calculating the
internal consistency and test-retest reliability (Uhlmann et al., 2012). To
assess the internal reliability of implicit measures that rely on reaction
times such as the Stroop task and IAT, researchers can separate the task
into multiple trials or blocks, and calculate the extent to which the reaction times are consistent across the trials or blocks (e.g., see Greenwald et
al., 1998 for an example of assessing the internal reliability of the IAT). In
contrast, it is more difficult to assess the internal reliabilities of interpretation-based implicit measures and it is common for interpretation-based
measures to exhibit lower internal and test-retest reliabilities than their
explicit counterparts (Uhlmann et al., 2012). Relatively lower internal and
test-retest reliabilities may have arisen in part due to participants’ perceived need to respond differently to each image or scenario presented in
an interpretation-based task, and to provide a different response each time
they participate in the task (Winter & Stewart, 1977). However, low testretest reliabilities may also indicate that the measure is assessing a state
instead of a stable trait (Blair, 2002). We encourage leadership researchers to refer to well-established reliability information reported in existing
literature to help them select the most reliable measure out of their chosen
category of implicit measures, and to conduct further empirical tests if
they encounter low internal or test-retest reliabilities with their selected
implicit measures.
SCHYNS_9781785367274_t.indd 30
10/11/2017 15:19
Implicit measures for leadership research ­
31
Is the Implicit Measure Adaptable Across Cultures?
Many leadership studies have non-English-speaking samples (Farh &
Cheng, 2000; Liden, 2012), which requires translating and back-­translating
explicit measures into parallel forms in different languages (Brislin,
Lonner, & Thorndike, 1973). However, there are unique challenges associated with adapting implicit measures into other languages (Uhlmann
et al., 2012). For example, it may be difficult or sometimes impossible to
find equivalent words that represent the construct of interest in another
language for use in an IAT (Brislin et al., 1973). In addition, implicit
measures such as the word completion task are not feasible in certain languages (e.g., Chinese, Korean) due to the different presentation of words
in these languages (i.e., where words are not formed from alphabetical
letters). However, there are some measures that have proven adaptable
across cultures, such as interpretation-based measures that rely on images
or vignettes (Uhlmann et al., 2012). Nevertheless, in these cases, researchers still have to expend considerable effort into modifying and validating
their coding or scoring schemes to take into account cultural differences
in interpretation and expressions. We advise leadership scholars to review
existing literature to gain a better understanding of the cross-cultural
generalizability of existing findings related to their constructs of interest.
This may help to inform them of how to develop a culturally appropriate coding scheme that allows them to capture their constructs of interest
accurately in another culture.
What Resources on Implicit Measures are Available for My Research?
The administration of some implicit measures requires the use of computers and specific software (Uhlmann et al., 2012). Researchers should then
make sure that computers are available to their sample and make arrangements to install the necessary software in the computers. This can be a
problem if leadership scholars are conducting field studies where respondents complete the implicit measure in their workplace. If the organization
does not allow researchers to install the software to administer the implicit
measure, researchers may have to abort their plans of using those implicit
measures that are administered through a computer (e.g., IAT) and rely on
measures that can be administered on pen and paper (e.g., word/sentencecompletion tasks, TAT, or CRT). Nevertheless, while resources may
shape the decision of which implicit measure to use, leadership researchers
should always ensure that the theoretical appropriateness of the selected
implicit measure is not compromised.
In the next section, we describe an empirical example in which a lexical
SCHYNS_9781785367274_t.indd 31
10/11/2017 15:19
32 Handbook of methods in leadership research
decision task (LDT) was used to examine how priming individuals to think
about leadership influenced the accessibility of implicit leader-related
attributes, conducted by Djurdjevic and Johnson (2009).
AN EMPIRICAL EXAMPLE: ASSESSING
LEADERSHIP SCHEMAS USING AN IMPLICIT
MEASURE
According to leader categorization theory, individuals associate certain
attributes with leaders through prior experiences and social interactions
with leaders, and these attributes are activated automatically to help individuals distinguish leaders from non-leaders (Lord et al., 1984). Although
the theory is founded on implicit leadership theories, scholars have mostly
tested it using explicit measures (e.g., participants rate how representative
attributes are of leaders; Epitropaki & Martin, 2004; Lord et al., 1984;
Offermann, Kennedy, & Wirtz, 1994). Doing so creates a disconnect
between theory and method because implicit leadership theories posit that
leader categorization operates at an implicit level, yet explicit measures
require participants to engage in deliberative information processing as
they access, search, and retrieve leader-related attributes from memory.
Djurdjevic and Johnson’s (2009) empirical example addresses this oversight by using an implicit measure to assess the accessibility of leaderrelated attributes.
The use of an implicit measure in this case offers three key advantages. First, using an implicit measure to test implicit leadership theory
creates a match between theory and method because the implicit measure
enables us to capture individuals’ implicit schemas with little or no biased
responding due to social desirability or impression management concerns
(Fazio & Olson, 2003). Second, Djurdjevic and Johnson’s (2009) findings
shed light on the extent to which implicit and explicit measures of leader
schemas converge. If discrepancies exist, it would imply that leader-related
attributes derived from studies that used explicit measures might not be
an accurate reflection of individuals’ implicit schemas. Third, Djurdjevic
and Johnson’s (2009) findings may help reconcile conflicting findings
from past research that used explicit measures to study implicit leader
schemas. For example, certain attributes, such as masculinity and dominance, were found to be prototypical of leaders by some researchers (e.g.,
Lord et al., 1986; Offermann et al., 1994) yet anti-prototypical by others
(e.g., Epitropaki & Martin, 2004). Such discrepancies may be the result of
response biases.
Djurdjevic and Johnson (2009) developed several hypotheses based on
SCHYNS_9781785367274_t.indd 32
10/11/2017 15:19
Implicit measures for leadership research ­
33
implicit theories of leader categorization. First, they hypothesized that
participants primed with the concept of leadership (vs those in a control
condition) would respond more quickly on the LDT to leader-related
prototypical words associated with intelligence, dynamism, and dedication
(H1), but more slowly to anti-prototypical words associated with tyranny
(H2). This is because priming individuals to think about leadership should
activate the prototypical attributes they associate with leaders, thus eliciting greater accessibility and faster recognition of these words (Lord et al.,
2001). In contrast, priming leadership should suppress anti-prototypical
attributes, resulting in less accessibility and slower recognition of these
words (ibid.). Djurdjevic and Johnson (2009) also predicted that participants primed with the concept of leadership (vs control) would respond
more quickly to the attribute words “visionary,” “innovative,” “resourceful,” and “flexible” (H3). Although these attributes are not included on
explicit measures of leadership schemas, they were expected to be relevant
because effective leaders foresee and adapt to changing environmental
demands and provide guidance for others (Bass, 1985; Zaccaro, Foti, &
Kenny, 1991). Last, Djurdjevic and Johnson (2009) also examined whether
attributes associated with masculinity, dominance, and sensitivity would
be more or less accessible when participants were primed with the concept
of leadership. Earlier leader categorization studies found that masculinity and dominance are prototypical leader attributes (Lord et al., 1984;
Offermann, et al., 1994), whereas more recent studies classified them as
anti-prototypical (Epitropaki & Martin, 2004, 2005). These mixed findings
might reflect a change in the concept of leadership over time (e.g., expecting
leaders to be more communal) or possibly a method artifact (e.g., pressure
on participants to respond in an egalitarian manner; Eagly & Karau, 2002).
Djurdjevic and Johnson (2009) also hypothesized that familiarity with
leadership roles, need for cognition, and participants’ gender would moderate the effects of the leadership prime on reaction times to attribute
words. Specifically, familiarity with leadership roles implies highly developed and elaborate implicit leadership schemas, hence making it easier
to access this information (H4). Conversely, individuals with a high need
for cognition tend to react more slowly because they are predisposed to
engage in deliberate and systematic thinking rather than automatic and
implicit processing (H5) (Cacioppo & Petty, 1982). Finally, Lord et al.
(2001) proposed that attributes such as “masculine” and “decisive” might
enjoy greater accessibility among male followers while attributes such as
“sensitive” and “helpful” might enjoy greater accessibility among female
followers. Hence, we hypothesized the relationship between leadership
prime and reaction times on agentic and communal leadership attributes
to be stronger for males and females, respectively.
SCHYNS_9781785367274_t.indd 33
10/11/2017 15:19
34 Handbook of methods in leadership research
To test these predictions, Djurdjevic and Johnson (2009) recruited a
sample of 199 undergraduate participants from a large university in USA.
According to prior research, implicit leadership schemas develop early in
life (Lord & Maher, 1991), and are similar across students and full-time
employees (Epitropaki & Martin, 2004). After reporting to the lab, participants completed explicit measures of familiarity with leadership roles
and need for cognition. They were then randomly assigned into an experimental or control condition. In the experimental condition, participants
were primed by instructing them to think and write a half-page description
about leadership and the traits, behaviors, and qualities of an ideal leader.
In the control condition, participants thought and wrote about their
favorite city. Priming is an accessibility-based implicit measure, and helps
to increase the accessibility of a target concept, in this case leadership, to
individuals. This enabled the authors to examine whether the increased
accessibility of the leadership concept influenced their reactions to leaderrelated attributes.
After the priming task, participants completed a computer-based LDT,
created using SuperLab Pro 2.0 (Cedrus Corp., 1999). The LDT is an
accessibility-based measure that enabled Djurdjevic and Johnson (2009) to
assess how accessible leader-related attributes were in participants. Leaderrelated attribute words validated in prior studies (Epitropaki & Martin,
2004; Lord et al., 1984; Offermann et al., 1994) were used to develop the
LDT content. These words are presented in Table 2.1. Participants were
presented with letter strings one at a time on the screen, and they had
pressed “<” on the keyboard if the letter string was a word (e.g., “smart”)
or “>” if it was a non-word (e.g., “smeet”). A faster reaction time indicates
greater accessibility of the word at implicit levels. Participants responded
to a total of 98 letter strings, consisting of 31 prototypical (e.g., “intelligent”) and anti-prototypical (e.g., “domineering”) words, ten non-leader
Table 2.1
Content of ILTs as captured by explicit measures
Authors
Characteristics
Lord, De Vader, & Alliger
(1986)
Offermann, Kennedy, &
Wirtz (1994)
Prototypical: Intelligence, masculinity, &
dominance
Prototypical (primary): Intelligence, dedication,
charisma, & sensitivity
Prototypical (secondary): Masculinity, tyranny,
strength, & attractiveness
Prototypical: Intelligence, dedication, charisma,
& sensitivity
Anti-prototypical: Masculinity & tyranny
Epitropaki & Martin (2004)
SCHYNS_9781785367274_t.indd 34
10/11/2017 15:19
Implicit measures for leadership research ­
35
Table 2.2
upported EFA factor structure of the leader attribute word
S
reaction times
Factor:
Sensitivity
Intelligence
Motivation
Dominance
Tyranny
Words:
Sincere
Helpful
Understanding
Sensitive
Clever
Knowledgeable
Intelligent
Educated
Motivated
Dedicated
Strong
Energetic
Diligent
Dynamic
Domineering
Aggressive
Loud
Masculine
Dominant
Selfish
Conceited
Manipulative
attribute words (e.g., “funny”), six non-attribute words (e.g., “roof”), and
47 non-words (e.g., “renkylo”).
The data were cleaned following the guidelines recommended by Bassili
(2001; e.g., deleted incorrect responses, and removing reaction times that
fell outside of three standard deviations of participants’ mean reaction
time). Djurdjevic and Johnson (2009) conducted an exploratory factor
analysis (EFA) with the reaction times, and their results supported a fivefactor solution, presented in Table 2.2. These results demonstrated some
consistency with those involving explicit measures of leadership schemas
(Epitropaki & Martin, 2004), where factors corresponding to intelligence
and sensitivity emerged. A third factor also emerged that combined the dedication and dynamism factors from Epitropaki and Martin’s (2004) study,
which the authors labeled “motivation.” However, results also deviated
from past findings because Epitropaki and Martin’s tyranny factor split
into two factors in Djurdjevic and Johnson’s data. They labeled one factor
“tyranny” (e.g., selfish, manipulative), and the other factor “dominance”
(e.g., domineering, loud). Attributes from Epitropaki and Martin’s masculinity factor (e.g., masculine, aggressive) also loaded on this dominance
factor. These findings suggest moderate convergence between the factor
structures derived from implicit and explicit leadership schema measures.
Hypotheses were tested via t-tests to compare the mean reaction times
of participants in the experimental versus control conditions. As hypothesized, participants primed with leadership responded more quickly to
attribute words associated with intelligence, dynamism, and dedication
(e.g., clever, intelligent, motivated, dedicated, energetic). Conversely,
participants in the experimental condition responded more slowly to antiprototypical attribute words (e.g., selfish, manipulative) as compared to
control participants. As expected, Djurdjevic and Johnson (2009) also
found that participants responded more quickly to the words “visionary,”
“innovative,” “resourceful,” and “flexible” when primed to think about
leadership.
SCHYNS_9781785367274_t.indd 35
10/11/2017 15:19
36 Handbook of methods in leadership research
Of greater interest were the findings pertaining to the words that have
received mixed support as prototypical of leaders, namely “masculinity,”
“dominance,” and “sensitivity.” When explicit measures of leadership
schemas are used, these three attributes are sometimes rated as prototypical and at other times as anti-prototypical. When the LDT was used,
Djurdjevic and Johnson (2009) found that participants in the experimental
condition responded faster to dominance-related words (e.g., domineering, aggressive, masculine) than individuals in the control condition,
which suggests that dominance is prototypical of leadership. Djurdjevic
and Johnson’s findings that involved an implicit measure therefore parallel earlier studies (e.g., Lord et al., 1984; Offermann et al., 1994), which
concluded that dominance is prototypical of leadership. The results for
sensitivity were less clean: some sensitive-related words appeared prototypical (e.g., sincere, sensitive) whereas others did not (e.g., understanding,
helpful).
Last, Djurdjevic and Johnson (2009) found support for the moderating effects of familiarity with leadership roles and need for cognition, and
partial support for gender. Specifically, participants with high (vs low)
familiarity with leadership roles or with low (vs high) need for cognition
tended to have faster reaction times when responding to prototypical leadership attributes. With respect to gender, the authors found that female
participants appeared to have slightly faster reaction times to communal
attribute words (e.g., words associated with sensitivity) while males had
slightly faster reaction times to agentic attribute words (e.g., words associated with dominance), but several differences did not reach statistical
significance.
Djurdjevic and Johnson’s (2009) study provides a nice illustration
of how implicit measures can be employed for leadership research. An
implicit measure was needed in this case because the process of leader
categorization and the content of leadership schemas are believed to
occur automatically and exist outside of awareness. While Djurdjevic and
Johnson’s results demonstrated some parallels across implicit and explicit
measures (e.g., intelligence and motivation are both prototypical), they
also revealed some notable divergence. Specifically, sensitivity did not
appear to be a prototypical attribute, and dominance and masculinity
emerged as prototypical rather than anti-prototypical attributes. These
findings suggest that leadership schemas examined using explicit methods
of measurement do not necessarily generalize to leadership schemas at
implicit levels.
These findings highlight a need to understand the source of the explicit–
implicit dissociation. In particular, the divergence could be a result of
responding with bias (e.g., social desirability or impression management)
SCHYNS_9781785367274_t.indd 36
10/11/2017 15:19
Implicit measures for leadership research ­
37
on explicit measures. If that is the case, designing explicit measures that
can reduce response bias may reconcile the explicit–implicit dissociation.
Alternatively, the divergence could be due as a result of leadership schemas
existing and operating in parallel on the explicit and implicit levels. If so,
future research should seek to examine the relative importance of explicit
and implicit leadership schemas in the prediction of follower cognitions
and behaviors.
The explicit–implicit dissociations also suggest the possibility that
certain prototypical leader attributes have been overlooked. This is
because much leadership categorization research used attributes established in prior studies to inform the design of their explicit or implicit
measures. For example, Epitropaki and Martin (2004) used attributes
from Lord et al. (1984) and Offermann et al. (1994) for inclusion in their
explicit measure, and Djurdjevic and Johnson’s (2009) study used attributes from those prior studies for inclusion in their LDT. However, these
prior studies have used explicit measures exclusively, suggesting the likelihood that some attributes operating primarily on the implicit levels may
not have been captured. This suggests that the pool of leadership-related
attributes that is commonly used in leader categorization studies may be
incomplete or unrepresentative. Indeed, attributes that did not emerge
in prior studies, such as flexibility and resourcefulness, did emerge as
prototypical leader attributes when an implicit measure was used, thus
­providing support for this argument.
In conclusion, we urge researchers to select measures designed to
capture constructs at an implicit level if there is reason to believe that
the focal leadership-related processes and/or content are believed to exist
outside of awareness. While Djurdjevic and Johnson’s (2009) discrepant
findings may be discouraging with respect to the generalizability of findings from leadership categorization research utilizing explicit measures,
we believe these explicit–implicit dissociations add to our understanding
of the topic. Hence, we hope this empirical example inspires greater use
of implicit measures in future leadership research. With this in mind, we
propose some directions for future research in the following section.
SCOUTING THE FUTURE OF IMPLICIT MEASURES
FOR LEADERSHIP RESEARCH
We devote this final section to discussing potential avenues for future
leadership research using implicit measures. To date, researchers have
typically assessed the main effects of implicit measures, often assessing
whether implicit measures have incremental predictive validity beyond
SCHYNS_9781785367274_t.indd 37
10/11/2017 15:19
38 Handbook of methods in leadership research
explicit measures (e.g., Johnson & Saboe, 2011; Johnson et al., 2010;
Johnson, Chang, Meyer, Lanaj, & Way, 2013). However, as implicit
research advances, we expect future research will explore how implicit and
explicit processes interact (i.e., moderation), influence each other indirectly (i.e., mediation), or change over time (i.e., iteration). The following
discussion is organized based on these three directions.
Moderation
Interactions between implicit and explicit measures can occur in several
ways (Uhlmann et al., 2012). First, explicit measures may facilitate or
inhibit the influence of implicit measures on outcomes. In particular,
McClelland and colleagues (1989) argued that implicit and explicit processes reflect two different motivational systems that operate in parallel.
Individuals may not always be aware of their implicit or unconscious
intentions, or they may even engage in explicit cognition and behavior to
guard against their true (implicit) intentions (ibid.). Such processes give
rise to the likelihood of explicit traits or behaviors to facilitate or inhibit
the influence of implicit measures on outcomes. For example, Schoel and
colleagues (2011) demonstrated how self-reported explicit self-esteem
moderated the implicit attitudes individuals had toward democratic and
autocratic leadership. Extending this work, leadership scholars may find
it fruitful to look at how various explicit measures of personality traits
or environmental factors influence individuals’ implicit attitudes toward
different types of leadership styles or different leaders. For example,
employees in a work team with a high power distance climate may have a
positive implicit attitude towards authoritarian leadership, or employees
with low agreeableness may show greater implicit intention to retaliate
against an abusive supervisor. The move beyond examining the main or
additive effects of implicit measures on their outcomes of variables will
allow researchers to elucidate how explicit and implicit processes coexist
and interact to shape leaders’ or followers’ cognition and behaviors.
Second, leadership scholars may study whether and how individuals
engage in explicit attempts to suppress or conceal their implicit intentions or motives. For example, individuals with implicit biases against the
minority actually showed greater explicit motivation to control prejudice
and provided overly favorable ratings to minority targets (Olson & Fazio,
2004). These findings illustrate an explicit overcompensation effect where
individuals attempt to hide their socially undesirable implicit attitudes. It
will be interesting to replicate this study in the context of leadership. For
example, will leaders who are implicitly biased against minorities engage
in greater explicit motivation to control their prejudice and provide extra
SCHYNS_9781785367274_t.indd 38
10/11/2017 15:19
Implicit measures for leadership research ­
39
resources to their minority followers? These findings will certainly shed
some light on how biases impact leader ratings of performance or leader
intention to support or promote certain followers.
In addition, leadership researchers can conduct research to uncover
factors that moderate the correlations between implicit and explicit
measures. For example, Nosek (2005) looked at the moderators of the
correlations between IAT and self-report measures across a diverse array
of attitudes, and demonstrated that the correlations between IAT and selfreport measures were lower in domains where social desirability concerns
were high, such as in gender or racial stereotyping as opposed to consumer
preferences. Applied to leadership research, we believe it will be rewarding
to study such convergence and divergence of implicit and explicit measures
in order to better identify specific leadership domains in which implicit
measures are necessary for an accurate and comprehensive understanding
of the construct of interest.
Mediation
Leadership scholars may also consider using a mediation framework to
study how implicit and explicit processes causally influence one another
to shape cognitions and behaviors. In particular, explicit processes may
influence outcomes indirectly through the effects on implicit processes
(Uhlmann et al., 2012). In support of this notion, affective events theory
(Weiss & Cropanzano, 1996) proposes that episodes of work events elicit
automatic affective responses in employees through an implicit process,
and that such affective responses go on to influence explicit work behaviors. Based on this theory, scholars could examine how leader–member
interactions activate implicit affective responses in both leaders and followers, and how such automatic responses spillover to impact subsequent
attitudes or behaviors of the two parties. In addition, implicit processes
may influence outcomes indirectly through their effects on explicit processes. For example, self-concept is often implicitly and spontaneously
activated, and this activation potentially informs individuals in their
explicit evaluation and reports of their self-concepts (Peters & Gawronski,
2011). In turn, explicit measures of self-concepts influence conscious
thoughts and behaviors (ibid.). Therefore, leadership researchers may
study how implicit values and beliefs influence the explicit judgments that
individuals form in different leadership scenarios. For example, followers
with low implicit self-esteem may unconsciously prefer leadership styles
or behaviors that offer them greater structure and clearer instructions as
compared to followers with high implicit self-esteem. In turn, their implicit
self-esteem may shape how they explicitly ascribe causal attributions to
SCHYNS_9781785367274_t.indd 39
10/11/2017 15:19
40 Handbook of methods in leadership research
certain leadership styles or behaviors, and predict their subsequent behaviors toward certain leaders. Future research on implicit–explicit mediation
frameworks will help extend leadership theory and research.
Iteration
Finally, future leadership research may benefit from examining the
iterative properties of implicit processes. One direction is to examine
whether content and processes at the implicit levels are consistent
over time and across situations. Scholars have argued that implicit
attitudes develop early in childhood and remain stable through life
(Wilson, Lindsey, & Schooler, 2000). However, several studies have
also demonstrated that implicit attitudes may be malleable and can be
easily manipulated by situational factors (Lowery, Hardin, & Sinclair,
2001). Such mixed findings present intriguing research questions for
leadership scholars. For example, how stable are followers’ implicit
leadership schemas, and to what extent do they change when followers
switch leaders, jobs, or organizations? As noted by Lord et al. (2001),
individuals form implicit schemas about leadership attributes through
socialization and life experiences. Hence, it is possible for their implicit
schemas to change as they work under different leaders or as they work
in jobs or organizations with different leadership cultures. For example,
a follower working under an abusive leader who actually manages to
get work done effectively for a prolonged period of time may gradually disassociate tyranny as an anti-prototypical attribute. The question
then falls on investigating when implicit leadership schemas start to
get modified, and whether the implicit schemas can only accommodate
incremental changes or also radical changes. This study of the stability
of implicit leadership schemas will certainly help to expand our understanding of how implicit leadership content develops and operates to
influence behaviors over time.
Another potential direction is to study the temporal relationships
between implicit processes, explicit processes, and behavior. Implicit processes, explicit processes, and behaviors are typically multidirectional and
play out in iteration over time (Uhlmann et al., 2012). Hence, researchers
may investigate whether and when implicit attitudes reflect explicit behaviors, and vice versa. For example, can newly adopted explicit cognitions
and behaviors shape implicit cognitions? Applied more specifically to
leadership research, an example would be whether explicit social movements such as gender equality shape followers’ implicit evaluations of
lesbian, gay, bisexual, and transgender (LGBT) leaders. Such research
questions could be addressed by longitudinal designs where researchers
SCHYNS_9781785367274_t.indd 40
10/11/2017 15:19
Implicit measures for leadership research ­
41
assess the constructs of interest with both implicit and explicit measures
at multiple time points, and statistically analyse whether the data fits the
argument of implicit attitudes at Time 1 predicting behavior at Time 2 or
behavior at Time 1 predicting implicit attitude at Time 2. These findings
will prove generative for understanding the potential reciprocal roles of
implicit and explicit processes.
CONCLUSION
Most existing leadership research is based on explicit aspects of followers and leaders and the interactions between them that are captured at a
deliberative level within awareness. However, our discussion of implicit
leadership content and processes suggests that implicit measures can
potentially address the limitations posed by explicit leadership measures and tap into unique cognitive and affective processes that operate
outside individuals’ awareness. We also believe that the relevance and
applicability of implicit theories and measures will only increase in the
future as complex organizational demands place higher cognitive load on
leaders and followers and push a significant proportion of information
processing to below conscious levels. Our literature review attests to the
applicability of implicit theories in leadership research, especially in the
areas of leader categorization, leadership motives, leader evaluation, and
self-concepts of leaders and followers. However, our review also reveals
that leadership scholars appear to be lagging in their adoption of implicit
measures in their research, and that a lot more can be done with such
measures in the field. We believe the use of implicit measures will enable
more leadership phenomena to be empirically and accurately tested. As a
means to encourage the use of implicit measures for leadership research,
we provided a comprehensive set of criteria that provides actionable
knowledge to guide leadership scholars in their selection and adoption
of implicit measures for research. We also presented an example study
that employed an implicit measure of leadership. Finally, we concluded
by proposing potential directions for future leadership research that
may benefit from the use of implicit measures. We hope this chapter will
serve as a jump-off point for organizational researchers to explore and
utilize implicit measures of leadership, and inspire a program of research
examining how implicit processes influence leadership-related issues in
the workplace.
SCHYNS_9781785367274_t.indd 41
10/11/2017 15:19
42 Handbook of methods in leadership research
NOTES
1. For example, http://www.millisecond.com/download/library/LexicalDecisionTask/; last
accessed July 15, 2017.
2. For example, http://www.millisecond.com/download/library/Stroop/; last accessed July
15, 2017.
3. For example, http://www.millisecond.com/download/library/AffectivePriming/ or http://
www.millisecond.com/download/library/SubliminalPriming; last accessed July 15, 2017.
4. For example, http://www.millisecond.com/download/library/IAT; last accessed July 15,
2017.
5. For example, http://www.millisecond.com/download/library/; last accessed July 15, 2017.
REFERENCES
Atkinson, J.W. (1958). Motives in fantasy, action, and society: A method of assessment and
study. Oxford: Van Nostrand.
Baldwin, M.W., Carrell, S.E., & Lopez, D.F. (1990). Priming relationship schemas: My
advisor and the Pope are watching me from the back of my mind. Journal of Experimental
Social Psychology, 26(5), 435–454. doi: 10.1016/0022-1031(90)90068-W
Balota, D.A., Yap, M.J., Hutchison, K.A., Cortese, M.J., Kessler, B., Loftis, B., . . .
Treiman, R. (2007). The English Lexicon Project. Behavior Research Methods, 39(3),
445–459. doi: 10.3758/BF03193014
Bass, B.M. (1985). Leadership and performance beyond expectations. New York: Free Press.
Bassili, J.N. (2001). Cognitive indices of social information processing. In A. Tesser &
N. Schwartz (Eds.), The Blackwell handbook of social psychology: Intraindividual processes
(Vol. 1, pp. 68–88). Oxford: Blackwell Publishing.
Blair, I.V. (2002). The malleability of automatic stereotypes and prejudice. Personality and
Social Psychology Review, 6(3), 242–261. doi: 10.1207/S15327957PSPR0603_8
Bowling, N.A., & Johnson, R.E. (2013). Measuring implicit content and processes at work:
A new frontier within the organizational sciences. Human Resource Management Review,
23(3), 203–204. doi: 10.1016/j.hrmr.2012.12.001
Brewer, M.B., & Gardner, W. (1996). Who is this “we”? Levels of collective identity and
self representations. Journal of Personality and Social Psychology, 71(1), 83–93. doi:
10.1037/0022-3514.71.1.83
Brislin, R.W., Lonner, W.J., & Thorndike, R.M. (1973). Cross-cultural research methods.
New York: John Wiley.
Burwen, L.S., Campbell, D.T., & Kidd, J. (1956). The use of a sentence completion test in
measuring attitudes toward superiors and subordinates. Journal of Applied Psychology,
40(4), 248–250. doi: 10.1037/h0044718
Cacioppo, J.T., & Petty, R.E. (1982). The need for cognition. Journal of Personality and
Social Psychology, 42(1), 116–131. doi: 10.1037/0022-3514.42.1.116
Cedrus Corp. (1999). SuperLab Pro: Experimental lab software (Version 2.0). San Pedro,
CA: Cedrus.
Chaiken, S., & Trope, Y. (1999). Dual-process theories in social psychology. New York:
Guilford Press.
Chung-Herrera, B.G., & Lankau, M.J. (2005). Are we there yet? An assessment of fit
between stereotypes of minority managers and the successful-manager prototype. Journal
of Applied Social Psychology, 35(10), 2029–2056. doi: 10.1111/j.1559-1816.2005.tb02208.
Collins, A.M., & Loftus, E.F. (1975). A spreading-activation theory of semantic processing.
Psychological Review, 82(6), 407–428. doi: 10.1037/0033-295X.82.6.407
Conrey, F.R., Sherman, J.W., Gawronski, B., Hugenberg, K., & Groom, C.J. (2005).
Separating multiple processes in implicit social cognition: The quad model of implicit
SCHYNS_9781785367274_t.indd 42
10/11/2017 15:19
Implicit measures for leadership research ­
43
task performance. Journal of Personality and Social Psychology, 89(4), 469–487. doi:
10.1037/0022-3514.89.4.469
Dasgupta, N., & Asgari, S. (2004). Seeing is believing: Exposure to counterstereotypic
women leaders and its effect on the malleability of automatic gender stereotyping. Journal
of Experimental Social Psychology, 40(5), 642–658. doi: 10.1016/j.jesp.2004.02.003
Davies, M., & Gardner, D. (2013). A frequency dictionary of contemporary American English:
Word sketches, collocates and thematic lists. New York: Routledge.
Davies, P.G., Spencer, S.J., & Steele, C.M. (2005). Clearing the air: Identity safety moderates
the effects of stereotype threat on women’s leadership aspirations. Journal of Personality
and Social Psychology, 88(2), 276–287. doi: 10.1037/0022-3514.88.2.276
De Hoogh, A.H., Den Hartog, D.N., Koopman, P.L., Thierry, H., Van den Berg, P.T., Van
der Weide, J.G., & Wilderom, C.P. (2005). Leader motives, charismatic leadership, and
subordinates’ work attitude in the profit and voluntary sector. The Leadership Quarterly,
16(1), 17–38. doi: 10.1016/j.leaqua.2004.10.001
De Houwer, J., Teige-Mocigemba, S., Spruyt, A., & Moors, A. (2009). Implicit measures:
A normative analysis and review. Psychological Bulletin, 135(3), 347–368. doi: 10.1037/
a0014211
Djurdjevic, E., & Johnson, R. (2009). Putting the “implicit” in implicit leadership theory
(ILT): Assessing ILTs using implicit measures. Paper presented at the 69th Academy of
Management Annual Meeting, Chicago, IL.
Eagly, A.H., & Karau, S.J. (2002). Role congruity theory of prejudice toward female leaders.
Psychological Review, 109(3), 573–598. doi: 10.1037/0033-295X.109.3.573
Ebrahimi, B.P. (1997). Motivation to manage in Hong Kong: Modification and test of Miner
Sentence Completion Scale-H. Journal of Managerial Psychology, 12(6), 401–414. doi:
10.1108/02683949710176151
Eden, D., & Leviatan, U. (1975). Implicit leadership theory as a determinant of the factor
structure underlying supervisory behavior scales. Journal of Applied Psychology, 60(6),
736–741. doi: 10.1037/0021-9010.60.6.736
Epitropaki, O., & Martin, R. (2004). Implicit leadership theories in applied settings: Factor
structure, generalizability, and stability over time. Journal of Applied Psychology, 89(2),
293–310. doi: 10.1037/0021-9010.89.2.293
Epitropaki, O., & Martin, R. (2005). From ideal to real: A longitudinal study of the role of
implicit leadership theories on leader–member exchanges and employee outcomes. Journal
of Applied Psychology, 90(4), 659–676. doi: 10.1037/0021-9010.90.4.659
Farh, J.-L., & Cheng, B.-S. (2000). A cultural analysis of paternalistic leadership in Chinese
organizations. In J.T. Li, A.S. Tsui & E. Weldon (Eds.), Management and organizations in
the Chinese context (pp. 84–127). London: Palgrave Macmillan.
Fazio, R.H., & Olson, M.A. (2003). Implicit measures in social cognition research: Their
meaning and use. Annual Review of Psychology, 54(1), 297–327. doi: 10.1146/annurev.
psych.54.101601.145225
Fitzsimmons, S., & Marcuse, F. (1961). Adjustment in leaders and non-leaders as measured
by the sentence completion projective technique. Journal of Clinical Psychology, 17(4),
380–381. doi: 10.1002/1097-4679(196110)17:4<380::AID-JCLP2270170412>3.0.CO;2-A
Greenwald, A.G., & Banaji, M.R. (1995). Implicit social cognition: Attitudes, self-esteem,
and stereotypes. Psychological Review, 102(1), 4–27. doi: 10.1037/0033-295X.102.1.4
Greenwald, A.G., McGhee, D.E., & Schwartz, J.L. (1998). Measuring individual differences in implicit cognition: The implicit association test. Journal of Personality and Social
Psychology, 74(6), 1464–1480. doi: 10.1037/0022-3514.74.6.1464
Gündemir, S., Homan, A.C., De Dreu, C.K., & Van Vugt, M. (2014). Think leader, think
white? Capturing and weakening an implicit pro-white leadership bias. PloS One, 9(1),
e83915. doi: 10.1371/journal.pone.0083915
Hanges, P., Lord, R., & Dickson, M. (2000). An information-processing perspective on
leadership and culture: A case for connectionist architecture. Applied Psychology, 49(1),
133–161. doi: 10.1111/1464-0597.00008
Hofmann, W., Gawronski, B., Gschwendner, T., Le, H., & Schmitt, M. (2005). A meta-analysis
SCHYNS_9781785367274_t.indd 43
10/11/2017 15:19
44 Handbook of methods in leadership research
on the correlation between the implicit association test and explicit self-report measures.
Personality and Social Psychology Bulletin, 31(10), 1369–1385. doi: 10.1177/0146167205275613
House, R.J., Spangler, W.D., & Woycke, J. (1991). Personality and charisma in the US presidency: A psychological theory of leader effectiveness. Administrative Science Quarterly,
36(3), 364–396. doi: 10.2307/2393201
Jacobs, R.L., & McClelland, D.C. (1994). Moving up the corporate ladder: A longitudinal
study of the leadership motive pattern and managerial success in women and men. Consulting
Psychology Journal: Practice and Research, 46(1), 32–41. doi: 10.1037/1061-4087.46.1.32
James, L.R. (1998). Measurement of personality via conditional reasoning. Organizational
Research Methods, 1(2), 131–163. doi: 10.1177/109442819812001
James, L.R., LeBreton, J.M., Mitchell, T.R., Smith, D.R., DeSimone, J.A., Cookson,
R., & Lee, H.J. (2012). Use of conditional reasoning to measure the power motive. In
R.S. Landis & J.M. Cortina (Eds.), Frontiers of methodology in organizational research.
London: Routledge.
James, L.R., McIntyre, M.D., Glisson, C.A., Green, P.D., Patton, T.W., LeBreton, J.M.,
. . .Mitchell, T.R. (2005). A conditional reasoning measure for aggression. Organizational
Research Methods, 8(1), 69–99. doi: 10.1177/1094428104272182
Johnson, R.E., & Lord, R.G. (2010). Implicit effects of justice on self-identity. Journal of
Applied Psychology, 95(4), 681–695. doi: 10.1037/a0019298
Johnson, R.E., & Saboe, K.N. (2010). Measuring implicit traits in organizational research:
Development of an indirect measure of employee implicit self-concept. Organizational
Research Methods, 14(3), 530–547. doi: 10.1177/1094428110363617
Johnson, R.E., & Tan, J.A. (2009). Explicit reasons for examining the implicit motive system.
Industrial and Organizational Psychology, 2(01), 103–105. doi: 10.1177/1094428110363617
Johnson, R.E., Chang, C.-H., Meyer, T., Lanaj, K., & Way, J.D. (2013). Approaching success
or avoiding failure? Approach and avoidance motives in the work domain. European
Journal of Personality, 27(5), 424–441. doi: 10.1002/per.1883
Johnson, R.E., Tolentino, A.L., Rodopman, O.B., & Cho, E. (2010). We (sometimes) know
not how we feel: Predicting work behaviors with an implicit measure of trait affectivity.
Personnel Psychology, 63(1), 197–219. doi: 10.1111/j.1744-6570.2009.01166.x
Johnson, S.K., Murphy, S.E., Zewdie, S., & Reichard, R.J. (2008). The strong, sensitive type:
Effects of gender stereotypes and leadership prototypes on the evaluation of male and
female leaders. Organizational Behavior and Human Decision Processes, 106(1), 39–60. doi:
10.1016/j.obhdp.2007.12.002
Kenney, R.A., Schwartz-Kenney, B.M., & Blascovich, J. (1996). Implicit leadership ­theories:
Defining leaders described as worthy of influence. Personality and Social Psychology
Bulletin, 22(11), 1128–1143. doi: 10.1177/01461672962211004
Konst, D., Vonk, R., & Van der Vlist, R. (1999). Inferences about causes and consequences
of behavior of leaders and subordinates. Journal of Organizational Behavior, 20(2), 261–
271. doi: 10.1002/(SICI)1099-1379(199903)20:2<261::AID-JOB889>3.0.CO;2-3
Koopman, J., Howe, M., Johnson, R.E., Tan, J.A., & Chang, C.-H. (2013). A framework for
developing word fragment completion tasks. Human Resource Management Review, 23(3),
242–253. doi: 10.1016/j.hrmr.2012.12.005
Krizan, Z., & Suls, J. (2008). Are implicit and explicit measures of self-esteem related?
A meta-analysis for the name-letter test. Personality and Individual Differences, 44(2),
521–531. doi: 10.1016/j.paid.2007.09.017
Larson, J.R. (1982). Cognitive mechanisms mediating the impact of implicit theories
of leader behavior on leader behavior ratings. Organizational Behavior and Human
Performance, 29(1), 129–140. doi: 10.1016/0030-5073(82)90245-8
Latu, I.M., Mast, M.S., Lammers, J., & Bombari, D. (2013). Successful female leaders
empower women’s behavior in leadership tasks. Journal of Experimental Social Psychology,
49(3), 444–448. doi: 10.1016/j.jesp.2013.01.003
LeBreton, J.M., Barksdale, C.D., Robin, J., & James, L.R. (2007). Measurement issues associated with conditional reasoning tests: Indirect measurement and test faking. Journal of
Applied Psychology, 92(1), 1–16. doi: 10.1037/0021-9010.92.1.1
SCHYNS_9781785367274_t.indd 44
10/11/2017 15:19
Implicit measures for leadership research ­
45
Liden, R.C. (2012). Leadership research in Asia: A brief assessment and suggestions for the
future. Asia Pacific Journal of Management, 29(2), 205–212. doi: 10.1007/s10490-011-9276-2
Liden, R.C., & Maslyn, J.M. (1998). Multidimensionality of leader–member exchange: An
empirical assessment through scale development. Journal of Management, 24(1), 43–72.
doi: 10.1177/014920639802400105
Liden, R.C., Wayne, S.J., Zhao, H., & Henderson, D. (2008). Servant leadership: Development
of a multidimensional measure and multi-level assessment. The Leadership Quarterly, 19(2),
161–177. doi: 10.1016/j.leaqua.2008.01.006
Lilienfeld, S.O., Wood, J.M., & Garb, H.N. (2000). The scientific status of projective techniques. Psychological Science in the Public Interest, 1(2), 27–66. doi: 10.1111/1529-1006.002
Locke, E.A. (1991). The essence of leadership. New York: Lexington Books.
Lord, R.G., & Brown, D.J. (2001). Leadership, values, and subordinate self-concepts. The
Leadership Quarterly, 12(2), 133–152. doi: 10.1016/S1048-9843(01)00072-8
Lord, R.G., & Maher, K.J. (1991). Leadership and information processing: Linking perceptions and performance. New York: Routledge.
Lord, R.G., Binning, J.F., Rush, M.C., & Thomas, J.C. (1978). The effect of performance
cues and leader behavior on questionnaire ratings of leadership behavior. Organizational
Behavior and Human Performance, 21(1), 27–39. doi: 10.1016/0030-5073(78)90036-3
Lord, R.G., Brown, D.J., Harvey, J.L., & Hall, R.J. (2001). Contextual constraints on
prototype generation and their multilevel consequences for leadership perceptions. The
Leadership Quarterly, 12(3), 311–338. doi: 10.1016/S1048-9843(01)00081-9
Lord, R.G., De Vader, C.L., & Alliger, G.M. (1986). A meta-analysis of the relation between
personality traits and leadership perceptions: An application of validity generalization
procedures. Journal of Applied Psychology, 71(3), 402–410.
Lord, R.G., Diefendorff, J.M., Schmidt, A.M., & Hall, R.J. (2010). Self-regulation at work.
Annual Review of Psychology, 61(1), 543–568. doi: 10.1146/annurev.psych.093008.100314
Lord, R.G., Foti, R.J., & De Vader, C.L. (1984). A test of leadership categorization theory:
Internal structure, information processing, and leadership perceptions. Organizational
Behavior and Human Performance, 34(3), 343–378. doi: 10.1016/0030-5073(84)90043-6
Lowery, B.S., Hardin, C.D., & Sinclair, S. (2001). Social influence effects on automatic
racial prejudice. Journal of Personality and Social Psychology, 81(5), 842–855. doi:
10.1037/0022-3514.81.5.842
McClelland, D.C., & Boyatzis, R.E. (1982). Leadership motive pattern and long-term success
in management. Journal of Applied Psychology, 67(6), 737–743.
McClelland, D.C., Koestner, R., & Weinberger, J. (1989). How do self-attributed and implicit
motives differ? Psychological Review, 96(4), 690–702. doi: 10.1037/0033-295X.96.4.690
McClelland, J.L., McNaughton, B.L., & O’Reilly, R.C. (1995). Why there are complementary learning systems in the hippocampus and neocortex: Insights from the successes and
failures of connectionist models of learning and memory. Psychological Review, 102(3),
419–457. doi: 10.1037/0033-295X.102.3.419
Megargee, E.I. (1969). Influence of sex roles on the manifestation of leadership. Journal of
Applied Psychology, 53(5), 377–382. doi: 10.1037/h0028093
Mervis, C.B., & Rosch, E. (1981). Categorization of natural objects. Annual Review of
Psychology, 32(1), 89–115. doi: 10.1146/annurev.ps.32.020181.000513
Meyer, D.E., & Schvaneveldt, R.W. (1971). Facilitation in recognizing pairs of words:
Evidence of a dependence between retrieval operations. Journal of Experimental
Psychology, 90(2), 227–234. doi: 10.1037/h0031564
Miner, J.B. (1978). Twenty years of research on role-motivation theory of managerial effectiveness. Personnel Psychology, 31(4), 739–760. doi: 10.1111/j.1744-6570.1978.tb02122.x
Miner, J.B., & Smith, N.R. (1982). Decline and stabilization of managerial motivation over a 20-year period. Journal of Applied Psychology, 67(3), 297–305. doi:
10.1037/0021-9010.67.3.297
Miner, J.B., Chen, C.-C., & Yu, K. (1991). Theory testing under adverse conditions:
Motivation to manage in the People’s Republic of China. Journal of Applied Psychology,
76(3), 343–349. doi: 10.1037/0021-9010.76.3.343
SCHYNS_9781785367274_t.indd 45
10/11/2017 15:19
46 Handbook of methods in leadership research
Morgan, C.D., & Murray, H.A. (1935). A method for investigating fantasies: The thematic
apperception test. Archives of Neurology and Psychiatry, 34(2), 289–306. doi: 10.1001/
archneurpsyc.1935.02250200049005
Nosek, B.A. (2005). Moderators of the relationship between implicit and explicit evaluation.
Journal of Experimental Psychology: General, 134(4), 565–584. doi: 10.1037/0096-3445.
134.4.565
Offermann, L.R., Kennedy, J.K., & Wirtz, P.W. (1994). Implicit leadership theories:
Content, structure, and generalizability. The Leadership Quarterly, 5(1), 43–58. doi:
10.1016/1048-9843(94)90005-1
Olson, M.A., & Fazio, R.H. (2004). Trait inferences as a function of automatically activated
racial attitudes and motivation to control prejudiced reactions. Basic and Applied Social
Psychology, 26(1), 1–11. doi: 10.1207/s15324834basp2601_1
Parks-Stamm, E.J., Heilman, M.E., & Hearns, K.A. (2008). Motivated to penalize: Women’s
strategic rejection of successful women. Personality and Social Psychology Bulletin, 34(2),
237–247. doi: 10.1177/0146167207310027
Peters, K.R., & Gawronski, B. (2011). Mutual influences between the implicit and explicit
self-concepts: The role of memory activation and motivated reasoning. Journal of
Experimental Social Psychology, 47(2), 436–442. doi: 10.1016/j.jesp.2010.11.015
Robinson, M.D., & Neighbors, C. (2006). Catching the mind in action: Implicit methods
in personality research and assessment. In M. Eid & E. Diener (Eds.), Handbook
of multimethod measurement in psychology (pp. 115–125). Washington, DC: American
Psychological Association.
Rosch, E. (1978). Principles of categorization. In E. Rosch & B.B. Lloyd (Eds.), Cognition
and categorization. Hillsdale, NJ: Erlbaum.
Rudman, L.A., & Kilianski, S.E. (2000). Implicit and explicit attitudes toward female
authority. Personality and Social Psychology Bulletin, 26(11), 1315–1328. doi:
10.1037/0022-3514.81.5.856
Rudman, L.A., & Phelan, J.E. (2010). The effect of priming gender roles on women’s
implicit gender beliefs and career aspirations. Social Psychology, 41(3), 192–202. doi:
10.1027/1864-9335/a000027
Rudman, L.A., Ashmore, R.D., & Gary, M.L. (2001). “Unlearning” automatic biases: The malleability of implicit prejudice and stereotypes. Journal of Personality and Social Psychology,
81(5), 856–868. doi: 10.1037/0022-3514.81.5.856
Rush, M.C., & Russell, J.E. (1988). Leader prototypes and prototype-contingent consensus
in leader behavior descriptions. Journal of Experimental Social Psychology, 24(1), 88–104.
doi: 10.1016/0022-1031(88)90045-5
Rush, M.C., Thomas, J.C., & Lord, R.G. (1977). Implicit leadership theory: A potential
threat to the internal validity of leader behavior questionnaires. Organizational Behavior
and Human Performance, 20(1), 93–110. doi: 10.1016/0030-5073(77)90046-0
Schoel, C., Bluemke, M., Mueller, P., & Stahlberg, D. (2011). When autocratic leaders become
an option – Uncertainty and self-esteem predict implicit leadership preferences. Journal of
Personality and Social Psychology, 101(3), 521–540. doi: 10.1037/a0023393
Scott, K.A., & Brown, D.J. (2006). Female first, leader second? Gender bias in the encoding
of leadership behavior. Organizational Behavior and Human Decision Processes, 101(2),
230–242. doi: 10.1016/j.obhdp.2006.06.002
Spangler, W.D. (1992). Validity of questionnaire and TAT measures of need for achievement:
Two meta-analyses. Psychological Bulletin, 112(1), 140–154. doi: 10.1037/0033-2909.112.1.140
Stahl, M.J., Grigsby, D.W., & Gulati, A. (1985). Comparing the job choice exercise and
the multiple choice version of the Miner Sentence Completion Scale. Journal of Applied
Psychology, 70(1), 228–232. doi: 10.1037/0021-9010.70.1.228
Stajkovic, A.D., Locke, E.A., & Blair, E.S. (2006). A first examination of the relationships
between primed subconscious goals, assigned conscious goals, and task performance.
Journal of Applied Psychology, 91(5), 1172–1180. doi: 10.1037/0021-9010.91.5.1172
Stroop, J.R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental
Psychology, 18(6), 643–662. doi: 10.1037/h0054651
SCHYNS_9781785367274_t.indd 46
10/11/2017 15:19
Implicit measures for leadership research ­
47
Taylor, S.E. (1983). Adjustment to threatening events: A theory of cognitive adaptation.
American Psychologist, 38(11), 1161–1173. doi: 10.1037/0003-066X.38.11.1161
Tepper, B.J. (2000). Consequences of abusive supervision. Academy of Management Journal,
43(2), 178–190. doi: 10.2307/1556375
Tomkins, S.S., & Tomkins, E.J. (1947). The thematic apperception test: The theory and technique of interpretation. New York: Grune & Stratton.
Uhlmann, E.L., Leavitt, K., Menges, J.I., Koopman, J., Howe, M., & Johnson, R.E. (2012).
Getting explicit about the implicit: A taxonomy of implicit measures and guide for their
use in organizational research. Organizational Research Methods, 15(4), 553–601. doi:
10.1177/1094428112442750
Wegner, D.M., & Bargh, J.A. (1998). Control and automaticity in social life. In D.T. Gilbert,
S.T. Fiske & G. Lindzey (Eds.), The handbook of social psychology (4th ed., Vols. 1 and 2,
pp. 446–496). New York: McGraw Hill.
Weiss, H.M., & Cropanzano, R. (1996). Affective events theory: A theoretical discussion of
the structure, causes and consequences of affective experiences at work. In B.M. Staw &
L.L. Cummings (Eds.), Research in organizational behavior: An annual series of analytical
essays and critical reviews (Vol. 18, pp. 1–74). Greenwich, CT: Elsevier Science/JAI Press.
Wilson, T.D., Lindsey, S., & Schooler, T.Y. (2000). A model of dual attitudes. Psychological
Review, 107(1), 101–126. doi: 10.1037/0033-295X.107.1.101
Winter, D.G. (1991). A motivational model of leadership: Predicting long-term management success from TAT measures of power motivation and responsibility. The Leadership
Quarterly, 2(2), 67–80. doi: 10.1016/1048-9843(91)90023-U
Winter, D.G., & Stewart, A.J. (1977). Power motive reliability as a function of retest
instructions. Journal of Consulting and Clinical Psychology, 45(3), 436–440. doi:
10.1037/0022-006X.45.3.436
Zaccaro, S.J., Foti, R.J., & Kenny, D.A. (1991). Self-monitoring and trait-based variance in
leadership: An investigation of leader flexibility across multiple group situations. Journal
of Applied Psychology, 76(2), 308–315. doi: 10.1037/0021-9010.76.2.308
Ziegert, J.C., & Hanges, P.J. (2005). Employment discrimination: The role of implicit attitudes, motivation, and a climate for racial bias. Journal of Applied Psychology, 90(3),
553–562. doi: 10.1037/0021-9010.90.3.553
SCHYNS_9781785367274_t.indd 47
10/11/2017 15:19
3.
uppet masters in the lab: experimental
P
methods in leadership research
Eric F. Rietzschel, Barbara Wisse and
Diana Rus
Leadership researchers typically aim to understand leadership processes
and the impact of leadership on a variety of organizationally important
outcome variables, such as individual and group emotions, motivation,
cognitions, and performance. Hence, most leadership theories either explicitly propose or implicitly assume cause-and-effect relationships (cf. Yukl,
2010). Yet, a surprisingly large proportion of the amassed knowledge in the
field stems from correlational field surveys (cf. Avolio, Reichard, Hannah,
Walumbwa, & Chan, 2009; Bass, 1990). These correlational designs not
only preclude conclusions regarding causality, but also (unnecessarily)
limit researchers in developing a more nuanced understanding of leadership processes and their effects. Consequently, there is an obvious need for
conducting experimental studies in leadership research (cf. Avolio et al.,
2009; Brown & Lord, 1999; Day, Zaccaro, & Halpin, 2004).
Our discussion of experimental methods is not a call for its undifferentiated use. Clearly, the choice of research method depends upon the nature
of the question one tries to answer. In addition, in line with those organizational researchers who advocated the benefits of employing multiple
research methodologies (cf. Jick, 1979), we believe that the field of leadership research would benefit from an increasing use of diverse research
methods, both qualitative and quantitative. One of those methods could
be experimental in nature, and the strengths of this method make it worthy
of consideration. In this chapter, we therefore specifically focus on experimental research methods as an empirical approach that can help us gain a
deeper understanding of leadership processes.
In the sections that follow, we first review the most commonly used
experimental paradigms in leadership research, such as individual laboratory experiments, group experiments, field experiments, and vignette
studies. Next, we address some of the strengths and potential pitfalls
associated with employing experimental methods in leadership research.
Finally, we discuss the future of leadership research by highlighting some
recent developments in the field and pointing out opportunities for further
development and refinement of these methods.
48
SCHYNS_9781785367274_t.indd 48
10/11/2017 15:19
Experimental methods in leadership research ­
49
EXPERIMENTAL METHODS USED IN LEADERSHIP
RESEARCH
First, we will turn our attention to commonly used experimental paradigms and methods in leadership research and provide some examples.
Specifically, we will focus on laboratory experiments with individual participants, group experiments, field experiments, and vignette experiments.
Given the increasing popularity of computer-mediated paradigms (especially for laboratory experiments with individual participants and group
experiments), we will devote specific attention to their use. However,
as an introduction, we will start by briefly describing the prototypical
­experiment in the behavioral sciences.
Usually, participants (individuals or groups) are invited to a laboratory,
where they are asked to read, watch, or listen to certain instructions and/
or other stimulus materials (such as instructions given by a task “leader,”
who may be present in person, or perhaps gives his or her messages
through video or audio recordings, or in written form). These materials
are often used to serve as independent variables. Different versions of
these instructions or stimulus materials (the experimental manipulation)
are used for different groups of participants (the experimental conditions).
When the experiment has multiple independent variables that are manipulated, these are usually combined in a factorial design, where researchers
“cross” the independent variables in such a way that each possible combination occurs. For example, suppose that researchers are interested in
the combined effects of leadership style and feedback valence (this study
would then have two independent variables). They could manipulate
leadership style by presenting participants with either a transformational
leader or a transactional leader, and they could manipulate feedback
valence by exposing participants to either positive feedback or negative
feedback. Thus, in this case the researchers would use a factorial design
with 2 × 2 5 4 conditions (transformational-positive, transactional-­
positive, ­transformational-negative, and transactional-negative). Usually,
participants are randomly assigned to one of these conditions (Rosenthal
& Rosnow, 2008; Shadish, Cook, & Campbell, 2002) and typically
perform certain tasks and/or are asked to respond to a number of questionnaires. The behavior of the participants, their task performance, and/
or questionnaire responses usually function as the main dependent variables. Statistical analysis (such as t-tests, analysis of variance, or regression
analysis) is conducted to assess whether participants responded differently
in the various conditions. Because everything in the experiment, except the
actual manipulation, is kept constant and controlled, and because random
assignment to conditions greatly diminishes the likelihood of pre-existing
SCHYNS_9781785367274_t.indd 49
10/11/2017 15:19
50 Handbook of methods in leadership research
differences between conditions, existing effects can only reasonably be
explained as being a function of the experimental manipulation(s). In
the following, we focus on different categories or types of experimental
paradigms that are often used in leadership research, and provide some
examples.
Laboratory Experiments with Individual Participants
A laboratory experiment with individual participants largely follows the
script of the typical experiment explained above. A good example of such
an experiment is a study by Howell and Frost (1989). In this experiment,
the researchers manipulated leadership style, exposing participants to
charismatic, structuring, or considerate leadership. They combined this in
a factorial design with a manipulation of group productivity norms, and
looked at the effects on individuals’ attitudes and their performance on a
decision-making task. This experiment had an elaborate set-up, involving
trained confederates (actors or individuals who work for the researcher)
who played two different roles (leaders and co-workers). The “leaders”
were trained to follow pre-developed scripts (which specified verbal and
non-verbal behaviors) for one of the different leadership styles in all their
interactions with the study participants. Further, the “co-workers” were
trained to display behaviors indicative of either a high or a low group
productivity norm (i.e., acting enthusiastic and eager in the case of a high
productivity norm, or acting bored and unwilling in the case of a low productivity norm). The experimental task consisted of an in-basket exercise
and participants were subsequently asked to fill out some questionnaires.
The study results showed that charismatic leadership led to good work
outcomes and favorable attitudes to the leader and group, regardless of
group productivity norms. In contrast, considerate and structuring leadership only led to good work outcomes and favorable attitudes when participants worked in a group with a high productivity norm.
Clearly, this experiment required substantial investments in terms of
time, energy, effort and money. In addition to setting up the experimental
rooms and creating all of the materials (instructions, protocols, experimental tasks, and measurement instruments), the confederates needed to
be trained to display the behaviors required by the experiment in a sufficiently uniform, yet believable manner. Often, such resources are simply
not available. One way to reduce this resource investment is to make use
of computers, rather than confederates, and this (in combination with the
control and standardization advantages that computers bring) has probably led to the upsurge of computer-mediated experiments. Indeed, probably the most popular type of experimental paradigm used in leadership
SCHYNS_9781785367274_t.indd 50
10/11/2017 15:19
Experimental methods in leadership research ­
51
research is the computer-mediated laboratory experiment with individual
participants.
Computer-mediated variants
A typical computer-mediated experiment requires participants (for
instance, students or people holding a leadership position) to come to the
laboratory, where they are placed in individual cubicles equipped with
a computer. All subsequent instructions and/or stimuli are presented on
the computer screens, and the software program records the dependent
measures. An “interaction” with other participants (such as a leader or a
follower) is often simulated (with messages from the “other participant”
having been pre-programmed) in order to make participants believe that
they are actually interacting with other individuals.
Computer-mediated experiments are (often) easier to run than experiments involving actual interactions between individuals, but this does
not mean that they cannot employ complex designs – in fact, using such
designs may even be easier in a computer-mediated setting, since the
experimental situation is easier to standardize and control. Rus, Van
Knippenberg, and Wisse (2010; Study 1) report one example of such an
experiment. The purpose of this study was to test the hypothesis that, with
higher leader power, leaders’ self-serving behaviors would be determined
more by internal states, and less by external cues. Participants (in this case
undergraduate students) were seated in individual cubicles with computers, and were led to believe that they were the leader of a four-person
group engaged in a computer-mediated task. This study used no less than
three different manipulations. First, the researchers manipulated power,
by providing participants with either more or less coercive power in the
simulated group task (i.e., giving them the opportunity to fire or reprimand subordinates, or not giving them this power). Second, effective leadership beliefs (internal states) were manipulated by providing participants
with a description of either a self-serving leader or a group-serving leader,
and subsequently asking them to list five reasons why this leader would
be effective in motivating his or her subordinates (thus instilling either
self-serving or group-serving leadership beliefs in participants). Third,
performance information (external cues) was manipulated by presenting
participants with bogus feedback on their task performance as compared
to that of their subordinates (they either did better or worse than their
subordinates). These manipulations were combined in a factorial design,
and participants were randomly assigned to one of the eight experimental conditions. The main dependent variable, self-serving behavior, was
assessed by asking participants to divide a total of 500 points that the team
could earn between themselves and their employees. Each point counted
SCHYNS_9781785367274_t.indd 51
10/11/2017 15:19
52 Handbook of methods in leadership research
as one lottery entry for several €50 prizes. Hence, the more points they
self-awarded, the higher their chances of winning one of the prizes. In line
with the researchers’ expectations, the results indicated that in conditions
of high power, leaders (i.e., the participants) were more self-serving when
they endorsed self-serving beliefs than when they endorsed group-serving
beliefs, while performance information had no effect on leader behavior.
In other words, their behavior was guided more by their internal state than
by external cues. The opposite was the case in conditions of low power.
In experiments on leader behavior such as the one described above,
participants are placed in a leader role and their behaviors, affective states
and/or cognitions are subsequently assessed. In other computer-mediated
experiments, participants are assigned a subordinate or follower role. They
are presented with a leader (for instance via e-mails, messages or film clips)
and their reaction to that leader (or their performance on a subsequent
task) is assessed (e.g., Damen, Van Knippenberg, & Van Knippenberg,
2008b; Grant & Hoffmann, 2011; Van Kleef, Homan, Beersma, Van
Knippenberg, & Van Knippenberg, 2009). Take, for instance, the experiments reported by Venus, Stam, and Van Knippenberg (2013). In these
experiments, the hypothesis was that follower performance would depend
on the match between the emotions expressed by a leader, and the regulatory focus (Higgins, 1997) implied in the leaders’ communication: followers were expected to perform best when the expressed emotions fit the
regulatory focus of the leaders’ messages. Participants in these studies
were shown a video clip of a leader. Each video clip showed a leader that
communicated either a promotion-oriented message (e.g., referring to the
importance of being “able to be flexible under fast changing conditions”
and the benefits of “enthusiastic, creative subordinates that are able to
cope with the complex problems of today”; Venus et al., 2013, p. 57) or
a prevention-oriented message (e.g., referring to the importance of not
being “inflexible and slow under fast changing conditions” and the undesirability of “conservative and bored subordinates that are not able to
cope with the complex problems of today”). Moreover, this leader did so
while displaying different emotional states (i.e., enthusiasm, agitation, or
frustration). After participants had watched the video, they engaged in a
performance task (specifically, a memory task that was presented as a test
of mental ability). Venus et al. (2013) found that, as expected, performance
was highest when leaders’ regulatory orientation matched their emotional
display (for instance, when a promotion orientation was accompanied by
enthusiasm).
A recent development in computer-mediated experiments is conducting experiments via Internet platforms. Such experiments do not require
participants to come to a laboratory at all; instead people can partake
SCHYNS_9781785367274_t.indd 52
10/11/2017 15:19
Experimental methods in leadership research ­
53
in the study at home or at work. Given that more and more people
are online and that the speed of Internet connections keeps increasing,
Internet recruitment methods have become increasingly popular among
researchers. Although running experiments online has the drawback of
relinquishing some experimental control (see below), research has shown
that data obtained with Internet platforms like Amazon’s Mechanical
Turk (MTurk) or ClearVoice are as reliable as those obtained via traditional methods (Buhrmester, Kwang, & Gosling, 2011; Mason & Suri,
2012; Paolacci, Chandler, & Ipeirotis, 2010).1 Similarly to laboratorybased computer-mediated experiments, online experiments often establish
a bogus connection with “other participants” to make participants believe
that they are actually interacting with other individuals.
Group Experiments
Sometimes, the focus of leadership research is not so much on how individuals respond to experimental manipulations, but on how groups of
individuals respond to them (for example, when studying leadership in
teams). This might be the case when researchers are interested in individual-level variables (such as motivation or performance) among individuals who happen to work within teams (and hence might influence each
other’s responses), but more often researchers are specifically interested
in the effects of leadership on variables at the team level, such as team
climate (e.g., Gil, Rico, Alcover, & Barrasa, 2005) or team innovation
(Eisenbeiss, Van Knippenberg, & Boerner, 2008). One possibility is to
aggregate all responses to the team level, thus using the team as the unit
of analysis. However, analytical techniques such as multilevel regression
(e.g., Hox, 2010; also see Yammarino & Gooty, Chapter 10 this volume)
allow researchers to combine variables at the individual and the team level
in a single model (e.g., testing cross-level interactions between team-level
variables such as climate and individual-level variables such as personality). Whenever the research question concerns variables that occur only
at the team level, or that are likely to be significantly affected by the team
context, researchers should consider the use of a group study.
Take, for instance, a study conducted by Sy, Cȏté, and Saavedra (2005).
Their study focused on the effect that the mood of a leader would have on
the affective tone of the group (the “collective,” aggregated mood state of
the group) and on group processes (such as coordination and investment
of effort). To investigate this, they had 56 pre-existing student groups participate in their study. In each group, one randomly chosen member was
assigned the leader role. To manipulate the mood of these leaders, they
were asked to watch one of two versions of an eight-minute video clip,
SCHYNS_9781785367274_t.indd 53
10/11/2017 15:19
54 Handbook of methods in leadership research
away from their group. Leaders in the positive mood condition viewed a
humorous clip of David Letterman, while leaders in the negative mood
condition viewed part of a TV documentary about social injustice and
aggression. Thereafter, leaders were reunited with their groups and were
given some time to interact and plan for the upcoming task, which was
to erect a tent while wearing a blindfold. Immediately after this planning
stage, group members engaged in the actual task. Group processes were
coded in vivo as groups worked on the task. Group affective tone was calculated by averaging the scores of individual mood measurements within
groups. The results showed, among other things, that groups had a more
positive affective tone and exhibited more coordination when leaders were
in a positive mood than when leaders were in a negative mood. Analyses
also suggested that group affective tone mediated the effect of leader mood
on group coordination.
Computer-mediated variants
Similarly to experiments with individual participants, group experiments
are sometimes also computer-mediated and some of those experiments are
even conducted via Internet platforms. In these cases, an actual (rather
than simulated) connection with other participants is established to ensure
group interaction. Carton, Murphy, and Clark (2014; Study 2) conducted
a nice example of such an experiment. The goal was to study the combined effects of two aspects of leader communication: the degree of vision
imagery (e.g., describing products as “crafted flawlessly” versus “made to
perfection”) and the number of values communicated by the leader (such
as “customer satisfaction,” “profitability,” and “integrity”). Specifically,
the researchers wanted to test whether the combination of strong vision
imagery and a small number of values would lead to better performance
than other combinations. Both were experimentally manipulated, but
while vision imagery was manipulated as a between-subjects factor (each
participant was exposed to messages that were either strong or weak in
imagery), the number of values was manipulated as a within-subjects
factor (each participant was exposed to a message with a high number of
values and to a message with a low number of values; the order of these
messages was counterbalanced). The researchers recruited employees via
the research platform ClearVoice, and assigned them randomly to one of
the conditions. Participants were then placed into virtual teams of three
members, and each member was given a different task in the development of a new toy for a toy company. All participants were told that it
was important that their actions were congruent with the leader’s statements regarding the vision and values of the company. The quality of
the toy design was the dependent variable; the sharedness of goal percep-
SCHYNS_9781785367274_t.indd 54
10/11/2017 15:19
Experimental methods in leadership research ­
55
tions within the team and coordination were the hypothesized mediators.
Results showed that when leaders used strong imagery in combination
with a low number of values, teams had more shared goals and better
coordination, which in turn predicted better task performance.
Field Experiments
Although the examples discussed above take place in different settings
(e.g., the laboratory versus online), none of them actually studies leaders
and subordinates in their natural environment: the organization. This is
what field experiments do: a field experiment aims to examine leadership
processes in naturally occurring environments rather than in the laboratory. Field experiments are sometimes seen as having higher external
validity than laboratory experiments, but it is often more difficult to
control extraneous variables than in the lab (see below for a more extensive discussion). Moreover, it is not always possible to use true random
assignment in field experiments. For example, when the manipulation
entails training or an intervention (see example below), organizations may
want to co-decide who gets to be trained, or which teams get assigned to
an intervention condition. This effectively turns the study into a quasiexperiment, which of course limits its internal validity.
The most common manipulation in field experiments on leadership
is some form of leader training/development, where leaders are either
assigned to a certain training program or not, and the subsequent behavior, cognitions, or feelings of their subordinates (or of the leaders themselves) are then assessed (see Avolio et al., 2009). A fairly recent example
comes from Martin, Liao, and Campbell (2013), who conducted a field
experiment in the United Arab Emirates. The experiment addressed the
effects of two kinds of leadership (directive and empowering leadership)
on task performance and proactive behaviors. Leaders who participated
in the study were assigned to one of two training conditions (for directive
or empowering leadership), or to a control condition without any training.
Leaders were asked to keep daily logs, and subordinates (from the leaders’
work units) and customers were asked to fill out several surveys on, among
other things, satisfaction with the leader, perceived task proficiency of
the leaders’ work units, and proactivity of the leaders’ work units. These
surveys were filled out both before and after the leadership training period
(a pre-test/post-test design). Thus, the complete study design not only
allowed the researcher to assess differences between the training conditions and the control group, but also to test for actual changes in the
dependent variables over time.
Results indicated that both directive and empowering leadership
SCHYNS_9781785367274_t.indd 55
10/11/2017 15:19
56 Handbook of methods in leadership research
increased work unit core task proficiency. Notably, directive leadership
only enhanced proactive behaviors in work units that were highly satisfied
with their leaders, whereas empowering leadership had stronger effects
on both core task proficiency and proactive behaviors for work units that
were less satisfied with their leaders.
One advantage of field experiments is that they focus on variables that
are directly relevant to organizations, such as actual performance in work
units. This can be an important “selling point” when contacting organizations for possible participation in a field experiment, and it is also considered a strength in the research community (e.g., by editors and reviewers).
Nevertheless, the heavy time investment and the difficulties in establishing
experimental control make the field experiment the most demanding of all
experimental methods in leadership research.
Vignette Experiments
If field experiments represent one end of the “ease of use” continuum in
experimental research, vignette experiments probably represent the other.
A vignette experiment, sometimes also called a scenario experiment, presents a hypothetical situation, to which research participants are asked
to respond. For example, they can be asked to indicate how they would
perceive the presented situation or persons, or how they expect they would
feel or behave in that situation. Of course, asking people how they think
they would respond is not the same as measuring actual responses in a
given situation, so although these kinds of experiments are conducted
quite frequently in leadership research, they are usually part of a series
of studies. Mostly, they are used to check whether a previously found
result is robust and can be replicated by using different methods, or as
an initial test of a possible causal relation that would then be replicated
using a more elaborate experimental setting. For instance, having demonstrated a causal relation in lab experiments with student participants,
one may conduct a vignette study with respondents from a working population to ­demonstrate that the obtained results generalize to this target
population.
Because vignette studies require very little in the way of experimental
materials, they can easily be used in a variety of settings, such as the lab,
the field, or even online. Then again, because vignette studies present
participants with a hypothetical situation rather than immersing them in
an actual task, they are more suited for measuring hypothetical reactions
(i.e., how participants think they would behave in a certain situation) than
for assessing actual reactions (such as work performance). Again, this
means that vignette studies are best used in combination with other, more
SCHYNS_9781785367274_t.indd 56
10/11/2017 15:19
Experimental methods in leadership research ­
57
elaborate studies, especially when researchers are interested in testing
behavioral hypotheses.
An example of this approach is presented in Van Knippenberg and
Van Knippenberg (2005), who conducted a vignette experiment on leader
self-sacrifice (the degree to which leaders place the interests of the work
group above their own) and leader prototypicality (the degree to which a
leader is representative of a work group’s identity). The authors expected
that self-sacrificing leaders would be considered to be more effective and
able to push subordinates to a higher performance level than non-­selfsacrificing leaders – especially if these leaders were not very prototypical (the underlying reasoning being that highly prototypical leaders get
more “leeway” from their workgroup anyway, and hence do not need to
display self-sacrificing behavior). Having first established these results in
a computer-mediated laboratory experiment, the authors then replicated
their main results in a vignette study, followed by two survey studies.
In the vignette study, participants (business school students) were told
that they would read about a situation in which leadership played a role,
that they were to imagine themselves being in that particular situation,
and that they were to answer the subsequent questions accordingly.
Participants were then handed the scenario. The scenario asked them to
envision that, having graduated, they had gone to work for an international consulting agency with a very good reputation. Depending on the
experimental condition, their leader in that organization was portrayed
as being high or low in group prototypicality. For example, in the low
prototypicality condition participants read that “the leader was somewhat of an ‘outsider’ in the team, that he/she was very different from
other team members and that he or she had a different background,
different interests, and a different attitude toward life and work than
the other team members” (Van Knippenberg & Van Knippenberg, 2005,
p. 31). In contrast, participants in the high prototypicality condition
read that “the leader was very representative of the kind of persons in
the team, that he was very similar to the other team members, and that
he had a similar background, similar interests, and a similar attitude
toward life and work as the other team members.” Another paragraph in
the scenario described the leader’s behavior as either self-sacrificing (vs
non-self-sacrificing) by giving (vs not giving) examples of self-sacrificial
behavior displayed by that leader. Participants were then asked to assess
the effectiveness of the leader. In line with the other studies, results
indicated that self-sacrificing leaders were seen as more effective than
non-self-sacrificing leaders, but only when the leaders were low in prototypicality. The authors argue that the replication over multiple studies
bolsters confidence in the finding, especially because these studies used
SCHYNS_9781785367274_t.indd 57
10/11/2017 15:19
58 Handbook of methods in leadership research
different methodologies (i.e., laboratory experiment, scenario experiment, and cross-sectional survey).
In short, there are many ways to run a leadership experiment, employing different settings (e.g., lab, field, or online), materials (e.g., confederates, computerized instructions, or scenarios), or respondents (employees,
undergraduate students, or online panel respondents). Researchers can
pick the type of experiment they want to do, depending on what the
research question or program requires and which resources are available.
In order to facilitate this choice, it may also be helpful to think about some
more general advantages and pitfalls associated with the experimental
method.
STRENGTHS AND PITFALLS OF EXPERIMENTS
No single research method is the optimal choice for every research question or setting, and researchers need to make an informed choice as to
which method best suits their goals and means. In this section, we will
therefore discuss the main advantages and (possible) disadvantages of the
experimental method in leadership research.
Advantages of Experimental Research in Leadership
Internal validity
The most important advantage of experiments is their high internal
validity: the combination of systematic manipulation (making sure that
only one aspect of the instructions or experimental situation differs
between conditions, hence ruling out so-called procedural confounds)
and random assignment (which minimizes the chance of pre-existing differences between conditions, hence ruling out person-related confounds)
makes it possible to eliminate alternative explanations for the effect
under study, and hence allows the researcher to draw causal conclusions (Rosenthal & Rosnow, 2008; Shadish et al., 2002; Stone-Romero,
2002). Other research designs, such as correlational, qualitative, or even
quasi-experimental designs (in which participants are assigned to different conditions, but not fully randomly), do not have this advantage
and hence always lead to weaker conclusions with regard to causality.
Certain leader behaviors may be associated with certain outcomes, or
may – in a statistical sense – predict certain outcomes, but in the absence
of truly experimental research, they cannot be concluded to cause or even
contribute to those outcomes, since alternative explanations (e.g., a third
variable, or reverse causation) cannot be ruled out. Thus, for example, a
SCHYNS_9781785367274_t.indd 58
10/11/2017 15:19
Experimental methods in leadership research ­
59
field survey study finding that, on the whole, supervisory close monitoring
behavior (constantly keeping close tabs on employees; Zhou, 2003) tends
to be associated with lower employee motivation and satisfaction (e.g.,
Rietzschel, Slijkhuis, & Van Yperen, 2014) could indicate a causal relationship, but might also be explained by, say, the presence of an unfavorable
team climate (influencing both supervisor behavior and employee dissatisfaction), or by a causal relation in the reverse direction (e.g., demotivated
versus motivated employees may well elicit different kinds of leader behaviors). Only by conducting an experiment in which leadership behaviors
are systematically manipulated and participants are randomly assigned to
conditions (i.e., to different leader behaviors), can actual causal conclusions be drawn.
Testing interventions
The ability to draw causal conclusions is not just important for theoretical
reasons (gaining a complete and accurate understanding of the phenomenon under study; Mook, 1983) but also from a more applied perspective. Designing effective interventions (such as managerial training; e.g.,
Arthur, Bennett, Edens, & Bell, 2003; Collins & Holton, 2004) requires
accurate knowledge of the factors that contribute to (or hinder) leader
effectiveness. Organizational change efforts based on studies that do not
rule out the possibility of third variables or inverse causation might turn
out to be a considerable waste of time and money (e.g., Spector, 2010), and
in the worst case may leave both leaders and followers disillusioned and
give rise to change cynicism (Thundiyil, Chiaburu, Oh, Banks, & Peng,
2015). Moreover, testing the effectiveness of interventions (e.g., if we want
to see whether a certain training indeed had the desired effect) is a causal
question in itself, requiring (if possible) the use of experimental methods
(Cascio & Aguinis, 2011). However, as remarked above, a real test of
causality requires random assignment, and this is often not possible, especially in the context of interventions. Organizations are not always willing
to allow true random assignment to intervention versus control conditions
because some teams or units may experience more problems than others
and hence may be seen to have a stronger need for the intervention. In
such cases, it is important to measure and partial out possible differences
between the experimental conditions other than the manipulation itself
(see Shadish et al., 2002, for a further discussion of quasi-experimental
methods).
Experimental control and isolating specific factors
Although experimental methods are not necessarily limited to a particular setting, many experimental studies in organizational psychology,
SCHYNS_9781785367274_t.indd 59
10/11/2017 15:19
60 Handbook of methods in leadership research
i­ ncluding the field of leadership, are conducted in the laboratory (Bryman,
2011; Robson, 2011; Stone-Romero, 2009), or in what Stone-Romero
(2009) calls special purpose (SP) settings. It is difficult to conduct true
experiments in naturalistic settings such as organizations (non–special
purpose or NSP settings; Stone-Romero, 2009), because an experiment
can easily interfere with the day-to-day business on-site (e.g., Robson,
2011), and the field setting makes it difficult to achieve the desired level of
experimental control. Moreover, it is almost impossible to prevent people
in different conditions (but working within the same organization) from
communicating with each other about the study, thus possibly leading to
spillover between conditions. For example, Blumberg and Pringle (1983)
report on an experiment (strictly speaking, a quasi-experiment, since
people were not randomly assigned to conditions) with the introduction of
autonomous workgroups in a Pennsylvania coal mine. Some groups were
given high autonomy over their work tasks and scheduling, and received
extensive training for this new way of working. Other groups kept working
as usual (the control groups). Over time, serious tensions arose between
the members of the autonomous groups and the control groups because
members of the control groups felt deprived of attention and rewards.
Eventually, although the intervention did seem to have the desired effects
in that the autonomous workgroups improved on several aspects in their
functioning (such as better intragroup coordination), the experiment had
to be terminated. Although this example is rather extreme, it does illustrate why most experimental researchers tend to prefer SP settings (such
as the lab) over NSP settings (such as organizations): they simply allow
for more control, and – as explained above – control is essential to the
experimental method.
The systematic manipulation essential to true experiments further
implies a relatively fine-grained focus on specific behaviors or other predictors, rather than on “clusters” of leadership behaviors that tend to cooccur. Experiments are aimed at identifying causal relationships, and this
also means isolating (potential) causal factors, in order to find out which
one actually contributes to the effect. Thus, researchers interested in the
effects of a particular leadership style, such as charismatic leadership (e.g.,
Howell & Frost, 1989), will need to operationalize that style in such a way
that (only) the crucial aspects are present in their manipulation (cf. Yukl,
1999). Moreover, since the independent variables in the study are at the
discretion of the researchers, it is possible to study the effects of factors
or behaviors that may be relatively rare or difficult to identify in reallife settings, for example because they do not take place very openly or
because employees may be reluctant to report them (such as certain kinds
of abusive leadership; Tepper, 2007).
SCHYNS_9781785367274_t.indd 60
10/11/2017 15:19
Experimental methods in leadership research ­
61
Testing complex models
Paradoxically, it is precisely the focus on single, specific predictors that
allows researchers to systematically study their combined effects as well
(e.g., Fisher, 1984). By orthogonally (i.e., independently) manipulating
such predictors in a factorial design (Shadish et al., 2002), researchers can
see whether the effects of one predictor depend on the level of another predictor (moderation), or whether these predictors independently (and possibly additively) exert their effects. As we have seen, many of the studies
described in the previous section employed such factorial designs, and the
results of these studies are informative in not just showing whether certain
leadership behaviors are effective (e.g., providing followers with promotion-focused or prevention-focused messages; Venus et al., 2013), but also
identifying boundary conditions for those benefits (such as congruence
with a leader’s emotional expression).
Further, independent variables can be manipulated in different ways:
for example, rather than merely testing the effects of presence versus
absence of certain behaviors or factors, it is also possible to test the effects
of different levels of certain independent variables – which in turn makes
it possible to test non-monotonic relations, such as inverted U-shaped
relations (Grant & Schwartz, 2011), where a certain predictor leads to
more positive outcomes at intermediate levels than at very low or very
high levels.
Low time investment
In addition to the potential for unique information (e.g., regarding
causality), experiments can also have a practical advantage: it does not
necessarily take very long to run an experiment. Of course, this only goes
for some types of experiments: as described above, vignette studies may
be relatively easy to run quickly, whereas experiments with more highfidelity simulations of leadership and using behavioral coding (e.g., to
study group processes, as in the study by Sy et al., 2005) may require much
more preparation, as well as extensive data processing. Nevertheless,
once the relevant materials (such as vignettes) have been developed, many
experiments can be run in a relatively short time frame. In contrast, many
organizational studies, even those using relatively uncomplicated designs
such as correlational surveys, require much longer time periods for data
collection: access to and cooperation from the organization need to be
secured, participants need to be invited, and reminders need to be sent
(e.g., Robson, 2011). All in all, this can easily take months – and even
then it is often highly uncertain how many respondents the researcher will
end up with. Thus, experiments allow the researcher a fast route to testing
specific causal hypotheses. A next step, of course, could be to conduct field
SCHYNS_9781785367274_t.indd 61
10/11/2017 15:19
62 Handbook of methods in leadership research
studies to see whether these causal effects indeed can be observed within
actual organizations.
Pitfalls of Experimental Research in Leadership
Despite the very clear and important advantages of the experimental
research method in leadership research, there are inevitably some drawbacks as well. Mirroring the advantages described above, many of these
revolve around the classic counterpart to high internal validity: (supposedly) low external validity. We will discuss several of these pitfalls below,
and critically discuss whether each of them forms as serious a risk to
experimental research as is sometimes assumed.
Lack of external validity
Systematic manipulation and experimental control are essential to safeguarding internal validity, but at the same time they can put the study’s
external validity (the ability to generalize to a broader array of populations
and settings) at risk. Although all experiments share this disadvantage to
a certain extent, it is most salient in lab experiments. A lab experiment by
definition takes place in a highly artificial setting: simplified and controlled,
without much obvious resemblance to organizational reality (Bryman,
2011; Fisher, 1984; Stone-Romero, 2009). Although this is, of course,
precisely the point of doing a lab experiment, it does raise questions with
regard to external validity. We cannot automatically assume that effects
observed in the lab will be identical, or even of the same order of magnitude, in organizations. The effects might be moderated by all kinds of
personal and organizational characteristics not taken into account in the
experiment. For example, the involvement of participants in lab studies
begins and ends with their participation in the experiment, there is no social
context to speak of (except to the extent that it is part of the experimental
situation), there are no serious consequences attached to participants’
behavior, group members (mostly) have no history of working together,
and participants often have only limited behavioral or other response
options at their disposal (Aronson, Ellsworth, Carlsmith, & Gonzales,
1990). All of this is obviously very different from organizational reality.
Although this objection to (lab) experiments is often made, one may
wonder whether the concern is theoretically and empirically justified. For
example, it is doubtful whether studies conducted in a real-life setting,
such as an organization, have higher external validity. Stone-Romero
(2009, p. 308) argues that “even when studies are conducted in NSP settings, they typically involve non-representative samples of subjects, settings, and operational definitions of manipulations and/or measures.
SCHYNS_9781785367274_t.indd 62
10/11/2017 15:19
Experimental methods in leadership research ­
63
Thus, the external validity of such studies is suspect. The fact that they
were conducted in NSP settings often does nothing to strengthen external
validity inferences.” Organizations, teams, supervisors, and employees
differ from setting to setting, so there is no reason to assume that research
results can easily be generalized when they are collected in the field as
opposed to the lab.
In addition, Aronson et al. (1990) argue that it is too simple to state
that experiments are not “realistic,” because there are different kinds of
“realism.” Most objections against (lab) experiments are based on their
low mundane realism: experimental settings do not resemble real-world
settings in most ways. However, Aronson and colleagues point out that
mundane realism is not always important: “The mere fact that an event
looks like one that occurs in the real world does not imply that it is important in the study of human behavior. Many events that occur in the real
world are boring or uninfluential” (Aronson et al., 1990, p. 70). What is
important, they argue, is experimental realism: the experimental situation
must be “believable” for the participants in order to have an impact on
behavior (e.g., when studying the effects of feedback styles, participants
must believe that they will actually receive feedback on their performance;
see Shalley & Perry-Smith, 2001). Empirically, moreover, the results
of “artificial” experiments and “realistic” field studies tend to be more
aligned than critics of the experimental method assume (e.g., Anderson,
Lindsay, & Bushman, 1999; Locke, 1986).
Actually, with some creativity, it often is possible to come up with
experimental manipulations and tasks that resemble the real-world situation in important ways. For example, researchers interested in competitive
behavior in the workplace can actually have participants engage in a competitive task (or lead them to believe they are competing; e.g., Lee, Kesebir,
& Pillutla, 2016, Studies 3 and 4), and researchers interested in sales
performance may ask participants to do a task that simulates what they
would have to do in a computer retail store (Damen, Van Knippenberg,
& Van Knippenberg, 2008a). Although experimental realism arguably is
more important than mundane realism, it is usually a good idea to try to
combine the two to a certain extent, and researchers increasingly do so.
Another important point is that even if experiments sometimes are low
on external validity, this need not be a problem, because generalization
may not be what the researchers are aiming for at that moment (Mook,
1983). In many experiments, the primary goal is to establish a causal
link (e.g., to replicate correlational field data, to conduct an initial test
of a tentative intervention, or to test a theoretical process explanation).
In such cases, the artificiality of an experiment is not problematic, since
­generalization simply is not the immediate concern.
SCHYNS_9781785367274_t.indd 63
10/11/2017 15:19
64 Handbook of methods in leadership research
Use of student samples
Another potential problem is the pervasive use of undergraduate students
(usually psychology or business students) as respondents. The reason for
relying on these populations is practical: they are available in relatively
large numbers, are easily accessible within the university context (where
most of this research is done), and are available at low costs (students may
participate for course credits or for modest financial reimbursement).
The problem with relying on student samples is that students are likely to
be different from the target population (leaders or followers) in a number
of respects, and this might affect results. For example, because students
typically have little work experience, let alone supervisory experience, they
may find it difficult to relate to situations sketched in a vignette or scenario
study. Notably, this again underscores the importance of experimental
realism described above: researchers need to create an experimental situation that is meaningful to these participants, and that will allow them to
study the phenomenon of interest in this particular sample. Moreover, as
Highhouse and Gillespie (2009, p. 258) have pointed out: “The degree to
which a sample matches the population of interest does not affect one’s
ability to detect a relation between variables of theoretical significance, as
long as that sample is unbiased on factors relevant to the research question.” For example, the fact that most undergraduate participants score
above average (as compared to the general population) on measures of
cognitive ability is not problematic as long as the phenomenon under
study is unrelated to cognitive ability. These differences in cognitive ability
do not imply that they will respond differently to, say, a manipulation of
leadership style. Of course, not all manipulations, tasks, or measurement
instruments are equally suited for use in all possible populations, and use
of methods that are not suited to the population can lead to floor or ceiling
effects (Osborne, 2013). Suppose, for example, that a researcher wants to
test the hypothesis that a certain leadership style will lead to lower performance because of reduced employee effort, but uses an experimental
task that happens to be extremely easy for the undergraduate participant
population. Since the task is so easy for these participants, performing well on the task will require little if any effort, and the hypothesized
effects will probably not be observed. Ultimately, however, this would not
be an example of research with low external validity because of the participant population sampled, but rather an example of inadequately tested
research materials.
Low-impact manipulations of high-impact situations
All psychological research is subject to ethical guidelines (American
Psychological Association, 2010). This means that there are limits to the
SCHYNS_9781785367274_t.indd 64
10/11/2017 15:19
Experimental methods in leadership research ­
65
situations participants can be exposed to. As a consequence, some situations that may occur in organizations cannot be reproduced with sufficient fidelity in the lab. For example, experimental research on unethical
or abusive leadership is limited by the impossibility and undesirability of
exposing participants to serious forms of such unethical behaviors (which
may, in extreme cases, include verbal and even physical violence). To be
able to study such phenomena at all, researchers have to resort to “lowimpact” manipulations of the same phenomena that are deemed to elicit
similar responses but that can still be considered ethical. The effects of
such “low-impact” manipulations may be much more subtle than those
that would occur in real-world settings. Luckily, the stringent control
inherent in experiments often enables researchers to pick up such subtle
effects, because extraneous noise is minimized (Aronson et al., 1990).
Moreover, experiments conducted in a laboratory often allow researchers
to monitor participants’ mood states and other reactions better (and in
real time) than studies conducted in a field or online settings do.
Short-term approaches to long-term phenomena
As mentioned above, one advantage of (certain types of) experiments is
that they can often be run within a short time frame. Such a short-term
approach can work well when studying phenomena that have immediate effects, such as affective reactions to feedback (e.g., Niemann, Wisse,
Rus, Van Yperen, & Sassenberg, 2014). However, it becomes more problematic when the phenomenon of interest unfolds over a longer period of
time. Some outcomes may take longer periods of time to develop (e.g.,
burnout; Maslach, Schaufeli, & Leiter, 2001) and some processes require
repeated interaction over a prolonged period of time (e.g., the development of leader–member exchange relationships; Graen & Uhl-Bien, 1995)
before their effects are visible. Moreover, even if immediate effects can be
observed, there is the question of how these might further develop over
time. For example, an initial strong affective reaction to negative feedback
may eventually be followed by the use of emotion regulation strategies
(Gross, 1998), such as reappraisal (“this feedback will help me learn”),
which may then lead to very different behavioral outcomes than the initial
affective response would lead the researcher to suspect.
Once again, whether or not this is problematic depends on the goal of
the research. If the goal is to demonstrate that negative feedback can lead
to certain immediate affective and attitudinal reactions (cf. Niemann et
al., 2014), the use of a short-term paradigm is fully defensible. If, however,
the goal is to present an account of how negative feedback affects subordinates’ feelings and behavior over a longer period of time, additional work
will probably be required.
SCHYNS_9781785367274_t.indd 65
10/11/2017 15:19
66 Handbook of methods in leadership research
In this section, we have given an overview of the main advantages and
potential pitfalls of the experimental method in leadership research. As
will have become clear, whether these advantages are realized, and the
pitfalls avoided, depends on the researchers’ choice for a specific type of
experiment, the population they want to use, the methodological rigor
of the experiment (e.g., designing manipulations without procedural
confounds, and attaining high experimental realism), and on the goals
of the study (e.g., with regard to generalization). In addition, researchers may want to consider recent developments in experimental research:
some types of experimental research have been made possible, or much
easier, by technological developments. Other types of research are now
appreciated less than they used to be, due to changes in methodological
and statistical norms. In the next section, we therefore present a brief look
towards the future.
A LOOK TO THE FUTURE OF EXPERIMENTAL
METHODS IN LEADERSHIP RESEARCH
Predicting future developments in any field is an educated guess at best,
and unfounded speculation at worst. That being said, in this section we
will briefly discuss a number of what may be the more impactful potential
developments in the field of experimental methods in leadership research
(also see Lord, Chapter 16 in this volume).
Upsurge in Usage and Sophistication of Technology-enabled Experiments
Computer-mediated and online experiments are already being performed
in leadership research. A number of converging factors suggest that such
technology-enabled experiments are not only here to stay, but most likely
will also see an upsurge in both usage and sophistication. Firstly, the
American Psychological Association (APA) has explicitly approved their
use and publications employing such designs have increasingly entered
mainstream journals. Given that the initial barrier to publication has been
(partially) removed, this bodes well for leadership researchers aiming to
make use of such designs in the future.
Secondly, increased connectivity across the world combined with rising
Internet speed and the proliferation of online platforms for experimental
research provide the opportunity for leadership researchers to collect data
they would not have been able to collect a decade ago. As a consequence,
researchers now have access to a large, ready pool of participants who
are demographically diverse (e.g., in terms of ethnicity, age, culture, work
SCHYNS_9781785367274_t.indd 66
10/11/2017 15:19
Experimental methods in leadership research ­
67
experience, etc.). Indeed, Buhrmester et al. (2011) found that MTurk
samples were more demographically diverse than typical American college
samples. Such increased diversity in participants would not only address
the often-mentioned pitfall of using undergraduate students as experimental participants, but would also allow researchers to test a different set of hypotheses as it enables them to delve into understanding the
effects of leadership across a range of diverse constituents. A more refined
understanding of leadership processes, especially as they play out in cross-­
cultural contexts, would be welcome given the increasingly dispersed
nature of work and teams. In addition, the increased speed and lower costs
of running experimental research online would allow leadership researchers to more quickly iterate between theory development and experimentation (cf. Mason & Suri, 2012), therefore potentially contributing to faster
advances in the field.
Thirdly, technological advances such as the increased sophistication
of online platforms, the upsurge of gamification simulations and virtual
reality developments provide opportunities to study leadership processes
in vivo, within a tightly controlled environment, without losing the “richness” of the context. For instance, the increased sophistication of online
platforms affords leadership researchers the opportunity to conduct group
experiments that establish real-time connections among a diverse set of
participants, something we could not have done in the past. In addition,
gamification simulations (Hamari & Koivisto, 2015) have started to gain
ground as being part and parcel of leadership development programs
(e.g., Deloitte, NTT Data), and surely could be designed to be used in
the more traditional, controlled laboratory settings. Such set-ups could
provide researchers with the opportunity to run field experiments in vivo
that would provide a wealth of data and unique opportunities to study
the effects of different leadership development interventions on employee
motivation, behavior, and performance. Similarly, technological advances
in virtual reality simulations would provide leadership researchers with the
opportunity to study leadership processes as they play out, for instance,
in group settings.
Changes in Methodological and Statistical Norms
Whereas traditionally the norms of what constitutes good practice in
(experimental) psychology research have gradually evolved over time,
the last few years have seen a dramatic shift in this respect. For instance,
whereas articles based on single-source, cross-sectional data or single
experiments with low sample sizes used to be prevalent in reputable
journals as late as the 1980s, this is no longer the case. In this respect,
SCHYNS_9781785367274_t.indd 67
10/11/2017 15:19
68 Handbook of methods in leadership research
there has been an increasing shift towards a norm of testing the robustness of findings via replication studies and across multiple methodologies (cf. Simmons, 2014; Stone-Romero, 2009). For instance, different
APA divisions and other professional organizations, such as the Society
for Personality and Social Psychology, have recently provided recommendations for improving the dependability of research such as the
use of larger, more powerful sample sizes, reporting effect sizes and
95 percent confidence intervals, making available research materials
necessary to replicate reported results, and encouraging publication of
high-quality replication studies (Funder et al., 2014). Similarly, some
journals have developed new guidelines for publication, specifically
focusing on the vulnerabilities of null-hypothesis testing and the need
for larger samples sizes. These shifts have major implications for leadership researchers engaging in experimental research. For one, conducting
multiple studies, employing different methodologies, will be the new
standard (if it is not already). In addition, even within the experimental methodology, a combination of different experimental paradigms,
where the potential downsides of the one paradigm are compensated
for by the potential upsides of the other, appears to emerge as the new
norm. For instance, the results of an online experiment would benefit
from being replicated within the lab and potentially in a vignette experiment. In sum, it seems that the norms and practices within our field are
shifting towards an increased focus on safeguarding the robustness of
our findings.
CONCLUDING THOUGHTS
In this chapter, we have provided an overview of experimental research
methods in leadership research, addressing commonly used methods and
designs, potential advantages and pitfalls, and future developments that
seem relevant for researchers who want to conduct a leadership experiment. In doing so, we have focused mostly on researchers with a strong
interest in, but little experience with experimental research. Because
successfully conducting experimental research of course requires extensive theoretical knowledge (e.g., regarding the nature and scope of the
variables under study) and practical skills (e.g., regarding the construction of manipulations or task instructions), in this chapter we could only
brush the surface of some of the major issues. Nonetheless, we hope to
have ­provided the reader with an overview that is both inspiring and
informative.
SCHYNS_9781785367274_t.indd 68
10/11/2017 15:19
Experimental methods in leadership research ­
69
NOTE
1. However, other researchers, such as Harms and DeSimone (2015), do point towards
possible risks of MTurk samples, and a recent article by Zhou and Fishbach (2016)
experimentally addresses the possible consequences of (selective) participant attrition in
online samples.
REFERENCES
American Psychological Association (2010). Publication manual of the American Psychological
Association (6th ed.). Washington, DC: American Psychological Association.
Anderson, C.A., Lindsay, J.J., & Bushman, B.J. (1999). Research in the psychological
laboratory: Truth or triviality? Current Directions in Psychological Science, 8(1), 3–9. doi:
10.1111/1467-8721.00002
Aronson, E., Ellsworth, P.C., Carlsmith, J.M., & Gonzales, M.H. (1990). Methods of research
in social psychology. Boston, MA: McGraw-Hill.
Arthur, W., Jr., Bennett W., Jr., Edens, P.S., & Bell, S.T. (2003). Effectiveness of training
in organizations: A meta-analysis of design and evaluation features. Journal of Applied
Psychology, 88(2), 234–245. doi: 10.1037/0021-9010.88.2.234
Avolio, B.J., Reichard, R.J., Hannah, S.T., Walumbwa, F.O., & Chan, A. (2009). A metaanalytic review of leadership impact research: Experimental and quasi-experimental
studies. The Leadership Quarterly, 20(5), 764–784. doi: 10.1016/j.leaqua.2009.06.006
Bass, B.M. (1990). Bass and Stogdill’s handbook of leadership (3rd ed.). New York: The Free
Press.
Blumberg, M., & Pringle, C.D. (1983). How control groups can cause loss of control in
action research: The case of Rushton coal mine. Journal of Applied Behavioral Science,
19(4), 409–425. doi: 10.1177/002188638301900402
Brown, D.J. & Lord, R.G. (1999). The utility of experimental research in the study of transformational/charismatic leadership. Leadership Quarterly, 10(4), 531–539. doi: 10.1016/
S1048-9843(99)00029-6
Bryman, A. (2011). Research methods in the study of leadership. In A. Bryman, D. Collinson,
K. Grint, B. Jackson, & M. Uhl-Bien (Eds.), The SAGE handbook of leadership (pp.
15–28). London: Sage Publications.
Buhrmester, M., Kwang, T., & Gosling, S.D. (2011). Amazon’s Mechanical Turk: A new
source of inexpensive, yet high-quality, data? Perspectives on Psychological Science, 6(1),
3–5. doi: 10.1177/1745691610393980
Carton, A.M., Murphy, C., & Clark, J.R. (2014). A (blurry) vision of the future: How leader
rhetoric about ultimate goals influences performance. Academy of Management Journal,
57(6), 1544–1570. doi: 10.5465/amj.2012.0101
Cascio, W.F., & Aguinis, H. (2011). Applied psychology in human resource management (7th
ed.). Upper Saddle River, NJ: Prentice Hall.
Collins, D.B., & Holton, E.F. (2004). The effectiveness of managerial leadership development programs: A meta-analysis of studies from 1982 to 2001. Human Resource
Development Quarterly, 15(2), 217–248. doi: 10.1002/hrdq.1099
Damen, F., Van Knippenberg, B., & Van Knippenberg, D. (2008a). Affective match: Leader
emotions, follower positive affect, and follower performance. Journal of Applied Social
Psychology, 38(4), 868–902. doi: 10.1111/j.1559-1816.2008.00330.x
Damen, F., Van Knippenberg, D., & Van Knippenberg, B. (2008b). Leader affective displays
and attributions of charisma: The role of arousal. Journal of Applied Social Psychology,
38(10), 2594–2614. doi: 10.1111/j.1559-1816.2008.00405.x
Day, D.V., Zaccaro, S.J., & Halpin, S.M. (2004). Leader development for transforming organizations: Growing leaders for tomorrow. Mahwah, NJ: Lawrence Erlbaum.
SCHYNS_9781785367274_t.indd 69
10/11/2017 15:19
70 Handbook of methods in leadership research
Eisenbeiss, S.A., Van Knippenberg, D., & Boerner, S. (2008). Transformational leadership
and team innovation: Integrating team climate principles. Journal of Applied Psychology,
93(6), 1438–1446. doi: 10.1037/a0012716
Fisher, C.D. (1984). Laboratory experiments. In S. Bateman, & G.R. Ferris (Eds.), Method
and analysis in organizational research (pp. 169–200). Reston, VA: Reston Publishing
Company, Inc.
Funder, D.C., Levine, J.M., Mackie, D.M., Morf, C.C., Sansone, C., Vazire, S., & West,
S.G. (2014). Improving the dependability of research in personality and social psychology: Recommendations for research and educational practice. Personality and Social
Psychology Review, 18(1), 3–12. doi: 10.1177/1088868313507536
Gil, F., Rico, R., Alcover, C.M., & Barrasa, Á. (2005). Change-oriented leadership, satisfaction and performance in work groups: Effects of team climate and group potency. Journal
of Managerial Psychology, 20(3/4), 312–328. doi: 10.1108/02683940510589073
Graen, G.B., & Uhl-Bien, M. (1995). Relationship-based approach to leadership:
Development of leader–member exchange (LMX) theory of leadership over 25 years:
Applying a multi-level multi-domain perspective. The Leadership Quarterly, 6(2), 219–247.
doi: 10.1016/1048-9843(95)90036-5
Grant, A.M., & Hofmann, D.A. (2011). Outsourcing inspiration: The performance effects of
ideological messages from leaders and beneficiaries. Organizational Behavior and Human
Decision Processes, 116(2), 173–187. doi: 10.1016/j.obhdp.2011.06.005
Grant, A.M., & Schwartz, B. (2011). Too much of a good thing: The challenge and
opportunity of the inverted U. Perspectives on Psychological Science, 6(1), 61–76. doi:
10.1177/1745691610393523
Gross, J.J. (1998). The emerging field of emotion regulation: An integrative review. Review of
General Psychology, 2(3), 271–299. doi: 10.1037/1089-2680.2.3.271
Hamari, J., & Koivisto, J. (2015). Why do people use gamification services? International
Journal of Information Management, 35(4), 419–431. doi: 10.1016/j.ijinfomgt.2015.04.006
Harms, P.D., & DeSimone, J.A. (2015). Caution! MTurk workers ahead – fines doubled.
Industrial and Organizational Psychology, 8(2), 183–190. doi: 10.1017/iop.2015.23
Higgins, E.T. (1997). Beyond pleasure and pain. American Psychologist, 52(12), 1280–1300.
doi: 10.1037/0003-066X.52.12.1280
Highhouse, S., & Gillespie, J.Z. (2009). Do samples really matter that much? In C.E. Lance,
& R.J. Vandenberg (Eds.), Statistical and methodological myths and urban legends:
Doctrine, verity and fable in the organizational and social sciences (pp. 247–265). New York:
Routledge/Taylor & Francis Group.
Howell, J.M., & Frost, P.J. (1989). A laboratory study of charismatic leadership. Organizational Behavior and Human Decision Processes, 43(2), 243–269. doi:
10.1016/0749-5978(89)90052-6
Hox, J.J. (2010). Multilevel analysis: Techniques and applications (2nd ed.). New York:
Routledge.
Jick, T. (1979). Mixing qualitative and quantitative methods: Triangulation in action.
Administrative Science Quarterly, 24(4), 602–611. doi: 10.2307/2392366
Lee, S.Y., Kesebir, S., & Pillutla, M.M. (2016). Gender differences in response to competition with same-gender coworkers: A relational perspective. Journal of Personality and
Social Psychology, 116(6), 869–886. doi: 10.1037/pspi0000051
Locke, E.A. (1986). Generalizing from laboratory to field: Ecological validity or abstraction
of essential elements? In E.A. Locke (Ed.), Generalizing from laboratory to field settings
(pp. 257–267). Indianapolis, IN: D.C. Heath.
Martin, S.L., Liao, H., & Campbell, E.M. (2013). Directive versus empowering leadership:
A field experiment comparing impacts on task proficiency and proactivity. Academy of
Management Journal, 56(5), 1372–1395. doi: 10.5465/amj.2011.0113
Maslach, C., Schaufeli, W.B., & Leiter, M.P. (2001). Job burnout. Annual Review of Psychology,
52(1), 397–422. Retrieved from https://doi.org/10.1146/annurev.psych.52.1.397
Mason, W., & Suri, S. (2012). Conducting behavioral research on Amazon’s Mechanical
Turk. Behavior Research Methods, 44(1), 1–23. doi: 10.3758/s13428-011-0124-6
SCHYNS_9781785367274_t.indd 70
10/11/2017 15:19
Experimental methods in leadership research ­
71
Mook, D.G. (1983). In defense of external invalidity. American Psychologist, 38(4), 379–387.
doi: 10.1037/0003-066X.38.4.379
Niemann, J., Wisse, B., Rus, D., Van Yperen, N.W., & Sassenberg, K. (2014). Anger and
attitudinal reactions to negative feedback: The effects of emotional instability and power.
Motivation and Emotion, 38(5), 687–699. doi: 10.1007/s11031-014-9402-9
Osborne, J.W. (2013). Best practices in data cleaning. Thousand Oaks, CA: Sage.
Paolacci, G., Chandler, J., & Ipeirotis, P.G. (2010). Running experiments on Amazon
Mechanical Turk. Judgment and Decision making, 5(5), 411–419. Retrieved from https://
papers.ssrn.com/sol3/papers.cfm?abstract_id51626226
Rietzschel, E.F., Slijkhuis, M., & Van Yperen, N.W. (2014). Close monitoring as a contextual stimulator: How need for structure affects the relation between close monitoring and
work outcomes. European Journal of Work and Organizational Psychology, 23(3), 394–404.
doi: 10.1080/1359432X.2012.752897
Robson, C. (2011). Real world research (3rd ed.). Chichester, UK: Wiley.
Rosenthal, R., & Rosnow, R.L. (2008). Essentials of behavioral research: Methods and data
analysis. Boston, MA: McGraw-Hill.
Rus, D., Van Knippenberg, D., & Wisse, B.M. (2010). Leader power and self-serving
behavior: The role of effective leadership beliefs and performance information. Journal of
Experimental Social Psychology, 46(6), 922–933. doi: 10.1016/j.jesp.2010.06.007
Shadish, W.R., Cook, T.D., & Campbell, D.T. (2002). Experimental and quasi-experimental
designs for generalized causal inference. Boston, MA: Houghton, Mifflin and Company.
Shalley, C.E., & Perry-Smith, J.E. (2001). Effects of social-psychological factors on creative
performance: The role of informational and controlling expected evaluation and modeling
experience. Organizational Behavior and Human Decision Processes, 84(1), 1–22. doi:
10.1006/obhd.2000.2918
Simmons, D.J. (2014). The value of direct replication. Perspectives on Psychological Science,
9(1), 76–80. doi: http://dx.doi.org/10.1177/1745691613514755
Spector, B. (2010). Implementing organizational change. Upper Saddle River, NJ: Prentice Hall.
Stone-Romero, E.F. (2002). The relative validity and usefulness of various empirical research
designs. In S.G. Rogelberg (Ed.), Handbook of research methods in industrial and organizational psychology (pp. 77–98). Malden, MA: Blackwell Publishing.
Stone-Romero, E.F. (2009). Implications of research design options for the validity of inferences derived from organizational research. In D.A. Buchanan & A. Bryman (Eds.), The
SAGE handbook of organizational research methods. London: Sage Publications.
Sy, T., Côté, S., & Saavedra, R. (2005). The contagious leader: Impact of the leader’s mood
on the mood of group members, group affective tone, and group processes. Journal of
Applied Psychology, 90(2), 295–305. doi: 10.1037/0021-9010.90.2.295
Tepper, B.J. (2007). Abusive supervision in work organizations: Review, synthesis, and
research agenda. Journal of Management, 33(3), 261–289. doi: 10.1177/0149206307300812
Thundiyil, T.G., Chiaburu, D.S., Oh, I.S., Banks, G.C., & Peng, A.C. (2015). Cynical about
change? A preliminary meta-analysis and future research agenda. Journal of Applied
Behavioral Science, 51(4), 429–450. doi: 10.1177/0021886315603122
Van Kleef, G.A., Homan, A.C., Beersma, B., Van Knippenberg, D., Van Knippenberg, B.,
& Damen, F. (2009). Searing sentiment or cold calculation? The effects of leader emotional
displays on team performance depend on follower epistemic motivation. Academy of
Management Journal, 52(3), 562–580. doi: 10.5465/AMJ.2009.41331253
Van Knippenberg, B., & Van Knippenberg, D. (2005). Leader self-sacrifice and leadership effectiveness: The moderating role of leader prototypicality. Journal of Applied
Psychology, 90, 25–37. doi: 10.1037/0021-9010.90.1.25
Venus, M., Stam, D., & Van Knippenberg, D. (2013). Leader emotion as a catalyst of effective leader communication of visions, value-laden messages, and goals. Organizational
Behavior and Human Decision Processes, 122(1), 53–68. doi: 10.1016/j.obhdp.2013.03.009
Yukl, G. (1999). An evaluation of conceptual weaknesses in transformational and charismatic leadership theories. The Leadership Quarterly, 10(2), 285–305. doi: 10.1016/
S1048-9843(99)00013-2
SCHYNS_9781785367274_t.indd 71
10/11/2017 15:19
72 Handbook of methods in leadership research
Yukl, G. (2010). Leadership in organizations (7th ed.). Upper Saddle River, NJ: Prentice-Hall.
Zhou, H., & Fishbach, A. (2016). The pitfall of experimenting on the web: How unattended
selective attrition leads to surprising (yet false) research conclusions. Journal of Personality
and Social Psychology, 111(4), 493–504. doi: 10.1037/pspa0000056
Zhou, J. (2003). When the presence of creative coworkers is related to creativity: Role of
supervisor close monitoring, developmental feedback, and creative personality. Journal of
Applied Psychology, 88(3), 413–422. doi: 10.1037/0021-9010.88.3.413
SCHYNS_9781785367274_t.indd 72
10/11/2017 15:19
4.
Assessing leadership behavior with
observational and sensor-based methods:
a brief overview
Alexandra (Sasha) Cook and Bertolt Meyer
In order to understand interactional and interpersonal behavior, researchers need to observe it (Kerlinger, 1973). This also applies to leadership
research: as leadership is a social process occurring between at least two
individuals (DeRue, 2011; DeRue & Ashford, 2010; DeRue, Nahrgang,
& Ashford, 2015), observations of leadership behavior are necessary for
understanding how leadership is exerted, how subordinates are affected
by leader behavior, and which behaviors lead to the perception of effective leadership. Accordingly, observation studies have a long tradition
in leadership research (e.g., Bass, 1949, 1954). Possible applications of
observations in leadership research span from leaderless group discussions
in laboratory settings (Bass, 1949) to field research on behavior within
­existing leadership hierarchies (Luthans & Lockwood, 1984).
However, despite their long tradition, today’s leadership researchers rarely observe leadership anymore (Gioia & Sims, 1986). Between
1985 and 2009, less than 2 percent of the studies in The Leadership
Quarterly employed observational methods (Hiller, DeChurch, Murase,
& Doty, 2011). Hence, today’s mainstream leadership research typically
tries to capture leadership behavior through questionnaires (e.g., Van
Knippenberg & Sitkin, 2013), resulting in the assessment of leadership
behaviors through subordinate perceptions (Yukl, 2013). Several researchers have criticized that this practice is unable to capture the interactive and
interpersonal nature of leadership (e.g., Luthans & Lockwood, 1984; Sims
& Manz, 1984) and that survey methods are prone to systematic biases
due to the subjectivity of perceptions (Hansbrough, Lord, & Schyns, 2015)
and due to implicit leadership theories (e.g., Lord, Foti, & De Vader, 1984;
Melwani, Mueller, & Overbeck, 2012).
The call for more applications and developments of behavioral methods
and measures in leadership research has been getting increasingly louder
(Baumeister, 2016; Baumeister, Vohs, & Funder, 2007; Furr, 2009).
Furthermore, insights into actual leadership behavior are not only valuable for researchers, but also for the application in practice, for example
in team training (Taylor, Russ-Eft, & Chan, 2005) and in many leadership
73
SCHYNS_9781785367274_t.indd 73
10/11/2017 15:19
74 Handbook of methods in leadership research
training programs employing behavior modeling for improving leadership skills (Santos, Caetano, & Tavares, 2015). With the advent of new
sensor-based methods for behavior observation such as wearable sociometric badges (Olguín Olguín, 2007) and the development of new software
solutions for analysing behavioral data (Mangold, 2014), we believe that
observational data is now more accessible for research than ever and will
thus experience a renaissance.
This chapter aims at providing an overview of current techniques for
observing and analysing behavior in leadership research and at giving
recommendations for applying observational methods in future research.
First, we describe existing coding systems that assess leadership on the
basis of specific predefined sets of behaviors. Leadership coding systems
provide researchers and observers with instructions on which behaviors to
record. We review the coding systems regarding their content, recording
techniques, and quality criteria. Subsequently, we turn to issues of coder
training and reliability. This first section of the chapter provides researchers with an idea of how to conduct observations and highlights some issues
and limitations regarding observational methods.
In the second section, we present new technologies for capturing and
recording behavior and novelties regarding the analysis of behavioral
data. Specifically, we introduce sociometric badges and motion sensor
recording, and illustrate their application in leadership research by providing recent examples. Subsequently, we introduce software that facilitates
the coding process and the analysis of behavioral data. The second section
aims at giving a thorough insight into recent developments, their respective limitations, and the ways in which these new technologies address the
limitations of traditional observational methods. The chapter concludes
with recommendations for researchers and with an outlook on the future
(and a possible renaissance) of observational research.
OBSERVATIONAL METHODS IN LEADERSHIP
RESEARCH
Which behaviors lead to the perception of an individual as a leader? Which
specific behaviors influence a subordinate’s perception of effective leadership? How should leaders behave to influence team effectiveness? How
can someone improve their leadership? The search for effective leadership
behaviors has resulted in a variety of classifications and typologies of
behaviors that are assumed to constitute leadership behavior (Fleishman
et al., 1991). However, the behavioral categories in these taxonomies are
often described on a very abstract level (Yukl, 2013) or are defined in the
SCHYNS_9781785367274_t.indd 74
10/11/2017 15:19
Observational and sensor-based methods ­
75
form of, often positive, perceptions of outcomes (Meyer et al., 2016). To
give an example, idealized influence, a subcategory of transformational
leadership is described as “behavior, that increases follower identification
with the leader” (Yukl, 2013, p. 313). This, however, is not a description
of leader behavior, but a description of the perception of a positive effect
of the leader’s behavior (Van Knippenberg & Sitkin, 2013). A similar
phenomenon exists in research on emergent leadership: when assessing
who emerges as a leader in a group, researchers operationalize emergent
leadership almost exclusively through the team members’ perceptions,
neglecting the processes that lead to these perceptions (Guastello, 2007).
Additionally, questionnaires require the participant to recall past behavior, which makes the data susceptible to biases and errors (Eby, Cader, &
Noble, 2003; Hansbrough et al., 2015; Rush, Thomas, & Lord, 1977).
The popularity of survey research on leadership behavior may have contributed to inhibiting research on more specific behaviors (Yukl, 2012),
for example through observations. The absence of observational methods
can be, at least to a certain extent, explained by the fact that behavioral
coding is both time-consuming and labor intensive (e.g., Eby et al., 2003).
Additionally, methods for coding and analysing behavioral data are
complex and are rarely taught in organizational behavior classes.
The coding systems and automated behavior recording methods presented in this chapter focus on behavioral expressions of leadership
behavior in a way that conceptualizes behavior as visible conduct and
interaction (Bonito & Sanders, 2011). They capture more or less narrowly
defined sets of behaviors, which allows researchers to describe sequences
of interactional events that are visible and can be measured in an objective way. This is an important precondition for sensor-based methods. If
we take a weekly team meeting with a supervisor as an example, observational methods allow capturing behaviors such as whether the supervisor
provided positive or negative feedback, to whom he or she provided the
feedback, or, on a more basic level, the turn taking, speaking time and listening. Observations may thus provide a more detailed picture of the communication between individuals compared to post hoc recollections by the
meeting participants. Additionally, these specific behaviors can provide
starting points for leadership development and training. In many commercial training programs, the importance of leader–follower communication
is emphasized and specific communication behaviors are practiced (Frese,
Beimel, & Schoenborn, 2003). Therefore, research on actual leader communication behaviors could provide important insights for professionals
in the field of coaching and human resource development.
Coding systems and automated behavior recordings have the potential
to address the questions asked at the beginning of the section. We proceed
SCHYNS_9781785367274_t.indd 75
10/11/2017 15:19
76 Handbook of methods in leadership research
with presenting coding systems that allow the structured observation and
recording of leadership behavior in the following section before reviewing
sensor technologies in the subsequent one.
Classical Approaches: Leadership Coding Systems
The use of coding systems (or coding schemes) in leadership research
dates back to early studies on the leaderless group discussion technique
in the first half of the twentieth century. The leaderless group discussion
was developed and used as a personnel selection technique in order to
evaluate candidates for leadership positions, especially in military contexts
(Ansbacher, 1951; Bass, 1954). The basic procedure is simple: participants
discuss a given topic while observers monitor the discussion and rate or
code the participants’ behavior. The observers themselves do not take part
in the discussion (Bass, 1954). While the first rating systems included very
global (and subjective) ratings of behavior, such as “Who do you think
led the discussion?” (Bass, 1949, p. 529), accompanied by the recording
of speaking times (Bass, 1949), the systems became more elaborate and
focused by coding more specific descriptions of behavior such as “clearly
defined or outlined the problem” (Bass, 1954, p. 468) as predictors for
future leadership status or leadership perceptions.
In general, leadership coding systems aim at assessing leadership in
terms of visible, and therefore observable behavior between at least two
individuals. This behavior is either observed on-site or recorded for later
analysis. By providing predefined behavior categories, often accompanied
by concrete behavioral examples, coding schemes guide the observer by
telling him or her what to look for. The observer classifies the behavior
according to categories or codes and records their occurrences. The list
of concrete behaviors that are to be coded, which are typically grouped
into broader categories, is either derived from a theoretical framework
(Bienefeld & Grote, 2014; Eby et al., 2003) or grounded in previous
unstructured observations (Luthans & Lockwood, 1984). Regardless
of their theoretical foundation, all systems intend to assess behaviors
­constituting leadership or are typical for leaders.
In this chapter, we review coding systems that aim at assessing leadership itself as a variable, meaning that the results of the observations intend
to indicate the degree to which an observed individual showed leadership
behavior or acted as a leader. Table 4.1 lists example coding systems that
can be applied across different professions, because their behavioral codes
are not formulated with regard to a specific professional context. Other
coding systems are designed to assess leadership in specific professional
contexts such as medical teams (e.g., Künzle et al., 2010; Parker, Yule,
SCHYNS_9781785367274_t.indd 76
10/11/2017 15:19
77
SCHYNS_9781785367274_t.indd 77
10/11/2017 15:19
Observation coding scheme for leader
verbal behavior
Formal leadership
hierarchies
Formal leadership
hierarchies
Formal leadership
hierarchies
Formal leadership
hierarchies
Leaderless teams
Leaderless teams
Leaderless teams
Context
0.71–0.97b
0.86
0.81
0.65–0.98
0.07–0.97
0.90
0.94
Interrater reliability
(Cohen’s Kappa)
Notes:
a. Coding systems designed for the application in laboratory studies are marked with a grey background. Coding systems with a white
background are designed for the application in specific field contexts.
b. Computed with Spearman–Brown interrater reliability.
Meinecke, Klonek, & Kauffeld
(2016)
Sims & Manz (1984)
Komaki, Zlotnick, & Jensen
(1986)
Luthans & Lockwood (1984)
Lord (1977)
Act4leadership
Observational Inventory for Leader
Behaviors (OILB)
Emergent leadership behaviors – coding
sheet
Functional leadership – behavior
coding system
Operant Supervisory Taxonomy and
Index (OSTI)
Leadership Observation System (LOS)
Eby et al. (2003)
Foti & Hauenstein (2007)
Name
Examples of existing leadership behavior coding systems including application contexts and reliability valuesa
Source
Table 4.1
78 Handbook of methods in leadership research
Flin, & McKinley, 2011) or cockpit and cabin crews (Bienefeld & Grote,
2014).
We proceed with reviewing the kinds of behaviors that are typically
included in leadership coding systems, before turning towards formal
issues of observations based on the different schemes. In order to structure
the content of the coding systems, we refer to the hierarchical taxonomy
of leadership behaviors according to Yukl, Gordon, and Taber (2002),
as several recent systems use this taxonomy as a theoretical framework
(see Bergman, Rentsch, Small, Davenport, & Bergman, 2012; Bienefeld &
Grote, 2014; Eby et al., 2003 for examples).
Task-oriented behavior in leadership coding systems
Similar to taxonomic approaches in descriptions of leadership behavior
(Fleishman et al., 1991), most leadership coding systems categorize leadership behavior into two broad categories: task-oriented leadership and
relation-oriented leadership (Yukl et al., 2002). Task-oriented behaviors
include behaviors such as “assigning tasks to subordinates, maintaining
definite standards of performance, asking subordinates to follow standard
procedures” (Yukl, 2013, p. 64) and are included in almost all leadership coding systems. An overview and examples of task-related behavior
assessed by leadership coding systems is given in Table 4.2.
Yukl and colleagues (2002) divide task-related leadership behavior into
the three subcategories: short-term planning, clarifying responsibilities,
and monitoring operations and performance. Short-term planning includes
decisions on “what to do, how to do it, who will do it, and when it will be
done” (Yukl et al., 2002, p. 18). In leadership coding systems, short-term
planning behaviors include making decisions on the next steps and communicating these decisions to others (e.g., Bienefeld & Grote, 2014). It can also
include the proactive gathering of information (Bienefeld & Grote, 2014;
Crockett, 1955), for example, to identify obstacles (Lord, 1977).
Clarifying responsibilities refers to communicating the planning process
and its results to others (Yukl, 2012). It includes the communication of
information that is relevant for carrying out tasks (Foti & Hauenstein,
2007), assigning tasks and responsibilities to individuals (e.g., Künzle et
al., 2010), and the distribution and management of resources (Künzle et
al., 2010; Parker et al., 2012). Behavior descriptions in leadership coding
systems often show elements of both short-term planning behaviors and
clarifying responsibilities in combination. This is because planning is a
cognitive process (i.e., thinking about possibilities, weighing options) and
can only be observed when its results are openly communicated. Examples
include describing a plan to others and proposing individual tasks for
carrying out the plan (Yukl et al., 2002).
SCHYNS_9781785367274_t.indd 78
10/11/2017 15:19
79
SCHYNS_9781785367274_t.indd 79
10/11/2017 15:19
Determining the sequence of action, coordinating the pace and timing of activities (Bienefeld & Grote, 2014; Künzle et
al., 2010)
Making an informed judgment based on information, situation and risk (Parker, Yule, Flin, & McKinley, 2012)
Seeking to obtain information of an objective, factual or technical nature (Crockett, 1955)
Presenting factual information regarding the purpose of the task (Foti & Hauenstein, 2007)
Giving information, questioning about knowledge (Meinecke et al., 2016)
An individual delegates tasks or roles to somebody else (Bienefeld & Grote, 2014; Künzle et al., 2010)
Providing information relevant to carrying out actions, removing obstacles (Lord, 1977)
Providing structure to the situation and monitoring the group’s progress towards task completion (Eby et al., 2003)
Behaviors that reinforce standards (Parker, Yule, Flin, & McKinley, 2013)
Validating or helping a specific person in the group; showing concern for others’ feelings and ideas, behaving politely
to others (Eby et al., 2003)
Indicating gratitude, general satisfaction, or positive affect; complimenting group performance or output; general
courtesies; friendly behavior (Lord, 1977)
Instructing others on how a task or procedure should be done or provides clarification about decisions or plans
(Bienefeld & Grote, 2014)
Asking others for an opinion (Bienefeld & Grote, 2014)
Orienting employees, arranging for training seminars; clarifying roles, duties, job descriptions; coaching, acting as
a mentor, “walking” subordinates through tasks; helping subordinates with personal development plans (Luthans &
Lockwood, 1984)
Stressing the importance of goals; exhorting group members to work harder; making rewards contingent upon good
task performance; complimenting an individual’s task performance (Lord, 1977)
Giving feedback to others (Bienefeld & Grote, 2014)
Planning and organizing
Note: Task-oriented coding categories are marked by a grey background. Sections in italics are subcategories of behavior, which are
independently coded.
Recognizing others
Motivating
Consulting
Training/developing
Developing a positive
group atmosphere
Coaching
Decision making
Gathering information
Giving information/
explaining
Assigning work
Managing resources
Monitoring
Maintaining standards
Sensitivity
Definitions/Descriptions
Task- and relation-oriented behaviors and their descriptions in leadership coding systems
Behavior Category
Table 4.2
80 Handbook of methods in leadership research
Several leadership coding systems for applied settings, such as management, surgery, and aircraft, assess monitoring of operations and performance (see Bienefeld & Grote, 2014; Luthans & Lockwood, 1984; Parker
et al., 2012 for examples). The descriptions of these behaviors are similar
to each other across these systems, as the authors assume that leaders
control the specific action or performance of other focal individuals
(Bienefeld & Grote, 2014; Luthans & Lockwood, 1984) or of the entire
team (Eby et al., 2003).
Relation-oriented behaviors in leadership coding systems
Although every leadership coding system included in Table 4.1 assesses at
least some form of task-related behavior, not all of them capture relationoriented behaviors. Relation-oriented, team-focused (Hu et al., 2015), or
socio-emotional behaviors (Lord, 1977) are comparable to consideration
behaviors (Fleishman, 1957) and supportive leadership (House, 1971).
They include attending to followers (Judge, Piccolo, & Ilies, 2004) or the
team (Lord, 1977) and have been linked to follower satisfaction and performance (see Judge et al., 2004 for a meta-analysis). The behavior coding
systems vary regarding the subcategories of relation-oriented behaviors
(Table 4.2). However, almost all coding systems include behaviors corresponding to the four central categories: supporting, consulting, developing, and recognizing as described by Yukl et al. (2002).
Supporting behaviors imply attending to “the needs and feelings of
other people” (Yukl et al., 2002, p. 20). Example categories for supporting leadership behavior include “fulfilling non-task needs to members,”
“developing a positive group atmosphere” (Lord, 1977, p. 122), and the
sensitivity and team-building categories of the Observational Inventory of
Leader Behaviors (OILB; Eby et al., 2003). Both coding systems differentiate between supporting behavior that is directed at single individuals and
supporting behavior that is directed at the entire group.
Consulting behavior refers to the involvement of others in decision
making (Yukl et al., 2002), for example by asking for input, allowing
input, and ensuring that every member has the opportunity to give input
(e.g., Bienefeld & Grote, 2014; Eby et al., 2003; Luthans & Lockwood,
1984; Parker et al., 2013).
Parker and colleagues (2013) assess consulting behaviors and developing behaviors in the form of one joint behavioral category that they call
guiding and supporting. In their system, developing behaviors refer to the
coaching and to the teaching of others, for example by giving instructions
on how to perform a task or procedure (Bienefeld & Grote, 2014) or by
guiding others through tasks (Luthans & Lockwood, 1984).
Recognizing behaviors involve showing appreciation for others (Yukl et
SCHYNS_9781785367274_t.indd 80
10/11/2017 15:19
Observational and sensor-based methods ­
81
al., 2002) and are described in coding systems as motivating or reinforcing
behaviors, which include giving positive feedback, and complementing or
rewarding performance (see Lord, 1977; Luthans & Lockwood, 1984 for
examples).
Leadership behavior in coding systems beyond task and relation orientation
Next to task- and relation-oriented behavior, some coding systems provide
additional categories (Table 4.3). The exact nature of these categories
depends on the theoretical framework of the respective coding system and
on the (professional) contextual focus of the given system.
One of the most common of these other categories is initiative taking.
Initiative refers to “overcoming inertia” (Lord, 1977, p. 122) – that is, that
the individual activates the group through his or her action or through
the request for an action. As inertia implies that there has been no visible
group activity before, taking initiative primarily refers to the initial action
(and not on the meaning or content of the action). In the OILB coding
scheme, however, initiative refers to “taking action” (Eby et al., 2003,
p. 1467). Here, initiative has a more task-related connotation, including
that an individual makes decisions for the group and controls the group’s
activities towards the team goal. As a third possibility, some coding
schemes conceptualize initiative in terms of independent or autonomous
working. The observed individual starts to carry out an activity or a task
without being asked, told, or being instructed (e.g., Künzle et al., 2010).
In sum, although taking initiative is recorded in several leadership coding
systems, the meaning of the category differs substantially between coding
systems.
Coding systems often include additional behavior categories with a
negative connotation. These range from giving negative feedback or evaluations (Luthans & Lockwood, 1984; Perkins, 2009) to disagreeing and
opposing (Crockett, 1955), or even to attacking (Perkins, 2009). Positive
and negative valences are sometimes combined into behavior categories,
for example by distinguishing between positive relation-oriented behaviors, such as supporting others, and negative relation-oriented behaviors, such as criticizing (Meinecke et al., 2016). In two coding systems,
Act4leadership (ibid.) and in the Leader Behavior Rating System (LBRS;
Rice & Chemers, 1975), negative behaviors fall into the same category as
self-promotion behaviors (e.g., directing attention towards oneself and
pushing own ideas on others).
In contrast to negative aspects of leader behavior, charismatic leadership hardly plays a role in leadership coding systems. The only coding
system that includes charisma as a category is the Observational Inventory
of Leader Behaviors (OILB; Eby et al., 2003). These authors define
SCHYNS_9781785367274_t.indd 81
10/11/2017 15:19
82
SCHYNS_9781785367274_t.indd 82
10/11/2017 15:19
Summarizing
Opposing
Initiative
Crockett (1955)
Eby et al. (2003)
Disagreeing/attacking
Summarizing
Prominence seeking
Summarizing the group’s progress to date
Opposition, resistance to, or disagreement with a solution,
suggestion, interpretation
Speaks first in the group, asks experimenter questions on the group’s
behalf, takes action when the group is at an impasse
Inspires the group through language and communication, generates
enthusiasm
Refers to his or her performance
Initiating an action without being asked
Requests for non-specific action, behaviors with an activational
rather than motivational connotation
Enforcing rules and policies; any formal organizational reprimand or
notice, giving negative performance feedback
Showing flexibility and changing plans if required to cope with
changing circumstances to ensure that goals are met; anticipating
possible complications and communicating them to staff
Negative emotional responses, evaluations, or attacks
Restating or enumerating content already presented
Pushing own ideas on the group, interrupting others, and directing
attention to own achievements
Description/Example Behaviors
Note: The non-italic text refers to broader descriptions of several behavioral codes, while the text in italics refers to specific examples
of behavioral codes.
Rice & Chemers (1975)
Perkins (2009)
Disciplining/punishing
Luthans & Lockwood
(1984)
Parker et al. (2013)
Coping with pressure
Reference to own performance
Initiate an action
Initiating behavior
Komaki et al. (1986)
Künzle et al. (2010)
Lord (1977)
Charisma
Behavior Categories
Examples of additional behavior categories in leadership coding systems
Source
Table 4.3
Observational and sensor-based methods ­
83
c­ harisma as a behavior that “inspires the group through language and
communication” (Eby et al., 2003, p. 1467) and that evokes enthusiasm
within the team. For assessing charisma, the OILB relies on behaviors
such as talking enthusiastically and expressing confidence and on whether
an individual succeeds in convincing others of his or her ideas.
Finally, there are leadership coding systems that include behavior categories that refer to everyday managerial activities. To give an example,
the Leadership Observation System (LOS; Luthans & Lockwood, 1984)
is based on unstructured observations of the natural activities of managers. It includes behavioral categories such as staffing (the organization
of application processes, processing paperwork) and exchanging routine
information (e.g., attending meetings).
Rating vs sampling
While the previous section addressed the content of leadership coding
systems, which is important when considering the adequacy of applying a
coding system to a research question, the following sections describe the
different methods for capturing and recording the occurring behaviors.
These methods dictate the kind of data that coding systems deliver.
In general, one can distinguish between sampling and ratings methods.
Sampling methods allow for the analysis of frequencies, sequences, and
sometimes even behavioral patterns. Sampling methods record data in
a continuous way and capture the frequencies of behaviors. Hence, each
occurrence of a certain behavior is marked and recorded as it occurs.
Leadership coding systems often apply event sampling, which focuses on
the occurrence of behaviors as “punctual, instantaneous events without
considering or recording the actual duration of the behavior” (Altmann,
1974). Researchers apply event sampling whenever the focus is on the
frequency and point of time of specific behaviors during a task or interaction. As an example, Künzle and colleagues (2010) sampled how often
members from different professional backgrounds in anesthesia teams
show problem-solving behaviors and compared their frequencies across
different stages of the task (see Bienefeld & Grote, 2014; Perkins, 2009 for
further examples). State sampling provides not only information on when
and how often certain behaviors occurred, but also with regard to the
duration of behaviors by recording the beginning and end of each behavior (Altmann, 1974). State sampling is especially important for interaction
and sequence analyses, which we describe below.
Rating systems, on the other hand, provide subjective assessments of
occurrences of behaviors, without providing information on the chronology or sequence of the behavior. They can be as simple as asking whether
a certain behavior has occurred during the observation (yes/no) (Marks &
SCHYNS_9781785367274_t.indd 83
10/11/2017 15:19
84 Handbook of methods in leadership research
Printy, 2003; Ray, Ray, Eckerman, Milkosky, & Gillins, 2011) or asking
for the degree to which the behavior descriptions fit the observed person
with Likert scales (e.g., Rice & Chemers, 1975). A more elaborate rating
method is the use of the Behaviorally Anchored Rating Scale (BARS).
The BARS allows the rating of a specific category of behavior; however,
instead of a Likert-type scale indicating the frequency or intensity of the
behavior, it provides specific behavioral examples for each rating possibility (Ohland et al., 2012). An example of a BARS that assesses the consulting behavior of managers in disabilities services is shown in Table 4.4. The
behavior descriptions aim to minimize possible biases due to ambiguity, as
they provide all observers with “a common frame of reference” (Bergman
et al., 2012, p. 25).
The different forms of sampling and rating systems affect the usability
of the coding systems for the observers and the reliability of the observations. The following sections discuss issues regarding observer training
that is necessary for objective, reliable and valid observations.
Observer training
The coding systems summarized to this point do not state how much time
the coding process takes. As a comparison, interaction coding systems that
are not specific to leadership such as the Discussion Coding System (DCS;
Schermuly & Scholl, 2011) require a ratio of 1:8 regarding the real time
captured on video and the time needed for coding. Hence, conducting an
observational study is challenging, especially for the observers. In order to
minimize biases and to increase reliability and accuracy, most researchers
train their observers. Observers, who are often student assistants, need to
be familiar with the coding system and with systematic errors and biases
that may occur during the behavior coding and rating (e.g., Luthans &
Lockwood, 1984). Therefore, in addition to learning the respective coding
procedure, observer training usually includes information on possible
distortions and biases, such as simplification, focusing on a single source,
contamination from prior information, contextual errors, prejudice and
stereotyping, and the halo effect (Thornton & Zorich, 1980) and how to
avoid them. The amount of training depends on the number of behavior
categories, the observers’ experience, and the recording methods (rating
vs sampling). Training times range from two (Parker et al., 2013) to 12
(Foti & Hauenstein, 2007) to 40 hours (Komaki et al., 1986). Observer
training is usually deemed successful if the raters reach agreement (e.g.,
Foti & Hauenstein, 2007; Komaki et al., 1986). For example, in their study
on leadership behavior among sailboat skippers (Komaki, Desselles, &
Bowman, 1989), observers received around 40 hours of training. In the
training, they learned to use the Operation Supervisory Team Taxonomy
SCHYNS_9781785367274_t.indd 84
10/11/2017 15:19
85
SCHYNS_9781785367274_t.indd 85
10/11/2017 15:19
No evidence that the manager spends any time coaching or modeling. He/she is never out of the office and when on
shift works alone. You might see not very good practice if the manager is supporting. Staff are doing things that they
shouldn’t be doing and manager is not noticing or is noticing, complaining to you about but not giving feedback to staff
Manager says that they spend time informally watching staff but can’t give very good examples of how he/she has used
that information to try to shape up performance. Their own practice if observed is ok but they don’t correct or support
other staff. They might say they talk about good practice in team meetings etc. and give some examples but there is no
sense that they model or coach staff in any way, even on an ad hoc basis
Manager does not spend time formally observing but can give examples of how they have spotted things while walking
about or working on shift and then used that information to give feedback to staff, feed into supervision etc. No use of
role play or videos in team meetings but do discuss the support for individuals. Their practice is quite good but they can’t
give good examples of how they have modeled for a particular staff or activity. Or they can give some examples but these
are about things other than active support or engagement – e.g., how to use a particular piece of equipment or how to
give medicine etc.
Manager spends some time in formal observation and giving feedback and is seen to provide modeling or coaching (or
gives a good account of how he/she has done this). However, opportunities for this are left somewhat to chance and
there is no system in place where manager/supervisor spends time regularly with each member of staff. So who receives
modeling will depend on who the manager is on shift with etc.
You observe or manager gives examples of how he/she regularly observes and works with staff. There is a system of some
sort in place to ensure that all staff are observed and receive feedback on a regular basis and manager can give good
examples of how he/she has done this
Beadle-Brown, Bigby, & Bould (2015).
1
Source:
5
4
3
2
Example Behavior Description
xample of a Behaviorally Anchored Rating Scale item for coaching behavior taken from Beadle-Brown,
E
Bigby, & Bould (2015)
Rating
Table 4.4
86 Handbook of methods in leadership research
and Index (OSTTI; Komaki et al., 1986), a coding system that includes
“a 20-page observational code, consisting of definitions, examples and
nonexamples of the taxonomic categories” (Komaki et al., 1986, p. 264) as
well as decision rules and recording instructions. Furthermore, the OSTTI
requires the observers to continuously sample behavior in one-minute
intervals. The coders observe the behavior in the first ten seconds, code
and categorize the behavior within the subsequent 40 seconds, and use
the remaining ten seconds to observe behavior to understand the context
of the following one-minute interval. The entire OSTTI coding system
consists of seven categories containing more than 40 specific behavior
descriptions (Komaki et al., 1986). Hence, the complexity of both the
behavior coding itself and the recording via sampling require an intensive
observation for achieving acceptable reliability.
Whenever possible, scholars use video recordings for observations
(Eames et al., 2010; Hu et al., 2015), often in combination with coding
software as discussed below. When video recordings are not possible, for
example due to privacy or security reasons, coders receive special training for real-time or live coding of behavior (see Bienefeld & Grote, 2014;
Curtis, Smith, & Smoll, 1979; Parker et al., 2012 for examples), which may
include field training (Komaki et al., 1986) and the use of real-time coding
software (Bienefeld & Grote, 2014) in order to enhance reliability.
Reliability and validity of leadership coding systems
Researchers can establish the reliability of observations in multiple ways.
These include the extent to which multiple observers agree on the observed
behaviors (interrater agreement or reliability) and standard reliability
measures such as re-test reliability and split-half reliability. However, interrater agreement or reliability are most common for observational data
(Mitchell, 1979). Most studies use a subset of observations for determining interrater reliability, ranging from 7 percent (Courtright, Fairhurst,
& Rogers, 1989) to 20 percent of the observations (Sims & Manz, 1984).
When using video recordings for reliability analyses, different observers
code some videos twice for computing the interrater reliability. When using
sampling techniques, a preliminary division of the video sample into behavioral units can be carried out before the observers code the units according
to the behavioral categories (see Künzle et al., 2010). If video recordings
are available, interrater reliability is computed on the basis of parallel live
observations by two coders (Komaki et al., 1986). Most leadership coding
systems exhibit good reliabilities, regardless of the kind of reliability (see
Table 4.1). However, reliability does not guarantee that a coding system
delivers a valid assessment of leadership qualities (Schermuly & Scholl,
2011). Therefore, coding systems should also be valid.
SCHYNS_9781785367274_t.indd 86
10/11/2017 15:19
Observational and sensor-based methods ­
87
Unfortunately, very few coding systems report empirical validities, and
if they do, validity is often determined through assessment of face validity by experts. As an example, the authors of the Surgeons’ Leadership
Inventory (SLI; Parker et al., 2013) asked consultants and attending surgeons to determine whether the presented behaviors represent intraoperative leadership.
Luthans and Lockwood (1984) applied a multirater approach for determining convergent and discriminant validity by comparing the results
of the field observations of managers from two sources of observations:
participant observers (the managers’ secretaries in this case), and outside
observers (graduate students in management). In addition, they compared the results from their LOS coding system to similar categories from
the Leader Behavior Description Questionnaire (LBDQ-XII; Stogdill,
1963) and to the Managerial Behavior Survey (MBS; Yukl & Nemeroff,
1979). The comparison yielded moderate convergent and discriminative
validities.
Perkins (2009) determined validity in a different way after observing
team meetings in organizations: after the observations, the head researcher
interviewed the participants and compared their answers to the meeting
profiles from the coding system. As interview statements supported the
findings from the observation, the coding system was deemed valid.
The previous examples primarily focus on external validity. However,
when deciding between using an existing system or developing a new
coding system (potentially with categories and subcategories), researchers
should consider the internal validity of the behaviors (DiStefano & Hess,
2005).
Although leadership observations with coding schemes require extensive preparation, effort, and resources, they provide valuable information
on different facets of leadership behavior and especially on the contents
of communication. Therefore, they enable the assessment of leadership
data beyond the limits of survey methods and are the method of choice in
response to recent calls for a focus on more concrete, specific, and observable behaviors in leadership research. In the following section, we introduce sensor-based behavior assessment methods, which present a recent
alternative to labor-intensive manual behavioral coding
New Approaches: Sensor-based Assessment Methods in Leadership
Research
All observational methods up to this point assess leadership behaviors
that are usually seen as typical or effective. However, whenever observers evaluate and categorize behavior shown by an individual, they
SCHYNS_9781785367274_t.indd 87
10/11/2017 15:19
88 Handbook of methods in leadership research
a­ utomatically draw conclusions about the individual’s personality and
intentions. These inferences, which can also be influenced by factors such
as the observer’s personality (Akert & Panter, 1988), are unavoidable in
interpersonal perception (Reeder & Brewer, 1979) especially when applying broadly defined behavior categories in coding systems. Hence, in order
to fully understand which specific leader behaviors evoke which effects,
for example on team performance, subordinate perceptions of leadership
efficacy, or even observer perceptions of leadership, researchers need to
analyse leadership behavior at a micro-level (Meyer et al., 2016). Microlevel behaviors, or honest signals (Pentland, 2008), are defined as “leaders’
verbal and non-verbal visible conduct and interactions with their followers
that are likely to affect followers’ attitudes and behavior” (Meyer et al.,
2016, p. 9). Analysing micro-level behaviors can complement traditional
observation and coding systems, especially when the selection of microlevel behaviors is based on broader descriptions of leadership behavior
(Meyer et al., 2016), such as in leadership coding systems.
As social interactions are a key issue in leadership (Schyns & Mohr,
2004; Yukl, 2013) and in leadership coding systems, research on microlevel leadership behaviors can benefit from current developments in the
field of social sensing (Meyer et al., 2016). Social sensing refers to the
automated recording and analysis of micro-level social interaction behavior (Schmid Mast, Gatica-Perez, Frauendorfer, Nguyen, & Choudhury,
2015). Today, a number of computer-based methods for detecting and
identifying interactional features such as body posture, eye gaze, facial
expressions, and speech qualities already exist (see Schmid Mast et al.,
2015 for a review). This data provides insights into the relations between
micro-level behavior and interpersonal perceptions, such as expressed
leadership and dominance (Sanchez-Cortes, Aran, Jayagopi, Schmid
Mast, & Gatica-Perez, 2013).
Sensor-based recording devices can be stationary, such as a laboratory
equipped with cameras and microphones or mobile, making them applicable for field research (Schmid Mast et al., 2015). Mobile recording devices
include smartphones and gaze-tracking glasses, which allow the assessment of everyday human interactions in natural settings (ibid.).
In the following section, we describe two possibilities for sensor-based
data capturing and their application possibilities in leadership research:
sociometric badges, wearable sensors capturing multiple interaction features
in field settings, and stationary motion sensors for laboratory purposes.
Capturing interactional data with sociometric badges
Sociometric badges are wearable social sensing devices that are equipped
with microphones, accelerometers, Bluetooth receivers and transmitters,
SCHYNS_9781785367274_t.indd 88
10/11/2017 15:19
Observational and sensor-based methods ­
89
and infrared sensors. With the badges and the corresponding software,
researchers can assess interaction behavior such as proximity between two
individuals and speech features (e.g., speech duration, interruptions, frequency, volume). The badges are worn around the neck by the participant
and can store up to 40 hours of data (Kim, McFee, Olguín Olguín, Waber,
& Pentland, 2012; Olguín Olguín, 2007; Olguín Olguín & Pentland, 2010).
They are therefore applicable in field studies where the observed individuals are not constantly at the same location and constant video recording or
live observations are inconvenient or impossible.
In order to evaluate the badges’ capability to capture leadership
emergence and leadership positions in a field setting, we carried out a
small pilot study in an organizational setting. We collected badge data
and questionnaire data from all 29 staff members of a single research
department at a German university over the course of two days. We
introduced the participants to the badges prior to the assessment, so
that they were aware of the sensors, the measurement variables, and the
handling of the badges (e.g., how to recharge them). At the end of both
days of the study, participants filled in an online questionnaire, which
asked for the initials of the five co-workers with whom they had the
most face-to-face communication at work. The participants also ranked
these co-workers according to the estimated amount of face-to-face
communication.
Based on this questionnaire data, we computed social networks for each
day, which we subsequently compared to the social network data that the
badges supplied. In the social network that we derived from the questionnaire data, nodes represented individual participants, and edge weights
represented daily communication frequency according to their rank order.
For these networks, we computed the in-degree centrality, closeness centrality, and betweenness centrality for directed weighted networks with
the igraph R package (Csardi & Nepusz, 2006) for each node (i.e., for
each participant). In-degree centrality describes the amount and weight of
received nominations (i.e., how often this person was named as one of the
five persons with whom another employee had the most communication).
Closeness centrality denotes the centrality of an individual on the basis
of the entire network. Finally, betweenness centrality captures how often
an individual is between two others within the network or rather how
often the individual lies on the shortest network path between two other
­individuals (Prell, 2012).
We compared the questionnaire-based centrality indices to those that
the badges deliver based on the occurrence and duration of interactions,
which are again derived from distance and proximity data from the
infra-red and Bluetooth sensors (see Olguín-Olguín, 2007 for a detailed
SCHYNS_9781785367274_t.indd 89
10/11/2017 15:19
90 Handbook of methods in leadership research
Table 4.5
ay 1 Pearson correlations between questionnaire-based and
D
badge-based centrality values
Badge-assessed Interaction Data
Degree centrality
Closeness centrality
Betweenness centrality
Day 1
Day 2
Total
Day 1
Day 2
Total
Day 1
Day 2
Total
Infrared data
Bluetooth data
Combined
0.54*
0.35
0.35
0.63*
0.63**
0.54**
0.46*
0.35
0.41*
0.43*
0.46*
0.44*
0.53*
0.56**
0.45*
0.43*
0.45*
0.54**
0.41*
0.46*
0.44*
0.53*
0.56**
0.45*
0.41*
0.42*
0.48**
Note: * p < 0.05, ** p < 0.01.
description). The software delivers the centrality values for infrared data,
Bluetooth data, and a combination of both. As not all members of the
staff were present on both days and because some participants did not fill
out the questionnaire, n 5 25 datasets from badges (n 5 25 questionnaires)
on day one and n 5 24 datasets from badges (n 5 18 questionnaires) on
day two were included in the analyses. Pearson correlations of the network
centrality data are summarized in Table 4.5. The results show significant
positive correlations, especially regarding the first day, indicating a good
validity of the badge-assessed interactions as a measure for face-to-face
interactions in an organizational field setting. Hence, the badges allow an
objective assessment of interactions and provide a more exhaustive picture
of an individual’s face-to-face contacts as it does not depend on the individual’s memory.
As the protection of privacy is an important issue in field observations,
we also assessed the degree to which the participants felt distracted or
monitored at the end of each day. Participants reported that they had a
moderate feeling of awareness of the badges during the first day of the
study, but they also indicated that they did not feel as if they were being
monitored or as if they were under surveillance.
This study illustrated the potentials of wearable social sensing devices
for leadership research as they reflect the communication within an organization (or team) and could therefore provide information on the duration
and frequency of leader–member, and team communication networks.
As a case in point, a recent study on leadership emergence employed
SCHYNS_9781785367274_t.indd 90
10/11/2017 15:19
Observational and sensor-based methods ­
91
badge-elicited speaking time and found a positive relation with leadership
perceptions (Chaffin et al., 2015). Additional potential of the sociometric
badges and similar technologies lies in the simultaneous recording of multiple data types, such as physical activity, proximity data, and speech data.
Motion sensing in leadership research
Motion sensing assesses data on body movement and posture through
videos, depth recording systems (Schmid Mast et al., 2015), or inertia
measurement units (Feese, Arnrich, Tröster, Meyer, & Jonas, 2011). As
it requires attaching sensors to the participants’ bodies, it is often conducted in laboratory settings. As the impact of posture or body language
on interpersonal perception is well documented in leadership research
(Darioly & Schmid Mast, 2014), automated motion sensing can contribute
to leadership research by identifying motion patterns and relations, such
as synchronicity, between participants.
As an example, Meyer and colleagues (2016) analysed the mediating
effect of non-verbal behavior on the relationship between participative
leadership and peer leader evaluations, as well as participative leadership
and team decision quality. The authors focused on mimicry, the mirroring of posture, movements, and gestures as an indicator of empathy.
Student teams consisting of one leader and two followers worked on a
simulated decision-making task. Prior to the task, the team leader received
leadership training, which instilled a directive or participative leadership style. Each participant wore six motion sensors on the upper arms,
wrists, head, and on the back, measuring acceleration, rate of turn and
orientation (ibid.). This allowed a digital recreation of the participants’
movements. The subsequent analysis detected the averaged times a certain
body posture by a follower was mirrored by the leader within 60 seconds
of its initial occurrence (Feese et al., 2011). Results showed that mimicry
behavior mediated the relationship between the leadership manipulation
and leadership evaluations, which was assessed with the transformational
leadership scale of the Multifactor Leadership Questionnaire (Bass &
Avolio, 1990). However, the authors found no effect of mimicry on team
decision quality.
Reliability and Validity of Sensor-based Assessment Methods
Determining the reliability and validity of sensor-based assessment
methods is highly dependent on the methods and specific devices used.
When using devices that are equipped with several different sensors,
the task is more difficult as researchers need to evaluate each component’s validity and reliability independently. Regarding reliability, both
SCHYNS_9781785367274_t.indd 91
10/11/2017 15:19
92 Handbook of methods in leadership research
­ easurement errors within (for example, due to fluctuations of measurem
ment of one single device when repeatedly exposed to the same stimuli)
and between the devices need to be evaluated for each sensor component.
Different sensitivities of error variance between the two devices’ sensors,
for example the built-in microphones, can reduce the reliability by causing
bias if not detected and taken account of (Chaffin et al., 2015). Therefore,
a starting point for researchers working with sensor-based assessment
methods should always be a structured evaluation of the sensors’ variability in order to avoid systematic measurement errors.
The validity of sensor-based assessment methods relies strongly on
the algorithms used by the software to process the raw data (ibid.).
Depending on the type of the device, the algorithms are not necessarily available or viewable. Researchers can evaluate the sensors’ validity
by matching the sensor-assessed behaviors to observer-coded behaviors
used as ground truth (Feese et al., 2011) or experimental set-ups (Chaffin
et al., 2015). While the sensing of behavioral mimicry used in the study
by Meyer and colleagues (2016) showed an average accuracy of 78.15
percent, the results regarding the sociometric badges vary. In a series
of studies conducted by Chaffin and colleagues (2015), the researchers
discovered limitations to the badges’ abilities when comparing actual
speaking time and the speaking time assessed though the badges and
the manufacturer-provided algorithm (r 5 0.15). Although the use of
an optimized algorithm improved the results (r 5 0.36, p < 0.01), the
authors value the speech detection capabilities of the badges as limited.
However, proximity sensing had a high validity (percentage agreement 5
94.4 percent). These results highlight that it is necessary to evaluate every
sensor component individually. The choice of the sensor-based assessment methods should be carefully considered regarding the environment
and setting in which the actual assessment should take place and should
be tested and validated in either the actual setting or a similar environment. Additionally, it is important to mention the importance of assessment duration when using wearable devices. Regarding the sociometric
badges the validity increased with the length of the assessment period for
all measures (Chaffin et al., 2015).
The previous examples show the ways in which recent technological
developments can facilitate observations in leadership research in both
field and laboratory settings. However, we still need human observers
and coders whenever we want to evaluate the validity of sensor-based
methods, or whenever we need to analyse the content and context of interactions. In the following sections, we review means by which the manual
behavior recording and the subsequent data analysis can be supported
through special software and software-aided analysis techniques.
SCHYNS_9781785367274_t.indd 92
10/11/2017 15:19
Observational and sensor-based methods ­
93
Software for Creating and Analysing Observational Data
The raw form of behavioral data is typically some kind of state or event
sequence (Bakeman & Gottman, 1997). Such data is a time-ordered listing
of behaviors or codes, where one line of data represents a given behavior,
accompanied by a time stamp, denoting the time of its occurrence or its
duration. On the basis of such data, researchers not only can calculate frequencies, lengths, and durations of specific behavior, but can also analyse
interactions and patterns, as we describe below. Therefore, researchers
need tools to create such data, which is commonly done with software for
video coding and annotation, and for analysing it. In the following, we
briefly describe two pieces of common software that we employ for these
tasks in our research practice.
Behavior coding and annotation with Mangold INTERACT
Commonly, video or audio recordings are the raw material for subsequent
observational assessment of (leadership) behavior. Trained coders then
use behavior coding software such as INTERACT (Mangold, 2014) to
create state- or event-sequence data from these recordings. In our own
research, we chose this particular software for two reasons: first, it is the
only software that allows opening and controlling several video files simultaneously. In our lab, we have three cameras that record digital video into
three separate files (one for each camera). The INTERACT tool allows
opening all three video files simultaneously and controls their playback
simultaneously. In other words, if the observer presses play, all three
videos play in synchrony in three separate windows. Second, INTERACT
is the only coding software that we are aware of that allows coding one
behavior as an occurrence of a main category, and to subsequently add
subcategories to that behavior. For example, when we code group interactions with the DCS coding system (Schermuly & Scholl, 2011) the main
category is the speech act. So when a person on the video starts to talk, the
observer presses a key, and when the person stops talking or is interrupted,
the observer presses the key again. This logs the beginning and the end
of the behavior “speech act,” and coders can subsequently code whether
this specific act constituted a question, a suggestion and how dominant or
friendly it was. In other words, the speech act is the main category, and
dominance, friendliness, and type are subcategories. Other simpler annotation software can only code main categories that are assigned to specific
keys on the keyboard. With such simpler systems, coders would need to
assign “speech: dominant question” and “speech: submissive question” to
different buttons and would have to remember these. With INTERACT,
coders simply need to remember the button for “speech start” and “speech
SCHYNS_9781785367274_t.indd 93
10/11/2017 15:19
94 Handbook of methods in leadership research
end.” The tool can be configured in such a way that it pauses the video
whenever a certain type of main category is logged (e.g., after someone on
the tape finished talking) and then prompts the coder to rate the subcategories. This so-called lexical coding approach distinguishes INTERACT
from other software for video coding and analysis, and has proven to be
very useful in our own research.
Analysis of state and event sequences with sequence analysis: TraMineR
The data resulting from coding or annotating video or audio recordings
is typically a sequence of events with time stamps or durations. These
ordered lists denote which of the behaviors that the given coding system
distinguishes happened when. Without any further treatment, researchers can calculate the frequencies or durations of specific behaviors.
However, this kind of data can reveal much more interesting findings,
such as patterns of reoccurring behavior and transition probabilities
(i.e., the probability of a specific behavior given the previous occurrence
of another specific previous behavior). For these kinds of questions, a
variety of methods exist (for reviews see Chiu, 2005; Fairbairn, 2016). Of
these, sequential or sequence analysis has the longest history (Bakeman
& Gottman, 1997). Sequence analysis refers to a set of methods and
analyses that allow comparing and classifying event sequences. These
include determining the similarity between sequences, the calculation
of transition probabilities, the identification of common patterns and
subsequences, and the numerical description of sequence features such
as entropy and turbulence. To give an example, imagine a researcher
who observes and codes the following behaviors: leader smiles, follower
smiles, leader asks a question, and follower shares information. The set
of these behaviors constitutes the so-called alphabet of behaviors. In the
dataset, these behaviors are typically abbreviated, e.g., ls, fs, lq, fi. In the
terminology of sequence analysis, the observed behavior from a specific
interaction constitutes an event sequence, e.g., fs-ls-fs-lq. Researchers
can analyse such event sequences with the (free) TraMineR package
(Gabadinho, Ritschard, Müller, & Studer, 2011) in the R environment
(R Development Core Team, 2015). For the given example, a sequence
analysis with TraMineR can reveal how likely it is that a follower shares
information after the leader smiles (in contrast to after the leader asks a
question), and whether these transition probabilities differ for different
circumstances. With TraMineR, researchers can even identify the most
common subsequences from a set of event sequences, and can perform
cluster analyses over a set of event sequences. While INTERACT also
allows performing some of these analyses, we feel that a combination of
INTERACT (for annotating the videos) and R (for analysing the result-
SCHYNS_9781785367274_t.indd 94
10/11/2017 15:19
Observational and sensor-based methods ­
95
ing sequence data) is the most powerful combination for conducting
observational studies.
CONTRASTING CLASSIC METHODS AND
NEW DEVELOPMENTS – ADVANTAGES AND
LIMITATIONS
In this chapter, we reviewed both established and novel behavior observation methods. Leadership coding systems aim at observing and recording
typical leadership behavior. Although many of them include similar or
comparable behavior dimensions, coding systems have been designed for
different applied contexts. Applying coding systems requires well-trained
observers and coders who are able to interpret and classify behaviors
according to the categories of the respective coding systems. Therefore,
behavioral assessments of leadership behavior are somewhat costly to
obtain.
Sensor-based observation methods can add additional insight to leadership research by capturing behavior on the micro-level. Visible verbal and
non-verbal behavior parameters can be automatically and simultaneously
recorded and evaluated, and are rapidly available for further analyses.
We presented two examples of how sensor-based assessment of leadership
behavior can be included in organizational field settings and in laboratory settings. We further discussed and introduced new developments
that facilitate the conduction of observations with behavioral coding
and ­introduced the coding software INTERACT and sequence analysis
software.
In summary, a variety of methods are available for researchers who plan
to include observational methods in their research. However, researchers
need to carefully consider the selection of each method before conducting
their study. As our review has shown, leadership coding systems allow
the analysis of the behavioral content (e.g., what is communicated during
the interaction). Although there is some overlap between the behavioral
dimensions in the different systems or schemes, one should keep in mind
that some behavior categories or descriptions were designed for specific
contexts. Therefore, the behavioral descriptions should be reviewed thoroughly regarding their applicability and transferability to the planned
research context. Furthermore, the respective recording technique (sampling vs rating) is important as it affects the usability and the reliability
of the observations. Additionally, observers need sufficient preparation
and criteria for assessing the accuracy of observers should be defined prior
to the coding procedure. An important issue also pertains to establishing
SCHYNS_9781785367274_t.indd 95
10/11/2017 15:19
96 Handbook of methods in leadership research
blindness of the observers for the hypotheses. Blindness of the observers
towards the research question is necessary in order to enhance objectivity of the observers while coding and for reducing biases (Bienefeld &
Grote, 2014; Eby et al., 2003). Whenever video recordings are possible,
we recommend the use of coding software to facilitate the coding process.
Similar planning should be carried out prior to the use of sensor-based
methods. Conducting research in field settings can limit the options for
social sensing options. For example, when observing employees who work
in different rooms on different tasks (such as an office and a laboratory),
stationary devices are often not appropriate. Even more importantly,
researchers should base the selection of the captured verbal and nonverbal micro-level behaviors on prior theory as much as possible. As social
sensing methods do not provide information on the behavioral content,
but on the ways in which the content is communicated in interactions, we
recommend deriving the specific micro-level behaviors for analyses from
the broader behavior descriptions of leadership coding systems. Whenever
possible, we recommend a combination of both observer-based and
sensor-based methods whenever the research context and the provided
resources make it possible.
When conducting behavioral observations, the issue of privacy arises –
especially for wearable sensors. The devices capture behavioral data as it
occurs and are not able to distinguish between work and non-work activities. We therefore recommend a thorough introduction of the recording
functions of social sensing devices to participants. Furthermore, researchers should seek written consent from the participants for using video and
sensor data in scientific studies.
Despite their advantages in comparison to questionnaire-based measures of leadership behavior, observational measures come with their
own set of limitations. To state the obvious, behavioral observational
methods that rely on trained observers are quite time-consuming and
costly. They require intensive preparation and elaborate data analysis
procedures. Observational studies also pose a challenge regarding issues
of reliability and validity, as the application of other methods besides the
determination of the interrater reliability, such as multitrait-multimethod
(MTMM) analyses, are rare. Although methods of social sensing provide
fast access to the captured data, they also require computational skills and
knowledge in order to be carried out correctly. Technical difficulties and
the loss of data can impact the analyses. Therefore, wearable sensors
and computer-based sensing techniques are great opportunities for interdisciplinary collaborations between social scientists and scientists from
IT-related disciplines.
SCHYNS_9781785367274_t.indd 96
10/11/2017 15:19
Observational and sensor-based methods ­
97
FUTURE DEVELOPMENTS
In order to understand what actually happens when people lead and how
leadership and leadership perceptions interact, observational methods
provide an informational value that goes beyond the possibilities of survey
methods.
The time and effort required to conduct observational studies can
be reduced by focusing on so-called thin slices of behavior (Ambady &
Rosenthal, 1992). Thin slices are very short segments (e.g., 6–30 seconds;
Ambady & Rosenthal, 1993) of expressive behavior, which have been
shown to be able to predict various outcomes, for example in negotiations (Curhan & Pentland, 2007) and regarding perceptions of leadership
(Tskhay, Xu, & Rule, 2014). Due to the capability of social sensing
methods to record behavior as it occurs, these methods may facilitate
the analyses of the temporal development and dynamic aspects of leadership behavior and leadership perceptions, such as changes in the quality
and quantity of exerted behavior, which would be of special interest for
the research areas of informal leadership development and leadership
emergence in groups. Additionally, as interpersonal interactions often
include different types of behavior, researchers can analyse combinations
of these behaviors (Sanchez-Cortes et al., 2013) in order to gain insight
into the way different behaviors are processed in order to shape interpersonal perceptions. Nevertheless, the type of social sensing method should
be carefully considered regarding the study’s setting and nature of the
measured behavior and pretested to secure the applicability (Chaffin et
al., 2015). We recommend testing the devices both in a set-up that is as
similar as possible to the planned study setting (environment and number
of subjects) and by comparing the captured behaviors to manual observer
recordings. Additionally, researchers should conduct a systematic evaluation of the variability between the devices (both within and between)
through systematic manipulation of the environment in an experimental
setting.
Wearable sensors have the ability to steer sensor-based assessment
methods out of the laboratory and into the field. In order to secure a
congruency between the behaviors captured by technological devices and
the requirements of leadership research, researchers should start including
sensor-based assessment into research, or at least expose themselves to the
possibilities of this technology. By actively taking part in the development
processes, researchers can help developers and provide new methods for
future leadership research.
SCHYNS_9781785367274_t.indd 97
10/11/2017 15:19
98 Handbook of methods in leadership research
REFERENCES
Akert, R.M., & Panter, A.T. (1988). Extraversion and the ability to decode nonverbal communi­cation. Personality and Individual Differences, 9(6), 965–972. doi: 10.1016/0191-8869(88)
90130-4
Altmann, J. (1974). Observational study of behavior: Sampling methods. Behaviour, 49(3),
227–266. doi: 10.1163/156853974X00534
Ambady, N., & Rosenthal, R. (1992). Thin slices of expressive behavior as predictors of
interpersonal consequences: A meta-analysis. Psychological Bulletin, 111(2), 256–274. doi:
10.1037/0033-2909.111.2.256
Ambady, N., & Rosenthal, R. (1993). Half a minute: Predicting teacher evaluations from
thin slices of nonverbal behavior and physical attractiveness. Journal of Personality and
Social Psychology, 64(3), 431–441. doi: 10.1037/0022-3514.64.3.431
Ansbacher, H. (1951). The history of the leaderless group discussion technique. Psychological
Bulletin, 48(5), 383–391. Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/14875813
Bakeman, R., & Gottman, J.M. (1997). Observing interaction: An introduction to sequential
analysis (2nd ed.). Cambridge, MA: Cambridge University Press.
Bass, B.M. (1949). An analysis of the leaderless group discussion. Journal of Applied
Psychology, 33(6), 527–533. doi: 10.1037/h0058164
Bass, B.M. (1954). The leaderless group discussion. Psychological Bulletin, 51(5), 465–492.
doi: 10.1037/h0056881
Bass, B.M., & Avolio, B.J. (1990). Manual for the multifactor leadership questionnaire. Palo
Alto, CA: Consulting Psychology Press.
Baumeister, R.F. (2016). Charting the future of social psychology on stormy seas: Winners,
losers, and recommendations. Journal of Experimental Social Psychology. Advance online
publication. doi: 10.1016/j.jesp.2016.02.003
Baumeister, R.F., Vohs, K.D., & Funder, D.C. (2007). Psychology as the science of selfreports and finger movements: Whatever happened to actual behavior? Perspectives on
Psychological Science, 2(4), 396–403. doi: 10.1111/j.1745-6916.2007.00051.x
Beadle-Brown, J., Bigby, C., & Bould, E. (2015). Observing practice leadership in intellectual
and developmental disability services. Journal of Intellectual Disability Research, 59(12),
1081–1093. doi: 10.1111/jir.12208
Bergman, J.Z., Rentsch, J.R., Small, E.E., Davenport, S.W., & Bergman, S.M. (2012). The
shared leadership process in decision-making teams. Journal of Social Psychology, 152(1),
17–42. doi: 10.1080/00224545.2010.538763
Bienefeld, N., & Grote, G. (2014). Shared leadership in multiteam systems: How cockpit
and cabin crews lead each other to safety. Human Factors, 56(2), 270–286. doi:
10.1177/0018720813488137
Bonito, J.A., & Sanders, R.E. (2011). The existential center of small groups: Member’s conduct
and interaction. Small Group Research, 42(3), 343–358. doi: 10.1177/1046496410385472
Chaffin, D., Heidl, R., Hollenbeck, J.R., Howe, M., Yu, A., Voorhees, C., & Calantone, R.
(2015). The promise and perils of wearable sensors in organizational research. Organizational
Research Methods. Advance online publication. doi: 10.1177/1094428115617004
Chiu, M.M. (2005). A new method for analyzing sequential processes: Dynamic multilevel
analysis. Small Group Research, 36(5), 600–631. doi: 10.1177/1046496405279309
Courtright, J.A., Fairhurst, G.T., & Rogers, L.E. (1989). Interaction patterns in organic and
mechanistic systems. Academy of Management Journal, 32(4), 773–802. doi: 10.2307/256568
Crockett, W.H. (1955). Emergent leadership in small, decision-making groups. Journal of
Abnormal Psychology, 51(3), 378–383. doi: 10.1037/h0046109
Csardi, G., & Nepusz, T. (2006). The igraph software package for complex network research.
Retrieved from https://pdfs.semanticscholar.org/1d27/44b83519657f5f2610698a8ddd177
ced4f5c.pdf
Curhan, J.R., & Pentland, A. (2007). Thin slices of negotiation: Predicting outcomes from
conversational dynamics within the first 5 minutes. Journal of Applied Psychology, 92(3),
802–11. doi: 10.1037/0021-9010.92.3.802
SCHYNS_9781785367274_t.indd 98
10/11/2017 15:19
Observational and sensor-based methods ­
99
Curtis, B., Smith, R.E., & Smoll, F.L. (1979). Scrutinizing the skipper: A study of leadership behaviors in the dugout. Journal of Applied Psychology, 64(4), 391–400. doi:
10.1037/0021-9010.64.4.391
Darioly, A., & Schmid Mast, M. (2014). The role of nonverbal behavior in leadership: An
integrative review. In R.E. Riggio and S.J. Tan (Eds.), Leader interpersonal and influence skills: The soft skills of leadership (pp. 73–100). New York: Taylor and Francis. doi:
10.4324/9780203760536
DeRue, D.S. (2011). Adaptive leadership theory: Leading and following as a complex adaptive
process. Research in Organizational Behavior, 31, 125–150. doi: 10.1016/j.riob.2011.09.007
DeRue, D.S., & Ashford, S. (2010). Who will lead and who will follow? A social process of
leadership identity construction in organizations. Academy of Management Review, 35(4),
627–647. doi: 10.5465/AMR.2010.53503267
DeRue, D.S., Nahrgang, J.D., & Ashford, S.J. (2015). Interpersonal perceptions and
the emergence of leadership structures in groups: A network perspective. Organization
Science, 26(4), 1192–1209. doi: 10.1287/orsc.2014.0963
DiStefano, C., & Hess, B. (2005). Using confirmatory analysis for construct validation: An
empirical review. Journal of Psychoeducational Assessment, 23(3), 225–241.
Eames, C., Daley, D., Hutchings, J., Whitaker, C.J., Bywater, T., Jones, K., & Hughes, J.C.
(2010). The impact of group leaders’ behaviour on parents’ acquisition of key parenting
skills during parent training. Behaviour Research and Therapy, 48(12), 1221–1226. doi:
10.1016/j.brat.2010.07.011
Eby, L.T., Cader, J., & Noble, C.L. (2003). Why do high self-monitors emerge as leaders in
small groups? A comparative analysis of the behaviors of high versus low self-monitors.
Journal of Applied Social Psychology, 33(7), 1457–1479. doi: 10.1111/j.1559-1816.2003.
tb01958.x
Fairbairn, C. (2016). A nested frailty approach for small group behavioral observation data. Small Group Research. Retrieved from http://journals.sagepub.com/doi/
abs/10.1177/1046496416648778
Feese, S., Arnrich, B., Tröster, G., Meyer, B., & Jonas, K. (2011). Detecting posture mirroring in social interactions with wearable sensors. Proceedings – International Symposium on
Wearable Computers, ISWC, 119–120. doi: 10.1109/ISWC.2011.31
Fleishman, E.A. (1957). A leader behavior description for industry. In R. Stogdill & A.E.
Coons (Eds.), Leader behavior: Its description and measurement (pp. 103–119). Columbus:
Ohio State University.
Fleishman, E.A., Mumford, M.D., Zaccaro, S.J., Levin, K.Y., Korotkin, A.L., & Hein, M.B.
(1991). Taxonomic efforts in the description of leader behavior: A synthesis and functional
interpretation. The Leadership Quarterly, 2(4), 245–287. doi: 10.1016/1048-9843(91)90016-U
Foti, R.J., & Hauenstein, N.M.A. (2007). Pattern and variable approaches in leadership emergence and effectiveness. Journal of Applied Psychology, 92(2), 347–355. doi:
10.1037/0021-9010.92.2.347
Frese, M., Beimel, S., & Schoenborn, S. (2003). Action training for charismatic leadership:
Two evaluations of studies of a commercial training module on inspirational communication of a vision. Personnel Psychology, 56(3), 671–697. Retrieved from bschool.nus.
edu/. . ./frese%20beimel%20train%20charisma%20person%20psych03.pdf
Furr, R.M. (2009). Personality psychology as a truly behavioural science. European Journal
of Personality, 2(5), 369–401. doi: 10.1002/per.724
Gabadinho, A., Ritschard, G., Müller, N.S., & Studer, M. (2011). Analyzing and visualizing
state sequences in R with TraMineR. Journal of Statistical Software, 40(4), 1–37. Retrieved
from http://www.jstatsoft.org/v40/i04
Gioia, D.A., & Sims, H.P. (1986). Cognition–behavior connections: Attribution and verbal
behavior in leader–subordinate interactions. Organizational Behavior and Human Decision
Processes, 37(2), 197–229. doi: 10.1016/0749-5978(86)90052-X
Guastello, S.J. (2007). Non-linear dynamics and leadership emergence. The Leadership
Quarterly, 18(4), 357–369. doi: 10.1016/j.leaqua.2007.04.005
Hansbrough, T.K., Lord, R.G., & Schyns, B. (2015). Reconsidering the accuracy of
SCHYNS_9781785367274_t.indd 99
10/11/2017 15:19
100 Handbook of methods in leadership research
follower leadership ratings. The Leadership Quarterly, 26(2), 220–237. doi: 10.1016/j.
leaqua.2014.11.006
Hiller, N.J., DeChurch, L.A., Murase, T., & Doty, D. (2011). Searching for outcomes
of leadership: A 25-year review. Journal of Management, 37(4), 1137–1177. doi:
10.1177/0149206310393520
House, R.J. (1971). A path goal theory of leader effectiveness. Administrative Science
Quarterly, 16(3), 321–339.
Hu, Y.-Y., Parker, S.H., Lipsitz, S.R., Arriaga, A.F., Peyre, S.E., Corso, K.A.,. . .Greenberg,
C.C. (2015). Surgeons’ leadership styles and team behavior in the operating room. Journal
of the American College of Surgeons, 222(1), 41–51. doi: 10.1016/j.jamcollsurg.2015.09.013
Judge, T.A., Piccolo, R.F., & Ilies, R. (2004). The forgotten ones? The validity of consideration and initiating structure in leadership research. The Journal of Applied Psychology,
89(1), 36–51. doi: 10.1037/0021-9010.89.1.36
Kerlinger, F.N. (1973). Foundations of behavioral research. New York: Holt, Rinehart &
Winston.
Kim, T., McFee, E., Olguín Olguín, D., Waber, B., & Pentland, A. (2012). Sociometric
badges: Using sensor technology to capture new forms of collaboration. Journal of
Organizational Behavior, 33(3), 412–427. doi: 10.1002/job
Komaki, J.L., Desselles, M.L., & Bowman, E.D. (1989). Definitely not a breeze: Extending
an operant model of effective supervision to teams. Journal of Applied Psychology, 74(3),
522–529. doi: 10.1037/0021-9010.74.3.522
Komaki, J.L., Zlotnick, S., & Jensen, M. (1986). Development of an operant based taxonomy and observation index of supervisory behavior. Journal of Applied Psychology,
71(2), 260–269.
Künzle, B., Zala-Mezo, E., Wacker, J., Kolbe, M., Spahn, D.R., & Grote, G. (2010).
Leadership in anaesthesia teams: The most effective leadership is shared. Quality & Safety
in Healthcare, 19(6), e46. doi: 10.1136/qshc.2008.030262
Lord, R.G. (1977). Functional leadership behavior: Measurement and relation to social
power and leadership perceptions. Administrative Science Quarterly, 22(1), 114–132. doi:
10.2307/2391749
Lord, R.G., Foti, R.J., & De Vader, C.L. (1984). A test of leadership categorization theory:
Internal structure, information processing, and leadership perceptions. Organizational
Behavior and Human Performance, 34(3), 343–378. doi: 10.1016/0030-5073(84)90043-6
Luthans, F., & Lockwood, D.L. (1984). Toward an observation system for measuring leader
behavior in natural settings. In J.G. Hunt, D. Hosking, C. Schriesheim, & R. Stewart
(Eds.), Leaders and managers (pp. 117–141). New York: Pergamon Press.
Mangold. (2014). INTERACT quick start manual V2.4. Retrieved from www.mangoldinternational.com
Marks, H.M., & Printy, S.M. (2003). Principal leadership and school performance: An
integration of transformational and instructional leadership. Educational Administration
Quarterly, 39(3), 370–397. doi: 10.1177/0013161X03253412
Meinecke, A.L., Klonek, F.E., & Kauffeld, S. (2016). Using observational research methods
to study voice and silence. German Journal of Research in Human Resource Management.
Retrieved from http://journals.sagepub.com/doi/abs/10.1177/2397002216649862
Melwani, S., Mueller, J.S., & Overbeck, J.R. (2012). Looking down: The influence of
contempt and compassion on emergent leadership categorizations. Journal of Applied
Psychology, 97(6), 1171–1185. doi: 10.1037/a0030074
Meyer, B., Burtscher, M.J., Jonas, K., Feese, S., Arnrich, B., Tröster, G., & Schermuly,
C.C. (2016). What good leaders actually do: Micro-level leadership behaviour, leader
evaluations, and team decision quality. European Journal of Work and Organizational
Psychology. Advance online publication. doi: 10.1080/1359432X.2016.1189903
Mitchell, S.K. (1979). Interobserver agreement, reliability, and generalizability of
data collected in observational studies. Psychological Bulletin, 86(2), 376–390. doi:
10.1037/0033-2909.86.2.376
Ohland, M.W., Loughry, M.L., Woehr, D.J., Bullard, L.G., Felder, R.M., Finelli,
SCHYNS_9781785367274_t.indd 100
10/11/2017 15:19
Observational and sensor-based methods ­
101
C.J.,. . .Schmucker, D.G. (2012). The comprehensive assessment of team member effectiveness: Development of a behaviorally anchored rating scale for self- and peer evaluation.
Academy of Management Learning & Education, 11(4), 609–630. doi: 10.5465/amle.2010.0177
Olguín Olguín, D. (2007). Sociometric badges: Wearable technology for measuring human
behavior. Master’s thesis. Retrieved from http://dspace.mit.edu/handle/1721.1/42169
Olguín Olguín, D., & Pentland, A. (2010). Sensor-based organisational design and engineering. International Journal of Organisational Design and Engineering, 1(1/2), 69–97. doi:
10.1504/IJODE.2010.035187
Parker, S.H., Flin, R., McKinley, A., & Yule, S. (2013). The Surgeons’ Leadership Inventory
(SLI): A taxonomy and rating system for surgeons’ intraoperative leadership skills.
American Journal of Surgery, 205(6), 745–751. doi: 10.1016/j.amjsurg.2012.02.020
Parker, S.H., Yule, S., Flin, R., & McKinley, A. (2011). Towards a model of surgeons’
leadership in the operating room. BMJ Quality & Safety, 20(7), 570–579. doi: 10.1136/
bmjqs.2010.040295
Parker, S.H., Yule, S., Flin, R., & McKinley, A. (2012). Surgeons’ leadership in the operating room: An observational study. American Journal of Surgery, 204(3), 347–354. doi:
10.1016/j.amjsurg.2011.03.009
Pentland, A.S. (2008). Honest signals: How they shape our world. Cambridge, MA: MIT
Press.
Perkins, R.D. (2009). How executive coaching can change leader behavior and improve
meeting effectiveness: An exploratory study. Consulting Psychology Journal: Practice and
Research, 61(4), 298–318. doi: 10.1037/a0017842
Prell, C. (2012). Social network analysis. London: Sage.
Ray, R.D., Ray, J.M., Eckerman, D.A., Milkosky, L.M., & Gillins, L.J. (2011). Operations
analysis of behavioral observation procedures: A taxonomy for modeling in an expert training system. Behavior Research Methods, 43(3), 616–634. doi: 10.3758/s13428-011-0140-6
R Development Core Team. (2015). R: A language and environment for statistical computing, version 3.2.3 computer software. Retrieved from http://www.r-project.org
Reeder, G.D., & Brewer, M.B. (1979). A schematic model of dispositional attribution in interpersonal perception. Psychological Review, 86(1), 61–79. doi: 10.1037/0033-295X.86.1.61
Rice, R.W., & Chemers, M.M. (1975). Personality and situational determinants of leader
behavior. Journal of Applied Psychology, 60(1), 20–27. doi: 10.1037/h0076362
Rush, M., Thomas, J., & Lord, R.G. (1977). Implicit leadership theory: A potential threat
to the internal validity of leader behavior questionnaires. Organizational Behavior and
Human Performance, 20(1), 93–110.
Sanchez-Cortes, D., Aran, O., Jayagopi, D.B., Schmid Mast, M., & Gatica-Perez, D.
(2013). Emergent leaders through looking and speaking: From audio-visual data to multimodal recognition. Journal on Multimodal User Interfaces, 7(1), 39–53. doi: 10.1007/
s12193-012-0101-0
Santos, J.P., Caetano, A., & Tavares, S.M. (2015). Is training leaders in functional leadership
a useful tool for improving the performance of leadership functions and team effectiveness? Leadership Quarterly, 26(3), 470–484. doi: 10.1016/j.leaqua.2015.02.010
Schermuly, C.C., & Scholl, W. (2011). IKD – Instrument zur Kodierung von Diskussionen [The
Discussion Coding Scheme]. Göttingen, Germany: Hogrefe.
Schmid Mast, M., Gatica-Perez, D., Frauendorfer, D., Nguyen, L., & Choudhury, T. (2015).
Social sensing for psychology: Automated interpersonal behavior assessment. Current
Directions in Psychological Science, 24(2), 154–160. doi: 10.1177/0963721414560811
Schyns, B., & Mohr, G. (2004). Non-verbal elements of leadership behaviour. German
Journal of Research in Human Resource Management, 18(3), 289–305. Retrieved from
http://journals.sagepub.com/doi/abs/10.1177/239700220401800303
Sims, H.P., & Manz, C.C. (1984). Observing leader behavior: Toward reciprocal determinism in leadership theory. Journal of Applied Psychology, 6(2), 222–232. doi:
10.1037/0021-9010.69.2.222
Stogdill, R.M. (1963). Manual for the Leader Behavior Description Questionnaire – Form XII.
Columbus, OH: Bureau of Business Research, Ohio State University.
SCHYNS_9781785367274_t.indd 101
10/11/2017 15:19
102 Handbook of methods in leadership research
Taylor, P.J., Russ-Eft, D.F., & Chan, D.W.L. (2005). A meta-analytic review
of behavior modeling training. Journal of Applied Psychology, 90(4), 692–709. doi:
10.1037/0021-9010.90.4.692
Thornton, G.C., & Zorich, S. (1980). Training to improve observer accuracy. Journal of
Applied Psychology, 65(3), 351–354. doi: 10.1037/0021-9010.65.3.351
Tskhay, K.O., Xu, H., & Rule, N.O. (2014). Perceptions of leadership success from nonverbal cues communicated by orchestra conductors. The Leadership Quarterly, 25(5),
901–911. doi: 10.1016/j.leaqua.2014.07.001
Van Knippenberg, D., & Sitkin, S.B. (2013). A critical assessment of charismatic – transformational leadership research: Back to the drawing board? The Academy of Management
Annals, 7(1), 1–60. doi: 10.1080/19416520.2013.759433
Yukl, G. (2012). Effective leadership behavior : What we know and what questions
need more attention. Academy of Management Perspectives, 26(4), 66–85. doi: 10.5465/
amp.2012.0088
Yukl, G. (2013). Leadership in organizations. Harlow, UK: Pearson Education.
Yukl, G., & Nemeroff, W. (1979). Identification and measurement of specific categories of
leadership behavior: A progress report. In J.G. Hunt & L.L. Larson (Eds.), Crosscurrents
in leadership (pp. 164–200). Carbondale, IL: Southern Illinois University Press.
Yukl, G., Gordon, A., & Taber, T. (2002). A hierarchical taxonomy of leadership behavior:
Integrating a half century of behavior research. Journal of Leadership & Organizational
Behavior, 9(4), 15–32. doi: 10.1177/107179190200900102
SCHYNS_9781785367274_t.indd 102
10/11/2017 15:19
5.
The contribution of sophisticated facial
expression coding to leadership research
Savvas Trichas
INTRODUCTION
Previous studies have shown that facial expressions can exert substantial
influence on general interpersonal perception (Zebrowitz & Montepare,
2008), and specifically on leadership perception (Stewart, 2010). Quite a
few professions entail the expression and keen perception of emotion as
part of their work role such as doctors, sales people, funeral directors,
flight attendants, police interrogators, and bill collectors (Hochschild,
1983; Rafaeli & Sutton, 1987; Sutton & Rafaeli, 1988). A number
of leadership studies have stressed the significance of facial displays
(e.g., Bucy, 2000; Bucy & Bradley, 2004; Masters & Sullivan, 1989).
Specifically, the importance of leaders’ emotional displays is emphasized in research on educational, organizational, and political leadership
(Beatty, 2001; Bono & Ilies, 2006; Humphrey, Pollack, & Hawver, 2008;
Stewart, Bucy, & Méhu, 2015). However, our knowledge regarding the
impact of leaders’ facial expressions on others’ perceptions of leadership
is still limited.
Although there is already a considerable amount of research on leaders’
emotional displays (e.g., Bucy & Bradley, 2004; Lewis, 2000) the majority
of studies ignore the benefits that can emerge from incorporating facial
expression analysis techniques currently used in the field of facial expression research (Ekman, Friesen, & Hager, 2002). The few studies that have
used sophisticated facial expression analysis in the context of leadership
demonstrate the importance of accurately describing facial expressions
when interpreting results. In such studies (Stewart, Waller, & Schubert,
2009; Trichas & Schyns, 2012) emphasis was placed on the accuracy of
describing facial expressions in order to increase the depth of analysis.
Research has shown that subtle differences between facial expressions can
result in different perceptions (Surakka & Hietanen, 1998), highlighting
the importance of accurate descriptions of facial displays. Furthermore,
beyond visual coding, there are also temporal aspects that need to be
considered, such as onset, offset, and apex duration, that all significantly
influence observers’ evaluations of facial expressions (see Krumhuber &
103
SCHYNS_9781785367274_t.indd 103
10/11/2017 15:19
104 Handbook of methods in leadership research
Kappas, 2005; Krumhuber, Manstead, Cosker, Marshall, & Rosin, 2009;
Krumhuber, Manstead, & Kappas, 2006).
The aim of the present work is to contribute to our understanding of the
interaction between leadership and leaders’ facial expressions. Specifically,
it is argued that the area of leadership could benefit from the incorporation
of detailed facial expression coding techniques. Sample research questions
that leadership scholars can investigate with the use of such methods are:
(1) What are the intensity ranges of leaders’ appropriate facial behavior
within specific organizational scenarios and how are these related with
leadership perceptions? Exploration of such issues is interesting as awareness of appropriateness of expression intensity ranges could contribute to
leadership perceptions, while being inattentive to the margins set by these
ranges could mean the opposite. (2) How do facial expressions that appear
on leaders’ faces for less than a second impact observers’ reactions? Leaders
who are aware of the influence of such brief displays to observers’ reactions
may be tuned to their own emotions and be more efficient in connecting
with their followers. (3) How is the temporal development, from onset
to apex, related to the perceived authenticity of a leader’s facial display?
Taking into consideration that authentic expressions are linked with leadership influence (Newcombe & Ashkanasy, 2002) exploring facial expressions’ temporal aspects could provide leaders with a deeper understanding
of the impact of their (and others’) facial expressions on characteristics
related with leadership perceptions such as trustworthiness, thus increasing their communicational precision. The integration of such methods can
eventually contribute not only to increasing research awareness but also to
improving our communicative knowledge in the context of leadership. The
above questions are a small sample of research potential created with the
use of detailed facial coding, and each can be explored by incorporating
sophisticated facial coding analysis into leadership research designs. The
increased precision of analysis may ultimately help to uncover new angles
that will add to our understanding of leadership processes.
To support the above argument, in the following sections, literature
from two areas is reviewed, namely leadership perception and facial
expression. After a general introduction to facial expressions and perceptual concepts, the main literature on leaders’ facial expressions is discussed. Additionally, an argument is constructed that supports the use of
sophisticated facial expression coding analysis to increasing credibility of
leadership research designs. Leadership studies that included such sophisticated facial expression techniques are then presented with an extended
discussion of the research of Trichas and Schyns (2012). Finally, the
general discussion and conclusions follow together with implications and
recommendations for leadership research and practice.
SCHYNS_9781785367274_t.indd 104
10/11/2017 15:19
Sophisticated facial expression coding ­105
THE COMMUNICATIVE FUNCTION OF FACIAL
EXPRESSIONS
Human beings begin to understand the value of facial expression early on
in life. Even infants are able to perceive faces and react to the facial expressions they observe (Field, Woodson, Greenberg, & Cohen, 1982; Tronick,
1989). The way the face functions is fascinating – many muscles together
transmit complex meanings by simple movement combinations. Facial
expression is a non-verbal communication channel that receives a lot of
research attention since it gathers together the vast majority of the sensory
organs, plus the brain, in one region (Cohn & Ekman, 2008).
Darwin (1872 [1962]) posits that while facial actions were originally
displayed as elements of, or indications of, survival movements, such
as protecting the eyes or uncovering the teeth to prepare an attack,
such behaviors have evolved into complex mechanisms of transmitting
emotions and interpersonal intentions. Parkinson (2005) stresses the
power of facial expressions to convey not just emotion-relevant information but also social intentions and practical meanings. These practical
meanings and inferences of intentions during social encounters are constructed automatically, effortlessly, and can be based on little information
(Schneid, Carlston, & Skowronski, 2015; Uleman, Saribay, & Gonzalez,
2008; Wang, Xia, & Yang, 2015). This is because there is a need to make
sense of information received during communication, which forces the
human brain to assign justification labels (Hassin, Bargh, & Uleman,
2002).
Drawing from Darwin’s seminal work, modern emotion research converges to the point that emotional facial expressions are used to communicate information relevant to the behavioral tendencies of the transmitter,
thus influencing person perception and impression formation (Adams,
Ambadi, Macrae, & Kleck, 2006; Cañadas, Lupiáñez, Kawakami,
Niedenthal, & Rodríguez-Bailón, 2016; Hendriks & Vingerhoets, 2006).
Put differently, observers use facial expressions to decode others’ emotions but also to assume intentions and permanent behavioral characteristics (Franklin & Zebrowitz, 2013; Fridlund, 1994; Hess, Blairy, &
Kleck, 2000; Montepare & Dobish, 2003; Shariff & Tracy, 2011; Tiedens,
2001). In addition, ecological theory maintains that facial expressions
of emotion are processed in an adaptive manner, through expectations
generated from interactions with other similar facial expressions, in order
to facilitate people’s adaptation to their social environment (Zebrowitz
& Montepare, 2008). Facial displays signal underlying emotions but
also have an approach, attack, or avoidance component that leads to
meaningful impressions (Marsh, Ambady, & Kleck, 2005; McArthur &
SCHYNS_9781785367274_t.indd 105
10/11/2017 15:19
106 Handbook of methods in leadership research
Baron, 1983; Monahan, 1998). In other words, the observation of facial
expressions generates overgeneralized, biased impressions (Zebrowitz &
Montepare, 2005). For example, when someone sees a person displaying
anger, besides inferring the transmitter’s emotional state at the moment,
the attack mode assumed by observers may be viewed as a characteristic of an aggressive, domineering, and dominant individual (Hareli &
Hess, 2010; Madera & Smith, 2009; Montepare & Zebrowitz-McArthur,
1998).
Facial displays influence perceivers’ reactions even when exposed for
a very short amount of time. Short-lived facial displays, often referred
to as micro-expressions, frequently range from 1/5th to 1/25th of a
second and are often associated with concealment of emotion (Ekman,
1992, 2001, 2003b; Hurley, Anker, Frank, Matsumoto, & Hwang, 2014).
Based on the premise that facial muscles are linked to brain activity, when
emotions are experienced, respective muscle movements may involuntarily appear on the face for a fraction of a second (Bhushan, 2015; Frank
& Ekman, 1997; Jenkins & Johnson, 1977). For instance, a person can be
smiling, change to a facial expression of disgust, and back to smiling in
less than a second. When micro-expressions are captured and presented
in slow motion or static frames, they can have a clear emotional meaning
to observers (Ekman, 2003a). However, these brief facial expressions are
often difficult to detect and may affect individuals without their awareness
(Winkielman & Berridge, 2004). Specifically, a considerable number of
studies demonstrate that brief exposure to facial displays of basic emotions
such as happiness, anger, and fear can influence observers’ reactions on a
subconscious level (Channouf, 2000; Marsh & Ambady, 2007; Perron,
Roy-Charland, Chamberland, Bleach, & Pelot, 2016; Winkielman, &
Berridge, 2003, 2004). Moreover, prior research indicates that brief exposure to facial expressions (as much as 30 milliseconds) can significantly
influence observers’ facial muscle tone, dermal reactions, cardiovascular
responses, and can cause alterations in amygdala regional blood flow
(Dimberg, Thunberg, & Elmehed, 2000; Wild, Erb, & Bartels, 2001;
Winkielman et al., 2005).
To conclude, facial displays and their impact on observers’ perceptions is an area that has been receiving increased research attention
within the past decades. Based on the last few paragraphs it seems that
facial expressions displayed at normal speed or in less than a second
significantly impact perceivers’ reactions. In the subsequent section,
the traditional research on emotions and leaders’ facial expressions is
reviewed.
SCHYNS_9781785367274_t.indd 106
10/11/2017 15:19
Sophisticated facial expression coding ­107
TRADITIONAL RESEARCH ON EMOTIONS AND
LEADERS’ FACIAL EXPRESSIONS
The significance of leaders’ facial expressions has been highlighted in a
variety of studies (e.g., Schyns & Morh, 2004; Stewart, 2010; Trichas,
2015). Three different types of studies are most pertinent to the perception
of leadership from facial expressions: political leaders’ emotional displays
(e.g., Bucy, 2000; Sullivan & Masters, 1988), leaders’ general emotional
displays (e.g., Damen, Van Knippenberg, & Van Knippenberg, 2008;
Glomb & Hulin, 1997), and exploration of the influence of facial displays
on the perception of traits linked with leadership such as charisma, power,
dominance, and trustworthiness (e.g., Awamleh & Gardner, 1999; Lau,
1982; Mazur & Mueller, 1996). In the following paragraphs, the traditional research on emotions and leaders’ facial expressions is introduced.
In the research on political leaders’ displays of emotion, observers were
asked to assess videos from renowned politicians’ public appearances
(e.g., Bill Clinton in Bucy & Newhagen, 1999; Barack Obama in Gong &
Bucy, 2016; George W. Bush in Cherulnik, Donley, Wiewel, & Miller,
2001). For instance, Bucy and Newhagen (1999) explored the influence of
video-recorded presidential displays, in the context of public appearance,
on recollection, thinking elaboration, and appropriateness assessments
of respondents. The respondents of the aforementioned study considered
negative and low-intensity leader displays more appropriate and hence
evaluated them more favorably. On the contrary, positive or intense
leader reactions contradicted participants’ expectations and were categorized as inappropriate (Bucy & Newhagen, 1999). Also, Gong and Bucy
(2016) used television news excerpts of the 2012 US presidential debates to
investigate the effect of leader appropriateness on observers’ non-verbal
expectancies. Their results indicated that inappropriate facial expressions
boosted viewers’ visual attention and educed negative reactions. Finally,
Cherulnik et al. (2001) asked observers to evaluate both professional
actors performing a fake campaign speech, and politicians (Bill Clinton
and George Bush) manipulating charisma-related facial behavior factors
such as frequency and valence of transmitters’ smiling, and duration and
frequency of eye contact with the audience. Their results show that leaders’
charismatic behavior leads to emotional contagion, which refers to the
tendency to automatically mimic and synchronize body movements, such
as facial expressions with those of another individual and, therefore, to
converge emotionally (Barsade, 2002; Howard & Gengler, 2001; Sy, Cȏté,
& Saavedra, 2005; Wild et al., 2001). Specifically, the participants in these
studies revealed higher levels of the behaviors exhibited by c­ harismatic
leaders, thus supporting emotional contagion effects.
SCHYNS_9781785367274_t.indd 107
10/11/2017 15:19
108 Handbook of methods in leadership research
A second line of research investigated the impact of leaders’ general
emotional displays on participants’ evaluations (e.g., Damen et al., 2008;
Gaddis, Connelly, & Mumford, 2004; Glomb, & Hulin, 1997), using
mainly manipulated facial expressions performed by leaders. A number
of studies found that observers evaluated leaders’ negative emotional
expressions negatively (Gaddis et al., 2004; Glomb & Hulin, 1997; Lewis,
2000; Medvedeff, 2008) and others showed that observers evaluated
leaders’ positive emotional expressions positively (Awamleh & Gardner,
1999; Medvedeff, 2008). For example, Newcombe and Ashkanasy (2002)
showed their participants video recordings of leaders providing positive
or negative feedback and displaying facial expressions compatible or
incompatible with the message communicated. They found that positive
feedback and facial expressions congruent with the verbal message communicated resulted in a more positive evaluation of the respective leader’s
negotiating latitude than negative or message-incongruent expressions.
In another study using an actor displaying angry or happy facial expressions, Van Kleef et al. (2009) showed that motivation to learn influenced
which type of affective expressions improved team task performance: participants high in epistemic motivation performed better when the leader
showed anger expressions. When motivation to learn was low, the participants showed better task performance after viewing happy expressions.
Furthermore, Cherulnik et al. (2001) found that emotional variables such
as intensity and frequency of smiles and eye gaze predicted perception of
charismatic leadership. Melwani, Mueller, and Overbeck (2012) investigated the effect of compassion and contempt on evaluations of leadership.
Specifically, participants’ social perception, leadership judgments, and discrete emotion ratings were evaluated in an interview context. Participants’
discrete emotions were evaluated taking into account facial expression,
verbal tone, and body language. Their results indicate that conveying
emotions of contempt and compassion is positively related to leadership
perceptions. Finally, Medvedeff (2008) experimented with leaders’ negative feedback from a web camera. The manipulation of affective displays
included, again: facial features, voice, and gestures for positive, negative,
and neutral affective displays. The participants judged positive affective
feedback more positively and negative affective feedback more negatively.
Besides the research mentioned above, a third line of studies used facial
expressions to investigate leadership-related traits such as dominance
(Keating, Mazur, & Segall, 1981; Kraus & Chen, 2013; Mazur & Mueller,
1996; Montepare & Dobish, 2003), status (Keating, 2003; Keating, Mazur,
& Segall, 1977), power (Dovidio, Heltman, Brown, Ellyson, & Keating,
1988), trustworthiness (Krumhuber, Manstead, Cosker, Marshall, &
Rosin, 2007), and charisma (Awamleh & Gardner, 1999; Shea & Howell,
SCHYNS_9781785367274_t.indd 108
10/11/2017 15:19
Sophisticated facial expression coding ­109
1999). For example, Kraus and Chen (2013) found that professional fighters during a prefight stare-down with their challenger were perceived as
less dominant when they smiled than when their expression was neutral.
In addition, Awamleh and Gardner (1999) linked leaders’ smiles in the
context of giving a speech with the perception of charisma. Finally,
Montepare and Dobish (2003) presented untrained actors posing emotions to participants and asked them to evaluate the actors in terms of
emotions and trait impressions. Their findings showed that the emotion
displayed in facial expressions shifted impressions of dominance and
affiliation. Specifically, happy and surprised facial expressions increased
perceived dominance and affiliation, angry facial expressions increased
perceived dominance and decreased perceived affiliation, and sad and fear
expressions decreased perceived dominance.
The studies reviewed in the last few paragraphs contributed to our
understanding of the perception of leaders’ emotional expressions (e.g.,
Bucy, 2000; Lewis, 2000; Mazur & Mueller, 1996). However, our knowledge regarding the actual impact of leaders’ facial displays is still restricted
as the traditional research in the field of leadership has neglected the added
contribution that can emerge from the integration of sophisticated facial
action coding analysis (e.g., Hess et al., 2000; Knutson, 1996). Specifying
exact facial muscle movement, intensity level, and timing can increase
depth of analysis, hence expanding the range of research angles to be
explored. To illustrate with an example, the use of detailed facial expression coding techniques enables scholars to determine whether a negative
leader display (e.g., Bucy & Newhagen, 1999) involves eyebrows pulling
together (indicator of anger) or eyebrows raising and pulling together
(indicator of sadness), identifying level of intensity (from a single trace to
maximum muscle activity), and determining the progress of expressions
in time from onset to offset. All three factors mentioned above have been
shown to significantly influence perceivers’ reactions, so by accurately
specifying them and/or controlling for them could add to the precision
of leadership research (e.g., Ambadar, Cohn, & Reed, 2009; Trichas &
Schyns, 2012). In the following section, basic concepts in measurement of
facial muscle movements are presented and the importance of integrating
sophisticated facial expression analysis in leadership research is discussed.
MEASURING FACIAL EXPRESSION
There are three main types of methods for measuring facial expression:
facial electromyography (EMG), automatic facial image analysis, and
manual facial expression coding (Cohn & Ekman, 2008). EMG involves
SCHYNS_9781785367274_t.indd 109
10/11/2017 15:19
110 Handbook of methods in leadership research
the use of electrodes in order to measure the activity of the facial muscles.
Although this method allows researchers to detect very subtle, non-visible
facial muscle movements, the attachment of electrodes to the area of the
face inhibits naturalistic facial expression and restricts researchers’ ability
to use the instrument outside laboratory settings. Automatic facial image
analysis uses computerized coding of facial expressions, which requires
a lesser degree of manual processing (Littlewort et al., 2011). Automatic
analysis is a promising field that can be applied with reasonable accuracy
but it is still a developing methodology that needs to better address issues
such as the evaluation of the full range of facial action units, intensities,
and temporal aspects (Valstar, Méhu, Jiang, Pantic, & Scherer, 2012).
Manual coding is the most frequent method used to measure facial
behavior in facial expression research. It allows for both live observation
and pre-recorded behavior (videos or photos) with high levels of validity
(Cohn & Ekman, 2008) and reliability (Sayette, Cohn, Wertz, Perrott, &
Parrot, 2001). The focus in this chapter is on manual coding.
According to Cohn and Ekman (2008), there are three aspects of facial
expressions that can and should be assessed in order to increase credibility
in describing facial expressions: type, intensity, and timing. Type indicates the facial muscle that moved. When a facial expression appears, we
can observe changes caused by muscle activity underneath the skin from
non-static facial parts such as eyebrows, cheeks, lips, chin, the eyelids,
and so on (Ekman et al., 2002). An evident facial expression is the movement observed as a result of modification of muscle positioning under
the skin and the wrinkling, pouching, and bulging that result from this
modification. For a smiling facial display, type would be described as:
pulling of the corners of the lips back and obliquely upward, deepening
of the oblique lines between the cheek and the mouth starting from the
nostril corners, deepening of the line below the lower eyelid before the
cheekbone, raising of the area between the oblique lines between the cheek
and the mouth starting from the nostril corners and below the lower eyelid
before the cheekbone (infraorbital triangle; see Ekman et al., 2002), skin
bagging under the lower eyelid, production of crow’s feet wrinkles at the
corners of the eyes, narrowing the eye aperture, possible nostril raising
and widening, and possible chin boss skin stretching (ibid.). Besides type,
a facial muscle movement might also differ in intensity. Intensity refers to
the strength of muscle movement in terms of the changes to facial appearance. Sophisticated facial expression coding distinguishes five different
levels of intensity: A – Trace, B – Slight, C – Marked, D – Extreme, and
E – Maximum (ibid.). Consequently, the degree of the deepening of the
oblique lines between the cheek and the mouth starting from the nostril
corners (nasolabial furrow; ibid.) in the smiling display described above
SCHYNS_9781785367274_t.indd 110
10/11/2017 15:19
Sophisticated facial expression coding ­111
Smile 1: low intensity
Figure 5.1
Smile 1: high intensity
Low-intensity versus high-intensity smiling facial expressions
can range from trace to maximum, indicating a lower or higher intensity
level respectively (Figure 5.1).
Finally, timing refers to the temporal development of the expression
from onset, to apex, and back to the expression’s offset. Specifically, from
one facial expression to another there might be differences in onset and/or
offset speed, and general duration (Ambadar, Cohn, & Reed, 2009). The
depth of analysis regarding the timing parameter increases when researchers explore individual facial action units (e.g., lip corner pull) as they can
investigate if these appear simultaneously or sequentially.
Sophisticated Facial Expression Description and the Facial Action
Coding System
There are a number of systems used for the coding of facial expression.
Such examples are the Affective Expressions Scoring System (AFFEX;
Izard, Dougherty, & Hembree, 1983) and the Maximally Discriminative
Facial Movement Coding System (MAX; Izard, 1983). Last, there is the
Facial Action Coding System (FACS; Ekman et al., 2002), considered to be
the most comprehensive system for coding and decoding facial expression
(Cohn, Ambadar, & Ekman, 2007). The basic methodology of these systems
is to describe changes in terms of facial muscle movement by viewing
recordings of facial behavior such as photographs or videos. Although the
information reported below regarding FACS is fundamental in understanding the underlying philosophy of the specific manual coding technique, the
next few paragraphs cannot teach the reader how to code facial behavior.
Understanding that facial expressions are the result of combinations
of facial muscle movements, Ekman and his colleagues constructed the
Facial Action Coding System (FACS; Ekman et al., 2002) to account
SCHYNS_9781785367274_t.indd 111
10/11/2017 15:19
112 Handbook of methods in leadership research
for every visible movement. FACS is one of the most widely used facial
expression coding instruments in the behavioral sciences (Littlewort et
al., 2011). It is an anatomically based instrument that is used to describe
all observable muscle movement changes in the appearance of the face
(Ekman et al., 2002). In order to adequately code an expression, observation of the facial behavior in slow motion and at a frame-by-frame rate
is required (Cohn, Zlochoher, Lien, & Kanade, 1999). FACS objectively
identifies 44 basic facial muscle movements called action units (AU), and
sketches a compound system for evaluating their occurrence (Ekman &
Rosenberg, 1997; Sayette et al., 2001). In other words, with the use of
the above instrument, coders can mark nearly all visible changes in facial
appearance as a result of facial muscle movement. The marking of these
changes includes all three aspects of facial expressions mentioned earlier
(type, intensity, and timing) as facial muscle movement, level of intensity,
and timing of expression are assessed in FACS coding. Also, the use of
a neutral frame of the face without any expression serves as a reference
point or a baseline, and is considered critical in pinpointing the exact facial
muscle movement and intensity (Figure 5.2).
In particular, FACS coding involves identifying the muscles that moved
in the frame with the facial expression and comparing the appearance
to the frame with the neutral face results. For example, in Figure 5.2, a
careful comparison of the frame with the facial expression to the frame
depicting the actor’s neutral face reveals: severe raising of the inner brow
corner and pulling together of eyebrows, extreme cheek raising and eyelid
tightening, slight lifting of the upper lip at an angle, and slight deepening
Neutral face
Figure 5.2
SCHYNS_9781785367274_t.indd 112
Facial expression
he neutral face as a reference point for identifying exact
T
facial muscle movement and intensity
10/11/2017 15:19
Sophisticated facial expression coding ­113
of the nasolabial furrow. Finally, the use of FACS allows for identification of AU combinations, which are argued to be the most frequent signs
of basic emotions such as happiness, anger, fear, contempt, surprise,
sadness, and disgust (Ekman et al., 2002). Based on the coded description
of Figure 5.2, it is also suggested that the facial expression contains indicators of sadness and distress. Because of its high level of validity and reliability (Cohn & Ekman, 2008; Sayette et al., 2001), FACS has been used
in several research areas including leadership, psychology, neuroscience,
forensics, and computer graphics (see Sayette et al., 2001; Trichas, 2015).
Importance of Sophisticated Facial Expression Description in
Leadership Research
As pointed out earlier in this chapter, modern FACS involves the marking
of exact facial muscle movement, intensity, and timing (Ekman et al.,
2002). The example below illustrates the relevance of such sophistication in coding for leadership research: as reported earlier in the section
regarding traditional research on emotions and leaders’ facial expressions,
Newcombe and Ashkanasy (2002) found that positive and messagecongruent facial expressions resulted in a more positive evaluation of
the respective leader’s negotiating latitude than negative and messageincongruent expressions. In terms of facial muscle movement, intensity,
and timing, the term “positive facial expression” is open to more than
one interpretation. Concerning facial muscle movement, sample results
reveal that subjects display more positive reactions to smiles with cheek
raising and crow’s feet wrinkles at the corners of the eyes (also called
Duchenne smiles) than they did when observing smiles without the cooccurring cheek and eye movements (e.g., Gunnery & Hall, 2014; Surakka
& Hietanen, 1998). Such results are relevant for leadership research as they
reveal that accurately identifying facial muscle movement may be crucial
to how a leader is perceived. The significance of intensity is also supported
empirically. For instance, research shows that observers’ formulate intensity expectations for facial expressions according to specific information
set by communication context (Trichas & Schyns, 2012). For example,
beholders might have quite different expectations of smile intensity from
a leader greeting an employee from when the same leader is meeting an
old friend. Leadership scholars can explore such research hypotheses, thus
adding new areas of investigation to the leadership literature. Finally,
with regard to the timing factor, Krumhuber et al. (2009) suggested that
the temporal development of dynamic facial expression from onset to
offset plays a significant role in the perceptual process. For example,
Krumhuber and Kappas (2005) showed that slower onsets in smiling facial
SCHYNS_9781785367274_t.indd 113
10/11/2017 15:19
114 Handbook of methods in leadership research
expressions are linked with more genuine impressions. Considering that
leader authenticity is an area that has always been considered important
to the concept of leadership (Bono & Ilies, 2006), linking temporal aspects
of facial expression with leader authenticity might be another significant
contribution of sophisticated facial coding analysis.
To conclude, all the examples of positive facial expressions mentioned
above might be labeled with a word or phrase, such as “smile.” The
word smile gives the reader a main idea of what observers see in the
facial action combinations involving a pull of the lips’ corners back and
obliquely upward. However, it is vital to understand that when dealing
with facial expressions, the perceptual impact of a smile could be different
as a result of minor differences in muscle movement, intensity, or timing
(e.g., Krumhuber & Kappas, 2005; Surakka & Hietanen, 1998; Trichas &
Schyns, 2012). Consequently, the identification of these facial expression
aspects in the coding procedure could contribute to increasing the profundity of analysis and the accuracy of results, thus adding to the credibility
of leadership research on facial expressions (see Rosenberg, 2005).
LEADERSHIP STUDIES USING SOPHISTICATED
FACIAL CODING ANALYSIS
To the author’s knowledge, there are only a few studies that use detailed
facial coding analysis to explore the influences of facial expressions in the
context of leadership (Stewart et al., 2009; Stewart & Dowe, 2013; Stewart,
Bucy, & Méhu, 2015; Stewart, Méhu, & Salter, 2015; Tsai et al., 2016). To
begin with, Stewart et al. (2009) conducted a study on political leadership
that used sophisticated facial expression coding to investigate observers’
responses. Specifically, in their study, smiling micro-momentary expressions (very briefly exposed emotional facial displays; see Ekman, 2009)
were isolated and removed from former US President George W. Bush’s
speech. These briefly exposed expressions depicting muscle movement
with emotional meaning were identified by coding the video footage of the
former president’s facial behavior in slow motion and at a frame-by-frame
rate using FACS. The study showed that removing the specific smiling
frames (positive micro-expressions) resulted in the respondents experiencing more feelings of anger and threat, thus indicating that viewers’ impressions can be influenced by leaders’ facial expressions of emotion even if
these last less than a second.
In a different study by Stewart and Dowe (2013), sophisticated facial
expression coding was also employed to determine the facial actions displayed during emotional expressions exhibited by President Obama. Their
SCHYNS_9781785367274_t.indd 114
10/11/2017 15:19
Sophisticated facial expression coding ­115
results revealed that observers were influenced by marked but also by
slight muscle facial movements displayed by the president. For instance,
President Obama displayed several distinct “types” of smiles. The president’s smiles that also activated the muscle causing the lower lip to move
upwards received lower happiness/reassurance and higher anger/threat
ratings than smiles that did not activate this lip muscle movement.
Stewart, Méhu, and Salter (2015) used FACS to examine sex differences
in leaders with regard to facial expression of emotion. Sample results
show that facial expressions of anger in different combinations of facial
muscle movement and intensity are more likely to be recognized precisely when displayed by male rather than female leaders. Furthermore,
Stewart, Bucy, and Méhu (2015) explored different types of smiles of
Republican presidential candidates, revealing the complexity of leader
smiling behavior and the effect these different smiles have on observers’
perceptions. An important finding of the study was the significant effect
subtle alterations regarding muscle movement and intensity of political
leaders’ smiles had on viewers’ perceptions. For example, smiles including a contraction of the muscles surrounding the eyes received higher
happiness/reassurance and low anger/threat ratings than smiles presented
without the eye muscle contractions (ibid.). Other simple changes to the
presidential candidates’ smiles such as tightening or pressing together the
lips, pressing up the lower lip, or downward pull of the lip corners were
found to result in significantly lower happiness/reassurance and to raise
anger/threat ratings.
In another study investigating leaders’ smiling behavior, Tsai et al.
(2016) used FACS to explore cultural differences between smiles that
pulled the corners of the lips back and obliquely upward with some contraction of the muscles surrounding the eyes (calm smiles) and smiles that
included more intense muscle activation including jaw dropping and teeth
showing (excited smiles). A significant finding of the above research is
that American leaders exhibited more smiles of excitement compared to
Tawainese/Chinese leaders. Extending from the above results Tsai et al.
(2016) found that the display of leaders’ excitement smiles was a reflection
of the extent the respective nation appreciated excitement as a positive
state of high stimulation. On the other hand, the more a nation appreciated calmness as a positive state of low stimulation, the more the leader of
that nation displayed calm smiles.
Finally, another study that employed sophisticated facial coding
methods to investigate leadership concepts is by Trichas and Schyns
(2012). The research by Trichas and Schyns (2012) will be examined more
thoroughly in the following section as an extended example to discuss the
integration of detailed facial expression methods in leadership research.
SCHYNS_9781785367274_t.indd 115
10/11/2017 15:19
116 Handbook of methods in leadership research
Extended Example: Trichas and Schyns (2012)
Trichas and Schyns (2012), in a multi-study design, used sophisticated
facial expression coding techniques to link observers’ reactions to leaders’
facial expressions with implicit leadership theories (mental concepts
people have regarding behaviors, characteristics, and attitudes of leaders;
see Dinh, Lord, Gardner, Meuser, Liden, & Hu, 2014; Dinh, Lord, &
Hoffman, 2014; Epitropaki & Martin, 2004; and Epitropaki, Sy, Martin,
Tram-Quon, & Topakas, 2013). Initially, Trichas and Schyns (2012)
evaluated participants’ implicit leadership theories (ILTs). Furthermore, a
business context was activated by framing photos of actors/leaders within
the organization and asking the participants to assess them in terms of
leadership. Specifically, the researchers asked observers to evaluate the
actors/leaders exhibiting several eyebrow movements. These included
eyebrow lowering and pulling together, and eyebrow raising and pulling
together in different intensities. The facial expressions used in study 1 of
Trichas and Schyns (2012) were coded in terms of facial muscle movement
and intensity using FACS.
The findings of study 1 revealed that simple brow movements such as
the lowering and pulling together of the eyebrows resulted in strong but
hostile leadership perceptions. In contrast, the raising and pulling together
of the eyebrows was perceived as a sign of weakness, which resulted in
decreasing overall ratings of leadership perception. In general, these results
suggest that appearing stern might constitute a more solid foundation for
leadership perception than appearing kind. In a broader view, these findings indicate that simple alterations of eyebrow movements can result in
significant differences in leadership perception. This finding is consistent
with previous findings highlighting the importance of subtle differences
in the perceptual process (e.g., Snodgrass, 1992; Surakka & Hietanen,
1998). It is essential for leaders to understand the importance of this, as
being aware of the impact of subtle details in facial expressions can help
to improve accuracy of communication and ultimately shape perception.
It is also vital for leadership scholars to understand that facial expression
coding not only provides such awareness but also enables researchers to
explore these subtle details in their designs. The use of sophisticated facial
expression analysis in Trichas and Schyns (2012) was crucial in study 1 as
the coding procedure allowed the researchers to isolate and use only the
few action units required, thus increasing both flexibility and precision on
stimuli construction.
In their second study, Trichas and Schyns (2012) aimed to explore the
effect of leadership perception considering additional factors such as
communication context, appropriateness of expression, and authenticity.
SCHYNS_9781785367274_t.indd 116
10/11/2017 15:19
Sophisticated facial expression coding ­117
Consequently, several pictures with facial actions/indicators of different
emotions were set up in a three-stage scenario (introduction, negotiation,
shaking hands before leaving) regarding a loan discussion between a bank
manager and a client. The facial expressions again included eyebrows
lowering and pulling together and eyebrows raising and pulling together.
Additionally, smiling facial expressions with different muscle combinations and intensities were used. Specifically, the actor displayed smiles
either with or without activation of the muscles around the eyes, a marker
often associated with authentic smiles in static facial expressions (e.g.,
Gunnery & Ruben, 2016). All the facial expressions used in study 2 of
Trichas and Schyns (2012) were also coded using FACS.
The respondents in study 2 indicated an overall preference for
leaders exhibiting smiles compared to the rest of the facial expressions.
However, the results indicate that these reactions were context dependent. Specifically, during the introduction respondents considered lowintensity smiles without activation of the muscles around the eyes more
appropriate. A probable reason for participants’ preferences towards pictures of non-authentic smiles might be that at the beginning of a meeting
with a client, a leader is expected to simply maintain a positive tone
without exaggeration. In the context of shaking hands before leaving,
the participants preferred high-intensity smiles with activation of the
muscles surrounding the eyes. This may be because after the negotiations,
just before saying goodbye, association between the two parties is at a
higher level and thus more intense, and authentic expressions are deemed
appropriate. These results are important because they show that aspects
of facial expression such as perceived authenticity and intensity of display
may depend on the level of association between a leader and a perceiver
at the time of expression. Therefore, understanding leader facial displays
becomes a matter of analysing both the facial expressions and (social or
situational) context.
Finally, leadership first impressions from the facial expressions in study
2 were compared to observers’ ILTs. Based on Hall and Lord’s (1995)
perspective arguing that people use their ILTs as a reference point when
they need to categorize someone as a leader or non-leader, Trichas and
Schyns (2012) anticipated that if expectations in the form of ILTs match
an individual’s facial expressions, then the perception of that individual
will be perceived as more leader-like than when these expectations did
not match the facial expressions displayed. Indeed, Trichas and Schyns
(2012) found that when ILTs are met by a person’s facial expressions, then
that person is categorized as “leader,” thus implying that ILTs are used in
the perception and evaluation of leaders. To put it more simply, a match
between a leader’s facial expressions and participants’ ILTs led to more
SCHYNS_9781785367274_t.indd 117
10/11/2017 15:19
118 Handbook of methods in leadership research
favorable leadership perceptions than when there was a mismatch. This
finding is in accordance with a previous theory holding that people use
their expectations (ILTs) as a reference point for the evaluation of good
leadership (Hall & Lord, 1995) and other research demonstrating that a
match between an individual’s expectations of a leader (a prototype) with
the leader’s actual behaviors leads to more favorable evaluations (Nye &
Forsyth, 1991).
Overall, Trichas and Schyn’s (2012) research reinforced the notion that
facial expressions have a commanding influence on leadership perception
(Stewart et al., 2009). Specifically, different facial expressions at several
intensity levels (from slight to extreme) were found to impact the perception of leadership. Consequently, the sophisticated facial expression
coding Trichas and Schyns (2012) used in their studies helped increase the
depth of analysis and the precision of findings, adding to the validity and
reliability of research (see Rosenberg, 2005). Finally, the same research
revealed that in order to understand the effect of leaders’ facial expressions
one needs to consider how people both produce and perceive facial expressions. Therefore, Trichas and Schyns (2012) are in line with earlier literature supporting the idea that leadership is, at least to a degree, constructed
by perceivers (Gray & Densten, 2007; Schyns, Felfe, & Blank, 2007).
IMPLICATIONS, CONTRIBUTIONS, AND FUTURE
RECOMMENDATIONS
The leadership research examined in this chapter reveals that facial expressions have a strong influence on leadership perception (e.g., Bucy, 2000;
Masters & Sullivan, 1989). However, our knowledge regarding the actual
impact of leaders’ facial displays is still restricted. The majority of relevant
research does not employ a detailed approach to facial expression analysis.
The present chapter used the few leadership studies that utilized sophisticated facial expression coding as reference points to draw attention to the
contribution of such methods in the area of leadership research.
One of the main arguments in the current chapter is that precise coding
can help to uncover both pronounced and subtle perceptual impacts
between displays, thus contributing to a deeper understanding of leaders’
facial expressions (Krumhuber et al., 2009). The FACS instrument, one of
the most broadly used instruments of detailed facial expression analysis
(Cohn & Ekman, 2008), is the method used in the contemporary leadership studies discussed in the chapter that underline the added value
of sophisticated facial expression coding. The precision of FACS can
facilitate the discrimination of fine perceptual influences between facial
SCHYNS_9781785367274_t.indd 118
10/11/2017 15:19
Sophisticated facial expression coding ­119
expressions, thereby extending our understanding of facial expressions in
the context of leadership (Trichas & Schyns, 2012). The identification of
all observable facial movements with high reliability even in low intensities increases the credibility of research designs (see Rosenberg, 2005) and
enables scholars to explore new angles in the context of leadership.
The results of the research that utilized such sophisticated facial coding
techniques discussed in the current chapter are of great academic value.
To be more specific, a significant finding revealed from the investigation
of leadership studies that used such detailed coding methods is that both
marked and slight alterations in leaders’ facial displays may lead to different perceptions (Stewart, Bucy, & Méhu, 2015; Stewart, Méhu, & Salter,
2015; Stewart & Dowe, 2013; Tsai et al., 2016). This was true even when
facial expressions appeared for fractions of a second (micro-expressions in
Stewart et al., 2009). These findings reinforce the necessity of coding precision in facial expression research designs within the context of leadership.
As is argued in the current chapter, such level of analysis allows scholars
to analyse facial expressions of actual leaders (e.g., accurately identifying
politicians’ micro-expressions and exploring their perceptual impact; see
Stewart et al., 2009), but also enables researchers to create experimental
stimuli by manipulating facial expressions (e.g., using coded facial expressions simulating business scenarios to examine the influences of certain
eyebrow movements and smiles on leadership perceptions; see Trichas &
Schyns, 2012). Finally, sophisticated facial expression coding can also be
used to treat emotional displays as an end product. Specifically, detailed
facial analysis may be used to construct dependent variables in order to
investigate complex leader concepts such as authenticity of expression.
Taking, for example, literature from other disciplines discriminating
between authentic and non-authentic facial expressions reviewed earlier in
this chapter (e.g., Gunnery & Ruben, 2016; Krumhuber & Kappas, 2005),
it would be interesting to see how such knowledge applies in several leadership contexts (e.g., leader exploration of authentic and non-authentic
emotional facial displays during public speaking on the basis of temporal
aspects).
With regard to the actual integration of sophisticated facial expression
coding to leadership research, a practical issue is whether the coding in a
design is done internally (the coding is conducted by one of the authors)
or externally (the coding is conducted by other certified coders unrelated
to the research). Either method provides high levels of credibility and
reliability; Cohn & Ekman, 2008. The current chapter aims to encourage any leadership scholars investigating facial expressions to become
FACS coders themselves in order to hone their awareness of the detail
and variety of facial expressions from anatomy to emotional perception.
SCHYNS_9781785367274_t.indd 119
10/11/2017 15:19
120 Handbook of methods in leadership research
Furthermore, internal coding facilitates the research process as the coder
is also one of the researchers so there is continuous and immediate access
to sophisticated facial action knowledge in important parts of the design
(e.g., instrumentation). Of course, it should be noted that internal coding
could also entail a potential disadvantage: in cases, for example, where
facial expression is a dependent variable, coders’ knowledge of conditions
and/or the hypotheses could be a potential threat to research credibility.
In such cases, authors/coders use FACS knowledge and mindset to contribute to the design but distance themselves from the actual analyses. This
can be done by involving independent FACS coders to perform the actual
analysis in order to ensure both credibility and reliability of the design.
Finally, another important finding revealed from the research reviewed
in the current chapter was that simple muscle movements in a leader’s
face resulted in context-dependent observer reactions (Trichas & Schyns,
2012). Specifically, using several FACS coded leader frames, the above
scholars demonstrated that facial expressions have a strong influence on
the perception of leadership but they also found that the nature of that
influence is a complex situational process. When perceivers observe facial
expressions, they act as “naive scientists”: they take accessible stimulus,
such as facial expression, context of communication, appropriateness,
and authenticity of expression and interpret them in the light of mental
schemas in an attempt to understand their environment (Hassin et al.,
2002; Trichas, 2015). As a result, an essential basis for exploring leadership concepts related to facial expressions is to accurately decode and
consider all available information. Therefore, the focus not only falls on
the expressed emotion per se, or even the aims of the leader/transmitter,
but also on accurately identifying exact facial muscle movement, and how
observers perceive these displays in the specific situation. The complexity
of how people perceive in Trichas and Schyns (2012) implies that understanding leadership becomes an issue of creative problem solving in terms
of realizing what is better under the given circumstances instead of seeking
out behavioral “recipes” to apply (Meindl, 1995).
CONCLUSION
The current chapter presents a set of methods for investigating facial
expression, different from what has been used so far in the area of leadership perception. The studies reviewed in the previous sections integrated
detailed psychological methods of facial expression coding to explore
leadership perception. The findings of this chapter show that even the
most subtle alterations in facial expressions matter in leadership percep-
SCHYNS_9781785367274_t.indd 120
10/11/2017 15:19
Sophisticated facial expression coding ­121
tions, implying that awareness of the sophistication of human facial
expressions can contribute to a different perspective on the leadership perceptual processes from both the transmitters’ and the perceivers’ points of
view. Cohn et al. (1999) argue that obtaining accurate facial expression
measurements is crucial for the credibility of research. Consequently, a
main conclusion of the current chapter is that the detailed coding of facial
expression can significantly contribute to our understanding of leadership perceptions.
REFERENCES
Adams, R., Ambadi, N., Macrae, C., & Kleck, R. (2006). Emotional expressions forecast
approach-avoidance behavior. Motivation and Emotion, 30(2), 179–188.
Ambadar, Z., Cohn, J., & Reed, L. (2009). All smiles are not created equal: Morphology
and timing of smiles perceived as amused, polite, and embarrassed/nervous. Journal of
Nonverbal Behavior, 33(1), 17–34.
Awamleh, R., & Gardner, W.L. (1999). Perceptions of leader charisma and effectiveness: The
effects of vision content, delivery, and organizational performance. Leadership Quarterly,
10(3), 345–373.
Barsade, S.G. (2002). The ripple effect: Emotional contagion and its influence on group
behavior. Administrative Science Quarterly, 47(4), 644–675.
Beatty, B. (2001). The emotions of educational leadership: Breaking the silence. International
Journal of Leadership in Education, 3(4), 331–357.
Bhushan, B. (2015). Study of facial micro-expressions in psychology. In M.K. Mandal &
A. Awasthi (Eds.), Understanding facial expressions in communication: Cross-cultural and
multidisciplinary perspectives (pp. 265–286). New Delhi: Springer.
Bono, J.E., & Ilies, R. (2006). Charisma, positive emotions and mood contagion. The
Leadership Quarterly, 17(4), 317–334.
Bucy, E.P. (2000). Emotional and evaluative consequences of inappropriate leader displays.
Communication Research, 27(2), 194–226.
Bucy, E.P., & Bradley, S.D. (2004). Presidential expressions and viewer emotion: Counter
empathic responses to televised leader displays. Social Science Information, 43(1), 59–94.
Bucy, E.P., & Newhagen, J.E. (1999). The emotional appropriateness heuristic: Processing
televised presidential reactions to the news. Journal of Communication, 49(4), 59–79.
Cañadas, E., Lupiáñez, J., Kawakami, K., Niedenthal, P.M., & Rodríguez-Bailón, R. (2016).
Perceiving emotions: Cueing social categorization processes and attentional control
through facial expressions. Cognition and Emotion, 30(6), 1149–1163.
Channouf, A. (2000). Subliminal exposure to facial expressions of emotion and evaluative
judgments of advertising messages. European Review of Applied Psychology, 50(1), 19–23.
Cherulnik, P.D., Donley, K.A., Wiewel, T.S., & Miller, S.R. (2001). Charisma is contagious:
The effect of leaders’ charisma on observers’ affect. Journal of Applied Social Psychology,
31(10), 2149–2159.
Cohn, J.F., & Ekman, P. (2008). Measuring facial action. In J.A. Harrigan, R. Rosenthal, &
K.R. Scherer, The new handbook of methods in nonverbal behavior research (pp. 9–64). New
York: Oxford University Press.
Cohn, J., Ambadar, Z., & Ekman, P. (2007). Observer-based measurement of facial expression with the Facial Action Coding System. In J.A. Coan, & J.J. Allen (Eds.), The handbook of emotion elicitation and assessment (pp. 203–221). Oxford/New York: Oxford
University Press.
Cohn, J.F., Zlochoher, A.J., Lien, J., & Kanade, T. (1999). Automated face analysis by feature
SCHYNS_9781785367274_t.indd 121
10/11/2017 15:19
122 Handbook of methods in leadership research
point tracking has high concurrent validity with manual FACS coding. Psychophysiology,
36(1), 35–43.
Damen, F., Van Knippenberg, B., & Van Knippenberg, D. (2008). Affective match in leadership: Leader emotional displays, follower positive affect, and follower performance. Journal
of Applied Social Psychology, 38(4), 868–902.
Darwin, C. (1872 [1965]). The expression of the emotions in man and animals. Chicago, IL:
University of Chicago Press.
Dimberg, U., Thunberg, M., & Elmehed, K. (2000). Unconscious facial reactions to emotional facial expressions. Psychological Science, 11(1), 86–89.
Dinh, J., Lord, R., Gardner, W., Meuser, J., Liden, R., & Hu, J. (2014). Leadership theory and
research in the new millennium: Current theoretical trends and changing perspectives. The
Leadership Quarterly, 25(1), 36–62.
Dinh, J.E., Lord, R.G., & Hoffman, E. (2014). Leadership perception and information
processing: Influences of symbolic, connectionist, emotion, and embodied architectures.
In D. Day, The Oxford handbook of leadership and organizations (pp. 29–65). New York:
Oxford University Press.
Dovidio, J.F., Heltman, K., Brown, C.E., Ellyson, S.L., & Keating, C.E. (1988). Power
displays between women and men in discussions of gender-linked tasks: A multichannel
study. Journal of Personality and Social Psychology, 55(4), 580–587.
Ekman, P. (1992). Facial expressions of emotion: New findings, new questions. Psychological
Science, 3(1), 34–38.
Ekman, P. (2001). Telling lies: Clues to deceit in the marketplace, marriage, and politics. New
York: W.W. Norton.
Ekman, P. (2003a). Darwin, deception and facial expression. In P. Ekman, R.J. Davidson,
& F. de Waals (Eds.), Annals of the New York Academy of Sciences: Vol. 1000: Emotions
inside out: 130 years after Darwin’s The Expression of the Emotions in Man and Animals
(pp. 205–221). New York: New York Academy of Sciences.
Ekman, P. (2003b). Emotions Revealed: Recognizing Faces and Feelings to Improve
Communication and Emotional Life, New York: Times Books.
Ekman, P. (2009). Lie catching and microexpressions. In C. Martin (Ed.), The philosophy of
deception (pp. 118–142). New York: Oxford University Press.
Ekman, P., & Rosenberg, E. (1997). What the face reveals: Basic and applied studies of spontaneous expression. New York: Oxford: Oxford University Press.
Ekman, P., Friesen, W.V., & Hager, J.C. (2002). Facial Action Coding System: The
manual. Salt Lake City, UT: Research Nexus Division of Network Information Research
Corporation.
Epitropaki, O., & Martin, R. (2004). Implicit leadership theories in applied settings: Factor
structure, generalizability, and stability over time. Journal of Applied Psychology, 89(2),
293–310.
Epitropaki, O., Sy, T., Martin, R., Tram-Quon, S., & Topakas, A. (2013). Implicit leadership and followership theories “in the wild”: Taking stock of information-processing
approaches to leadership and followership in organizational settings. The Leadership
Quarterly, 24(6), 858–881.
Field, T.M., Woodson, R., Greenberg, R., & Cohen, D. (1982). Discrimination and imitation
of facial expressions by neonates. Science, 218(4568), 179–181.
Frank, M.G., & Ekman, P. (1997). The ability to detect deceit generalizes across different types of high-stake lies. Journal of Personality and Social Psychology, 72(6),
1429–1439.
Franklin, R., & Zebrowitz, L. (2013). Older adults’ trait impressions of faces are sensitive to
subtle resemblance to emotions. Journal of Nonverbal Behavior, 37(3), 139–151.
Fridlund, A.J. (1994). Human facial expression: An evolutionary view. San Diego, CA:
Academic Press.
Gaddis, B., Connelly, S., & Mumford, M.D. (2004). Failure feedback as an affective
event: Influences of leader affect on subordinate attitudes and performance. Leadership
Quarterly, 15(5), 663–686.
SCHYNS_9781785367274_t.indd 122
10/11/2017 15:19
Sophisticated facial expression coding ­123
Glomb, T.M., & Hulin, C.L. (1997). Anger and gender effects in observed supervisor–­
subordinate dyadic interactions. Organizational Behavior and Human Decision Processes,
72(3), 281–307.
Gong, Z.H., & Bucy, E.P. (2016). When style obscures substance: Visual attention to display
appropriateness in the 2012 presidential debates. Communication Monographs. Retrieved
from http://www.tandfonline.com/doi/full/10.1080/03637751.2015.1119868
Gray, J.H., & Densten, I.L. (2007). How leaders woo followers in the romance of leadership.
Applied Psychology: An International Review, 56(4), 558–581.
Gunnery, S.D., & Hall, J.A. (2014). The Duchenne smile and persuasion. Journal of
Nonverbal Behavior, 38(2), 181–194.
Gunnery, S.D., & Ruben, M.A. (2016). Perceptions of Duchenne and non-Duchenne smiles:
A meta-analysis. Cognition and Emotion, 30(3), 1–15.
Hall, R.J., & Lord, R.G. (1995). Multi-level information-processing explanations of followers’ leadership perceptions. Leadership Quarterly, 6(3), 265–287.
Hareli, S., & Hess, U. (2010). What emotional reactions can tell us about the nature of others:
An appraisal perspective on person perception. Cognition & Emotion, 24(1), 128–140.
Hassin, R.R., Bargh, J.A., & Uleman, J.S. (2002). Spontaneous causal inferences. Journal of
Experimental Social Psychology, 38(5), 515–522.
Hendriks, M., & Vingerhoets, A. (2006). Social messages of crying faces: Their influence on
anticipated person perception, emotion and behavioral responses. Cognition & Emotion,
20(6), 878–886.
Hess, U., Blairy, S., & Kleck, R.E. (2000). The influence of facial emotion displays, gender,
and ethnicity on judgments of dominance and affiliation. Journal of Nonverbal Behavior,
24(4), 265–283.
Hochschild, A.R. (1983). The managed heart: Commercialization of human feeling. Berkeley,
CA: University of California Press.
Howard, D.J., & Gengler, C. (2001). Emotional contagion effects on product attitudes.
Journal of Consumer Research, 28(2), 189–201.
Humphrey, R.H., Pollack, J.M., & Hawver, T. (2008). Leading with emotional labour.
Journal of Managerial Psychology, 23(2), 151–168.
Hurley, C.M., Anker, A.E., Frank, M.G., Matsumoto, D., & Hwang, H.C. (2014). Background
factors predicting accuracy and improvement in micro-expression recognition. Motivation
and Emotion, 38(5), 700–714.
Izard, C. (1983). The Maximally Discriminative Facial Movement Coding System (MAX).
Unpublished manuscript. University of Delaware.
Izard, C.E., Dougherty, L.M., & Hembree, E.A. (1983). A system for identifying affect
expressions by holistic judgments. Unpublished manuscript. University of Delaware.
Jenkins, A.M., & Johnson, R.D. (1977). What the information analyst should know about
body language. MIS Quarterly, 1(3), 33–47.
Keating, C. (2003). Messages from face and body: Women, men, and the silent expression
of social status. In D.S. Cobble, B. Hutchison, & A.B. Chaloupka (Eds.), Femininities,
masculinities, and the politics of sexual difference(s) (pp. 65–70). Brunswick, NJ: Rutgers
University Press.
Keating, C.F., Mazur, A., & Segall, M.H. (1977). Facial gestures which influence the perception of status. Sociometry, 40(4), 374–378.
Keating, C.F., Mazur, A., & Segall, M.H. (1981). Culture and the perception of social dominance from facial expression. Journal of Personality and Social Psychology, 40(4), 615–626.
Knutson, B. (1996). Facial expressions of emotion influence interpersonal trait inferences.
Journal of Nonverbal Behavior, 20(3), 165–182.
Kraus, M.W., & Chen, T.D. (2013). A winning smile? Smile intensity, physical dominance,
and fighter performance. Emotion, 13(2), 270–279.
Krumhuber, E., & Kappas, A. (2005). Moving smiles: The role of dynamic components for
the perception of the genuineness of smiles. Journal of Nonverbal Behavior, 29(1), 3–24.
Krumhuber, E., Manstead, A.S., Cosker, D., Marshall, D., & Rosin, P.L. (2007). Facial
dynamics as indicators of trustworthiness and cooperative behavior. Emotion, 7(4), 730–735.
SCHYNS_9781785367274_t.indd 123
10/11/2017 15:19
124 Handbook of methods in leadership research
Krumhuber, E., Manstead, A.S., Cosker, D., Marshall, D., & Rosin, P.L. (2009). Effects
of dynamic attributes of smiles in human and synthetic faces: A simulated job interview
setting. Journal of Nonverbal Behavior, 33(1), 1–15.
Krumhuber, E., Manstead, A., & Kappas, A. (2006). Temporal aspects of facial displays in
person and expression perception: The effects of smile dynamics, head-tilt, and gender.
Journal of Nonverbal Behaviour, 31(1), 39–56.
Lau, S. (1982). The effect of smiling on person perception. The Journal of Social Psychology,
117(1), 63–67.
Lewis, K.M. (2000). When leaders display emotion: How followers respond to negative emotional expression of male and female leaders. Journal of Organizational Behavior, 21(2),
221–234.
Littlewort, G., Whitehill, J., Wu, T., Fasel, I., Frank, M., & Movellan, J. (2011). The
Computer Expression Recognition Toolbox (CERT). Proceedings of the IEEE International
Conference on Automatic Face and Gesture Recognition. Retrieved from mplab.ucsd.edu/
wp-content/uploads/2011-LittlewortEtAl-FG-CERT.pdf
Madera, J.M., & Smith, B.D. (2009). The effects of leader negative emotions on evaluations
of leadership in a crisis situation: The role of anger and sadness. The Leadership Quarterly,
20(2), 103–114.
Marsh, A., & Ambady, N. (2007). The influence of the fear facial expression on prosocial
responding. Cognition and Emotion, 21(2), 225–247.
Marsh, A.A., Ambady, N., & Kleck, R.E. (2005). The effects of fear and anger facial expressions on approach and avoidance-related behaviors. Emotion, 5(1), 119–124.
Masters, R.D., & Sullivan, D.G. (1989). Nonverbal displays and political leadership in
France and the United States. Political Behavior, 11(2), 123–156.
Mazur, A., & Mueller, U. (1996). Facial dominance. In A. Somit & S. Peterson (Eds.),
Research in Biopolitics (pp. 99–111). London: JAI Press.
McArthur, L.Z., & Baron, R.M. (1983). Toward an ecological theory of social perception.
Psychological Review, 90(3), 215–238.
Medvedeff, M.E. (2008). Leader affective displays during a negative work event: Influences
on subordinate appraisals, affect, and coping strategies. Doctoral dissertation. University
of Akron. Retrieved from http://etd.ohiolink.edu/send-pdf.cgi/Medvedeff%20Megan.pdf?​
akron1207753447
Meindl, J.R. (1995). The romance of leadership as a follower-centric theory: A social constructionist approach. Leadership Quarterly, 6(3), 329–341.
Melwani, S., Mueller, J.S., & Overbeck, J.R. (2012). Looking down: The influence of
contempt and compassion on emergent leadership categorizations. Journal of Applied
Psychology, 97(6), 1171–1185.
Monahan, J.L. (1998). I don’t know it but I like you: The influence of nonconscious affect on
person perception. Human Communication Research, 24(4), 480–500.
Montepare, J.M., & Dobish, H. (2003). The contribution of emotion perceptions and
their overgeneralizations to trait impressions. Journal of Nonverbal Behavior, 27(4),
237–254.
Montepare, J.M., & Zebrowitz-McArthur, L. (1998). Impressions of people created by
age-related qualities of their gaits. Journal of Personality and Social Psychology, 55(4),
547–556.
Newcombe, M.J., & Ashkanasy, N.M. (2002). The role of affect and affective congruence in perceptions of leaders: An experimental study. The Leadership Quarterly, 13(5),
601–614.
Nye, J., & Forsyth, D.R. (1991). The effects of prototype biases on leadership appraisals: A
test of leadership categorization theory. Small Group Research, 22(3), 360–379.
Parkinson, B. (2005). Do facial movements express emotions or communicate motives?
Personality and Social Psychology Review, 9(4), 278–311.
Perron, M., Roy-Charland, A., Chamberland, J., Bleach, C., & Pelot, A. (2016). Differences
between traces of negative emotions in smile judgment. Motivation and Emotion, 40(3),
478–488.
SCHYNS_9781785367274_t.indd 124
10/11/2017 15:19
Sophisticated facial expression coding ­125
Rafaeli, A., & Sutton, R.I. (1987). Expression of emotion as part of the work role. Academy
of Management Review, 12(1), 23–37.
Rosenberg, E. (2005). The study of spontaneous facial expressions in psychology. In
P. Ekman & E. Rosenberg (Eds.), What the face reveals: Basic and applied studies of spontaneous expression using the facial action coding system (2nd ed., pp. 3–17). New York:
Oxford University Press.
Sayette, M.A., Cohn, J.F., Wertz, J.M., Perrott, M.A., & Parrot, D.J. (2001). A psychometric evaluation of the Facial Action Coding System for assessing spontaneous expression.
Journal of Nonverbal Behavior, 25(3), 167–185.
Schneid, E.D., Carlston, D.E., & Skowronski, J.J. (2015). Spontaneous evaluative inferences
and their relationship to spontaneous trait inferences. Journal of Personality and Social
Psychology, 108(5), 681–696.
Schyns, B., & Mohr, G. (2004). Nonverbal elements of leadership behaviour. German Journal
of Human Resource Research, 18(3), 289–305.
Schyns, B., Felfe, J., & Blank, H. (2007). Is charisma hyper-romanticism? Empirical evidence
from new data and a meta-analysis. Applied Psychology: An International Review, 56(4),
505–527.
Shariff, A.F., & Tracy, J.L. (2011). What are emotion expressions for? Current Directions in
Psychological Science, 20(6), 395–399.
Shea, C.M., & Howell, J.M. (1999). Charismatic leadership and task feedback: A laboratory
study of their effects on self-efficacy and task performance. Leadership Quarterly, 10(3),
375–396.
Snodgrass, J. (1992). Judgment of feeling states from facial behavior: A bottom-up approach.
Unpublished doctoral dissertation. University of British Columbia.
Stewart, P. (2010). Presidential laugh lines: Candidate display behavior and audience laughter in the 2008. Politics and the Life Sciences, 29(2), 55–72.
Stewart, P., & Dowe, P. (2013). Interpreting president Barack Obama’s facial displays of
emotion: Revisiting the Dartmouth Group. Political Psychology, 34(3), 369–385.
Stewart, P.A., Bucy, E.P., & Méhu, M. (2015). Strengthening bonds and connecting with
followers: A biobehavioral inventory of political smiles. Politics and the Life Sciences,
34(1), 73–92.
Stewart, P.A., Méhu, M., & Salter, F.K. (2015). Sex and leadership: Interpreting competitive
and affiliative facial displays based on workplace status. International Public Management
Journal, 18(2), 190–208.
Stewart, P., Waller, B., & Schubert, J. (2009). Presidential speechmaking style: Emotional
response to micro-expressions of facial affect. Motivation and Emotion, 33(2), 125–135.
Sullivan, D.G., & Masters, R.D. (1988). “Happy warriors”: Leaders’ facial displays, viewers’
emotions, and political support. American Journal of Political Science, 32(2), 345–368.
Surakka, V., & Hietanen, J.K. (1998). Facial and emotional reactions to Duchenne and nonDuchenne smiles. International Journal of Psychophysiology, 29(1), 23–33.
Sutton, R.I., & Rafaeli, A. (1988). Untangling the relationship between displayed emotions and organizational sales: The case of convenience stores. Academy of Management
Journal, 31(3), 461–487.
Sy, T., Cȏté, S., & Saavedra, R. (2005). The contagious leader: Impact of the leader’s mood
on the mood of group members, group affective tone, and group processes. Journal of
Applied Psychology, 90(2), 295–305.
Tiedens, L.Z. (2001). Anger and advancement versus sadness and subjugation: The effect of
negative emotion expressions on social status conferral. Journal of Personality and Social
Psychology, 80(1), 86–94.
Trichas, S. (2015). New methods of exploring facial expressions in the context of leadership
perception: Implications for educational leaders. In K. Beycioglu & P. Pashiardis (Eds.),
Multidimensional perspectives on principal leadership effectiveness (pp. 205–225). New
York: IGI Global.
Trichas, S., & Schyns, B. (2012). The face of leadership: Perceiving leaders from facial expression. The Leadership Quarterly, 23(3), 545–566.
SCHYNS_9781785367274_t.indd 125
10/11/2017 15:19
126 Handbook of methods in leadership research
Tronick, E.Z. (1989). Emotions and emotional communication in infants. American
Psychologist, 44(2), 112–119.
Tsai, J., Ang, J., Blevins, E., Goernandt, J., Fung, H., Jiang, D., . . .Haddouk, L. (2016).
Leaders’ smiles reflect cultural differences in ideal affect. Emotion, 16(2), 183–195.
Uleman, J.S., Saribay, S.A., & Gonzalez, C. (2008). Spontaneous inferences, implicit impressions, and implicit theories. Annual Review of Psychology, 59(1), 329–360.
Valstar, M.F., Méhu, M., Jiang, B., Pantic, M., & Scherer, K. (2012). Meta-analysis of the
first facial expression recognition challenge. IEEE Transactions on Systems, Man, and
Cybernetics – Part B, 42(4), 966–979.
Van Kleef, G., Homan, A., Beersma, B., Van Knippenberg, D., Van Knippenberg, B., &
Damen, F. (2009). Searing sentiment or cold calculation? The effects of leader emotional
displays on team performance depend on follower epistemic motivation. Academy of
Management Journal, 52(3), 562–580.
Wang, M., Xia, J., & Yang, F. (2015). Flexibility of spontaneous trait inferences: The interactive effects of mood and gender stereotypes. Social Cognition, 33(4), 345–358.
Wild, B., Erb, M., & Bartels, M. (2001). Are emotions contagious? Evoked emotions while
viewing emotionally expressive faces: Quality, quantity, time course and gender differences. Psychiatry Research, 102(2), 109–124.
Winkielman, P., & Berridge, K. (2003). Irrational wanting and subrational liking: How
rudimentary motivational and affective processes shape preferences and choices. Political
Psychology, 24(4), 657–680.
Winkielman, P., & Berridge, K. (2004). Unconscious emotion. Current Directions in
Psychological Science, 13(3), 120–123.
Winkielman, P., Berridge, K., & Wilbarger, J.L. (2005). Unconscious affective reactions to
masked happy versus angry faces influence consumption behavior and judgments of value.
Personality and Social Psychology Bulletin, 31(1), 121–135.
Zebrowitz, L.A., & Montepare, J.M. (2005). Appearance does matter. Science, 308(5728),
1565–1566.
Zebrowitz, L.A., & Montepare, J.M. (2008). Social psychological face perception: Why
appearance matters. Social and Personality Psychology Compass, 2(3), 1497–1517.
SCHYNS_9781785367274_t.indd 126
10/11/2017 15:19
6.
Behavioral genetics and leadership
research
Wen-Dong Li, Remus Ilies and Wei Wang
The use of behavioral genetics in scientific inquiries may date back to the
study conducted by Sir Francis Galton, Darwin’s half cousin. Galton
(1869) published his study in the book titled Hereditary Genius: An Inquiry
into its Laws and Consequences. Galton examined so-called “eminent
men” in England, ranging from judges, musicians, to wrestlers, and
found that genetic factors played a prominent role in shaping the differences in their genius. Behavioral genetics approaches are broadly defined
as including twin and adoption studies and molecular genetics research
that utilize DNA information (Plomin, DeFries, Knopic, & Neiderhiser,
2013). Modern behavioral genetics approaches have been widely adopted
in an array of disciplines in social sciences, including psychology (e.g.,
Bouchard, Lykken, McGue, Segal, & Tellegen, 1990; Plomin, Owen,
& McGuffin, 1994), education (e.g., Eaves et al., 1997), sociology (e.g.,
Freese, 2008), economics (e.g., Miller, Mulvey, & Martin, 1995), finance
(e.g., Cesarini, Johannesson, Lichtenstein, Sandewall, & Wallace, 2008)
and even political sciences (e.g., Fowler, Baker, & Dawes, 2008).
Organizational researchers have also capitalized on the behavioral
genetics approaches in their investigation of the “nature versus nurture”
or “nature and nurture” issue. This issue is not a trivial one. Kurt Lewin
(1935) proposed the famous equation that human behavior is jointly
affected by both the person and the environment, that is, B 5 f (P, E) (B,
P, and E denote behavior, person, and environment respectively). Genetic
influences are typically employed to disentangle influences from the
person, or nature, from situational influences that are portrayed to indicate influences from the environment, or nurture. Organizational researchers have used the behavioral genetics approaches in their examinations of
job satisfaction (e.g., Arvey, Bouchard, Segal, & Abraham, 1989; Ilies &
Judge, 2003), entrepreneurship (e.g., Nicolaou, Shane, Cherkas, Hunkin,
& Spector, 2008; Zhang, Ilies, & Arvey, 2009), work-related personality
traits (e.g., Judge, Ilies, & Zhang, 2012; Li, 2011), and work characteristics
(Li, Zhang, Song, & Arvey, 2016), to name a few (for reviews, see Arvey,
Li, & Wang, 2016; Ilies, Arvey, & Bouchard, 2006).
The behavioral genetics approaches have also proven useful in
127
SCHYNS_9781785367274_t.indd 127
10/11/2017 15:19
128 Handbook of methods in leadership research
a­ dvancing leadership research (see also Chapter 7 in this volume). In the
field of leadership, researchers have long recognized the fundamental role
of individual difference variables (e.g., intelligence and personality traits)
in shaping leadership emergence and leadership effectiveness (e.g., DeRue,
Nahrgang, Wellman, & Humphrey, 2011; Judge, Bono, Ilies, & Gerhardt,
2002; Stogdill, 1948). Researchers have adopted twin studies and molecular genetics approaches in examining critical questions in this field (e.g.,
Arvey, Rotundo, Johnson, Zhang, & McGue, 2006; Arvey, Zhang,
Avolio, & Krueger, 2007; Li, Arvey, Zhang, & Song, 2012; Li et al., 2015).
Such questions include, but are not limited to, to what extent can genetic
factors influence individual differences in leadership, to what extent can
relationships between different leadership constructs be accounted for by
the same genetic factors, and are specific genes involved in shaping leadership, and how? Perhaps for these reasons, Bass and Bass (2008) state that
“the genetic factor needs to be taken into account in any complete examination of leadership” (p. 1203).
With the rapid advancement of leadership research and the behavioral
genetics field, we firmly believe that behavioral genetics approaches offer
leadership research useful theoretical and methodological perspectives to
address crucial questions about leadership. We therefore write this chapter
to partially serve this purpose. In the following sections, we first introduce
the classic twin studies in behavioral genetics research. We detail univariate and bivariate biometric genetic analyses. We then discuss the use of a
molecular genetics approach, which utilizes people’s specific DNA information. We hope this chapter can spur further research interest in using
behavioral genetics approaches to advance our knowledge on leadership.
CLASSIC TWIN STUDIES
Univariate Biometric Models
Early twin studies, especially those conducted by researchers at the University
of Minnesota, used a very special type of design involving identical (i.e.,
monozygotic) twins who are reared apart (e.g., Arvey et al., 1989; Bouchard
et al., 1990). A basic premise of this approach is that because identical twins
share 100 percent of their genes, when they are raised in different environments their similarities should be attributable only to their similarities in
their genes, not to their environments. This approach, though well known,
is not without limitations. One limitation is that, although identical twins
may be reared by different families, they may still experience similar effective
environments because of the similarities in their abilities, personality traits,
SCHYNS_9781785367274_t.indd 128
10/11/2017 15:19
Behavioral genetics and leadership research ­
129
values, and interests. This is neatly explained by a form of ­gene–­environment
interplay: gene–environment correlation (Plomin, DeFries, & Loehlin, 1977;
Scarr & McCartney, 1983). On the one hand, identical twins may be treated
similarly by others (e.g., family members, friends, and teachers) due to their
similar personality traits, even though they are raised in different families.
This is termed as passive gene–environment correlation, because in such
situations individuals are painted as passive recipients of environmental
influences. On the other hand, identical twins may also proactively seek out
congruent environments (e.g., in making friends with others, looking for jobs
and organizations to work in) with their individual characteristics. These
associations are called active gene–environment correlations, because they
underscore the agentic role of the person in selecting and building up his or
her own environmental niche.
Of late, more sophisticated methods have been developed using multigroup structural equation modeling (SEM) (Neale & Cardon, 1992;
Plomin et al., 2013). Typically, two types of twins are used: identical
twins and fraternal (i.e., dizygotic) twins. Identical twins share all of their
genetic endowments, and fraternal twins, on average, share 50 percent of
the genes that make them similar. This approach does not pose restrictions
on whether twins are reared together or apart, because it helps to distinguish two types of environmental factors: shared environmental factors
and unshared environmental factors. Shared environmental factors (C1
and C2 in Figure 6.1), also referred to as common environmental factors,
represent those same environments that make twins similar (e.g., the same
family and socialization experiences). Unshared environmental factors (E1
and E2), on the other hand, represent those environments that are unique
λ
A1
C1
a
c
Leadershiptwin1
δ
E1
e
A2
C2
a
c
E2
e
Leadershiptwin2
Note: A 5 additive genetic factors; C 5 shared environmental factors; E 5 unique
environmental factors; l 5 1 for identical twins and 0.5 for fraternal twins; d 5 1 for both
types of twins.
Figure 6.1
Univariate biometric analyses for leadership
SCHYNS_9781785367274_t.indd 129
10/11/2017 15:19
130 Handbook of methods in leadership research
to each twin, which make them different from each other (and potential
measurement error). Figure 6.1 presents a univariate biometric model. In
addition to the two types of environmental factors, this type of analysis
also models influences from additive genetic factors (A1 and A2).
This method has been widely used in the recent behavioral genetics literature (Arvey et al., 2016; Arvey, Wang, Song, Li, & Day, 2014; Ilies et
al., 2006; Plomin et al., 2013). With this approach, a heuristic way to detect
whether genetic factors may play a role in affecting a leadership variable,
for instance, is that identical twins are more similar than fraternal twins.
In practice, the similarity is indicated by co-twin correlation. Specifically,
researchers calculate the correlations for one variable between the two
co-twins within one twin pair separately for the identical twin group and
the fraternal twin group. If the co-twin correlation is greater for identical
twins than for fraternal twins, then it suggests likely genetic influences.
Note that often, additive genetic factors are presumed to cause genetic
influences, meaning that the influences of genetic factors are proportionate to the number of a specific genetic variant. However, if the co-twin
correlation for identical twins is more than twice in magnitude than
the correlation for fraternal twins, this often suggests the possibility for
dominant genetic factors. Dominant genetic influences arise from the situation in which the presence of even one genetic variant (e.g., one genetic
allele) would have substantial influences regardless of the numbers of this
genetic variant. In organizational genetics research, the majority of the
prior research has observed additive genetic influences (Arvey et al., 2016)
and it seems rare to observe dominant genetic influences (see Zyphur,
Narayanan, Arvey, & Alexander, 2009, for an exception).
In such analyses, an observed variable, L (e.g., leadership role occupancy), is often modeled as the following algebraically:
L 5 u + a*A + c*C + e*E
(6.1)
where A, C, and E, as discussed above, are standardized latent genetic and
environmental factors (means and variance specified at 0 and 1) respectively; and a, c, and e are corresponding coefficients for them; u represents
the intercept.
Therefore, in SEM, the variance of a leadership variable can be modeled
as:
Variance 5 a2 +c2 +e2(6.2)
It is important to note that as displayed in Figure 6.1, the correlation
between the two genetic variables A1 and A2 (l) is specified as 1 for the
SCHYNS_9781785367274_t.indd 130
10/11/2017 15:19
Behavioral genetics and leadership research ­
131
identical twin group and 0.5 for the fraternal twin group, because identical
and fraternal twins share 100 percent and 50 percent of their genes respectively. In addition, the correlation between the two shared environmental
factors C1 and C2 (d) is specified as 1, because they are the same environmental factors. The correlation between the two unshared environmental
factors, E1 and E2, is thus set as zero, because they are distinctive environmental factors by definition.
Based on the above specifications, the covariance between a leadership
variable for twin 1 and the same leadership variable for twin 2 is calculated
as the following for identical twins:
MZcovariance 5 a2 +c2(6.3)
For fraternal twins, the covariance is modeled as the following:
DZcovariance 5 0.5*a2 +c2(6.4)
The extent of genetic influence, which is also termed heritability, can be
calculated (5a2/[a2 +c2 +e2]), which indicates the amount of total variance in one variable that can be accounted for by individual differences
in genetic factors. Environmental influences can also be determined in
a similar manner. Influences from shared environmental factors are c2/
[a2 +c2 +e2] and influences from unshared environmental factors are e2/[a2
+c2 +e2]. Two examples are the unique variance analyses conducted in the
studies by Arvey and colleagues (Arvey et al., 2006, 2007).
In summary, in univariate biometric analyses, researchers typically
capitalize on information from identical and fraternal twins. Specifically,
researchers conduct multi-group (i.e., identical twin group and fraternal
twin group) SEM and compare the similarities between identical and fraternal twins as detailed above. Then the influences from genetic factors as
well as influences from environmental factors can be estimated. Typically,
such estimates of genetic and environmental factors are used to address
questions such as, to what extent does nature contribute to leadership and
what is the role of nurture?
It is important to point out that the classic behavioral genetics approach
using the two types of twins essentially capitalizes on a quasi-natural
experiment (Plomin et al., 1994). For example, researchers utilize information from two types of twins with different genetic similarities. The
two different types of twins experience similar environmental influences and exhibit differentiated similarities on individual characteristics
and also work outcomes (e.g., leadership). This advantage distinguishes
the behavioral genetics approaches from other approaches and also
SCHYNS_9781785367274_t.indd 131
10/11/2017 15:19
132 Handbook of methods in leadership research
­ rovides ­organizational researchers with unique advantages in investigatp
ing ­important questions related to leadership.
Like all research approaches, behavioral genetics approaches are not
without limitations. One limitation of the classic twin studies pertains
to the equal environments assumption, meaning that, under behavioral
genetics approaches, identical twins are assumed to be exposed to roughly
the same amount of shared environment as fraternal twins. Put differently, the shared environments experienced by identical twins are not
more similar than those encountered by fraternal twins. Researchers have
argued, however, that identical twins share more genes than fraternal
twins, and thus they may be treated more similarly than fraternal twins,
which may pose a challenge to the classic behavioral genetics approach of
twin studies (Gottesman & Shields, 1972; Kendler, Neale, Kessler, Heath,
& Eaves, 1994). If this argument holds, it may lead to inflated estimates
of genetic influences. This critique prompted behavioral geneticists to
conduct research to look deeper into this issue more directly. Researchers
found that indeed identical twins may experience more similar environments than fraternal twins (Bouchard & Propping, 1993; Scarr & CarterSaltzman, 1979). Advocates of behavioral genetics approaches fought
back and argued that, even if this was true, it may not be directly pertinent
to the constructs we study and thus may have little substantial influence
on the magnitudes of genetic influences (Gottesman & Shields, 1972). The
counter-argument makes sense and indeed it is primarily what previous
research has found (Plomin et al., 2013).
Genetic influences, or heritability estimates, are often misinterpreted.
Thus we would like to further clarify the interpretation of genetic influences. Most important, significant genetic influences on one variable
should not be explained as indicating we cannot change this variable
(Johnson, Turkheimer, Gottesman, & Bouchard, 2009). Genetic influences reflect the amount of variance in one variable that can be accounted
for by individual differences in genetic make-up. Genetic influences thus
pertain to inter-individual differences, while change is related to intra-­
individual differences. It is also important to note that, although most
individual difference variables and many work-related variables are
found to be substantially affected by genetic factors (Bouchard, 2004;
Turkheimer, 2000), it is still important to study how much variance in one
work-related variable can be accounted for by genetic variation – that is,
heritability. As contended by Johnson et al. (2009), “heritability studies do
continue to have some importance in areas of the social sciences in which
genetic influences have not been acknowledged” (p. 218).
A final point we would like to make regarding univariate biometric
analyses is that the influences of shared environmental factors, C, have
SCHYNS_9781785367274_t.indd 132
10/11/2017 15:19
Behavioral genetics and leadership research ­
133
often been found to be negligible, especially for studies using adult twins
(Arvey et al., 2016; Turkheimer, 2000). At first glance it seems strange,
given the importance of family environments (an often-assumed shared
environmental factor) on child development. There are several explanations offered by the previous research (Hoffman, 1991; Loehlin, 2007;
Plomin et al., 2013). First, children in the same family do not necessarily
experience environments in the same manner. For instance, one family
may have 500 books for their children, but one child may read 200 of
them and the other may read 100, depending on their reading abilities and
interests. In this vein, family influences are captured by unshared environmental factors, the E factor. Second, the influences of family environments
may gradually become less important as individuals grow up and even
insubstantial in adulthood because over time individuals have more and
more control over their environments. These are important reasons for
most organizational behavioral genetics research to fix influences from
shared environmental factors to zero in their analyses.
Bivariate Biometric Models
As revealed in research in psychology (Kendler & Baker, 2007), organizational behavior (Li, Zhang et al., 2016), and particularly in leadership
(Arvey et al., 2007), individuals are not randomly assigned to various
environments; often, the person also shapes his or her experiences, including work experiences and leadership experiences. For example, Arvey et
al. (2007) reported that approximately 30 percent of the variance in one’s
developmental experiences (i.e., work experience and family experience)
was influenced by genetic differences. Similar findings were reported in
a study on employees’ work experiences (e.g., job demands, job control,
and job complexity; Li, Zhang et al., 2016). A crucial reason for such
substantial genetic influences on individuals’ experiences lies in human
agency: the person can select or craft his or her own environments according to his or her abilities, interests, personality traits, and values (Johnson
et al., 2009). Of course, this notion is not entirely new to organizational
researchers. Schneider’s (Schneider, 1987; Schneider, Goldstein, & Smith,
1995) attraction-selection-attrition model suggests that people can make an
organization what it is through their own actions. The broader literature on
person–organization fit also suggests that people can select their occupations and jobs according to their individual characteristics, such as personality traits and abilities (Edwards, 2008; Kristof-Brown & Guay, 2010). The
burgeoning literature on proactivity has also underscored the agentic role
of the person in shaping work environments (Bindl & Parker, 2010; Frese
& Fay, 2001; Grant & Ashford, 2008; Li, Fay, Frese, Harms, & Gao, 2014).
SCHYNS_9781785367274_t.indd 133
10/11/2017 15:19
134 Handbook of methods in leadership research
Such substantial genetic influences on work experiences are not trivial,
considering that typically those work experience variables have been considered as purely environmental factors, meaning that these factors are
not related to people’s individual characteristics (Oldham & Hackman,
2010). By extension, the traditional view assumes that the relationships
between work experiences and work outcomes may have also been primarily attributed to environmental influences, and thus the person plays little
role here. One such area in organizational research is work design. Classic
work design theories and research have portrayed employee work characteristics as mainly influenced by organizations and managers, not the
person (Grant, Fried, & Juillerat, 2010; Oldham & Hackman, 2010; Parker,
Andrei, & Li, 2014). Thus a conventional wisdom in this area is that the
relationships between work characteristics and work outcomes are mostly
environmentally influenced. However, a recent study (Li, Zhang et al., 2016)
demonstrated that genetic factors also play indispensible roles (e.g., through
core self-evaluations) in explaining the relationships between work characteristics (e.g., job demands, job control, and job complexity) and well-being.
Thus, the most important contribution that the behavioral genetics
approaches can make to leadership research is perhaps to shed light on
the causal explanation of important relationships studied in leadership
research. That is, to identify to what extent such relationships can be
accounted for by genetic factors through selection or environmental influences. As Johnson et al. (2009) point out, an important advantage that
twin studies can bring to our scientific inquiries is that that they are able
to “distinguish selection from environmental causation” (p. 218). For
example, Arvey et al. (2007) reported that the majority of the relationship between family experience (a putative environmental variable) and
leadership role occupancy (i.e., the extent to which people hold leadership
positions) was mainly explained by genetic influences. This was a crucial
reason why work experience was related to leadership role occupancy as
well. Twin studies can also be utilized to address other leadership-related
issues (e.g., leadership experiences and leader development) in a similar
way through disentangling the distinctive roles of selection and environmental causation that underlie such relationships.
One may argue, why is it so important to use genetic influences to reflect
influences from the person through selection? Can’t we just use individual
characteristics such as intelligence (Judge, Colbert, & Ilies, 2004), personality traits (Judge et al., 2002), and physical characteristics (Judge &
Cable, 2004)? The answer is that (1) we cannot investigate influences from
those variables simultaneously in one study, and (2) given that virtually
all individual difference variables are under genetic influences, genetic
influences thus reflect influences from all person-related variables, that is,
SCHYNS_9781785367274_t.indd 134
10/11/2017 15:19
Behavioral genetics and leadership research ­
135
λ
A1
E1
a21
A2
e21
a11
λ
E2
a22
e22
e11
Predictortwin1
A3
E3
A4
E4
a21
e21
a11
a22
e22
e11
Leadershiptwin1
Predictortwin2
Leadershiptwin2
Note: A 5 additive genetic factors; E 5 unique environmental factors; effects of shared
environmental factors (C) were not modeled because the effects are typically not significant,
which is also a consistent finding in previous research; l 5 1 for identical twins and 0.5 for
fraternal twins.
Figure 6.2
ivariate biometric analyses for leadership with a predictor
B
based on Cholesky decomposition
influence from the person as a whole (Arvey et al., 2016; Johnson et al.,
2009; Li, Stanek, Zhang, Ones, & McGue, 2016).
Bivariate biometric models are an extension of univariate models. One
of the most widely used approaches is the one based on Cholesky decomposition (Plomin et al., 2013). Figure 6.2 represents a bivariate biometric
model based on Cholesky decomposition for the relationship between a
predictor (e.g., a personality trait or a measured environmental variable)
and leadership.
With respect to model specification, the cross-twin relationships (e.g.,
predictor_twin1 and predictor_twin2) are modeled similarly as in univariate analyses. The major difference between bivariate and univariate biometric models lies in the within-twin relationships (e.g., between
predictor_twin1 and leadership_twin1). Chiefly, the rationale of this
approach is to decompose the observed relationship between a predictor and leadership into two components: one related to the same genetic
factors (A1) and the other to the same environmental factors (E1).
Statistically, the genetic component of one relationship is the product of
the two coefficients presenting the influences of the same genetic factors
on the two observed constructs (5a11* a21). Likewise, the environmental
component of the observed relationship between a predictor and a leadership variable can also be calculated (5c11* c21). Thus, researchers could
further compute to what extent the observed relationship is accounted for
by genetic influences through selection (5|a11* a21|/[|a11* a21|+ |c11* c21|]) and
SCHYNS_9781785367274_t.indd 135
10/11/2017 15:19
136 Handbook of methods in leadership research
to what extent is due to environmental causation (5|c11* c21|/[|a11* a21|+
|c11* c21|]). This approach has been used previously in psychological (e.g.,
Plomin & Spinath, 2002) and organizational research (Li, Zhang et al.,
2016; Shane, Nicolaou, Cherkas, & Spector, 2010).
Biometric Growth Curve Models
A natural extension of bivariate biometric models is to integrate them
with growth curve models (see also Chapter 13 in this volume). Given the
rapid developments of change-related issues in organizational research
(e.g., Li, Song, & Arvey, 2011; Ployhart & Vandenberg, 2010; Preacher,
Briggs, Wichman, & MacCallum, 2008), it seems intriguing to investigate
whether change in work-related variables (e.g., leader development) can
be modulated by genetic variables. Coupled with previous research on
leadership development (e.g., Day, Fleenor, Atwater, Sturm, & McKee,
2014; Day, Harrison, & Halpin, 2009; DeRue, Nahrgang, Hollenbeck, &
Workman, 2012; Dragoni, Tesluk, Russell, & Oh, 2009), it seems informative to probe whether genes can play a role in leadership experiences and
leadership development over time. In psychology, researchers utilized this
methodology and reported significant genetic influences on change in academic achievement (Johnson, McGue, & Iacono, 2006) and personality
(Hopwood et al., 2011; McGue, Bacon, & Lykken, 1993).
Given that relatively little empirical research in psychology has adopted
this new approach and no organizational research, to our knowledge, has
so far applied this approach, we only briefly discuss the model specifications here. Figure 6.3 presents such an example with a leadership variable
captured three times.
This model is composed of two major parts. One part is related to the
latent growth curve model (on the bottom) in which a leadership variable
is modeled as two latent change variables for each twin, i.e., an intercept
and a slope. The intercept represents the starting point of the leadership
variable and the slope is often used to indicate change in the leadership
variable (given that in this example, because leadership is only measured
three times, non-linear change cannot be modeled). The other part is a
bivariate biometric model (on the top), predicting the latent change variables, intercept and slope. Simply put, a biometric growth curve model is a
­combination of a biometric model and a growth curve model.
Examples of Biometric Models
One good example of univariate biometric models applied in the area of
leadership was the univariate model in the study by Arvey et al. (2006).
SCHYNS_9781785367274_t.indd 136
10/11/2017 15:19
Behavioral genetics and leadership research ­
137
λ
δ
AL
CL
EL
λ
δ
AS
CS
Level1
1
1
1
LDtwin1.time1
0
LDtwin1.time2
ES
AL
CL
EL
Slope1
Level2
γ
1
1
LDtwin1.time3
LDtwin2.time1
1
AS
CS
ES
Slope2
1
0
LDtwin2.time2
γ
1
LDtwin2.time3
Note: LD 5 leadership; A 5 additive genetic factors; C 5 shared environmental factors;
E 5 unique environmental factors; l 5 1 for identical twins and 0.5 for fraternal twins; d 5
1 for both types of twins; g is an estimated parameter capturing the effects of time; Level1
and Level2 are initial status of leadership for twin1 and twin2 respectively, Slope1 and Slope2
are change in leadership for twin1 and twin2 respectively.
Figure 6.3
Biometric growth curve model for a leadership variable
Their analyses were based on 119 pairs of identical twins and 94 pairs
of fraternal twins collected through the Minnesota Twin Registry. The
authors employed univariate biometric models to estimate genetic influences on leadership role occupancy (pp. 9–10). They found that there was
no significant difference between a model with all the three factors, A, C,
and E (i.e., the ACE model) and another model with only A and E factors
(i.e., the AE model). In fact, the influence of C factors was not significant
in this case. Thus, the AE model was selected as the best-fitting model and
genetic factors were found to account for 30 percent of the variance in
leadership role occupancy.
One example of bivariate biometric models is from the study conducted
by Arvey et al. (2007). They reported the results of a multivariate biometric analysis (p. 702) and here we highlight the results for work experience and leadership role occupancy. Again, the results showed that the
influences of C were not significant and thus only A and E factors were
modeled. As shown in Figure 2 on page 702 of Arvey et al. (2007), the
same genetic factor, Aw, significantly influenced both work experience
(coefficient 5 0.18) and leadership (coefficient 5 0.24). Likewise, the same
environmental factor, Aw, also had significant influences on the two variables (coefficients 5 0.30 and 0.31 respectively). Thus, genetic influences
accounted for 31.7 percent of the relationship between work experience
SCHYNS_9781785367274_t.indd 137
10/11/2017 15:19
138 Handbook of methods in leadership research
and leadership role occupancy (50.18*0.24/(0.18*0.24+0.30*0.31)), and
accordingly, environmental factors explaining the remaining 68.3 percent
of the relationship (50.30*0.31/(0.18*0.24+0.30*0.31)).
MOLECULAR GENETICS RESEARCH
Twin studies enable researchers to investigate the relative contributions
of the person (through genetic influences) and the environment to the
variance in one variable or to the relationships/covariance between two
or among multiple variables. However, it cannot provide information
regarding which specific DNA markers are responsible for the significant
genetic influences (Plomin et al., 2013). Given the prevalence of selection
manifested through significant genetic influences on most individual difference variables and on many work-related variables, it is intriguing to
find out which gene or set of genes play a role in explaining such genetic
influences. With the rapid development of DNA sequencing technology, such information on specific genes is becoming increasingly available and such research can satisfy our scientific search for knowledge.
Moreover, if such genes can be identified, the next step is to examine
possible multiple mechanisms related to brain functions, hormones, physical characteristics, and psychological characteristics in the relationship
between those genes and work-related outcomes. Obtaining knowledge on
the relationship between specific genes and leadership variables, and mediating mechanisms, as well as moderating conditions, sheds light on the
biological foundation of important phenomena in organizational research
(Arvey et al., 2016; Cropanzano & Becker, 2013; Heaphy & Dutton, 2008;
Senior, Lee, & Butler, 2011; Waldman, Balthazard, & Peterson, 2011).
So far, organizational research has lagged behind the other social sciences in molecular genetic research. Much research has been conducted,
for example, in psychology (see Turkheimer, Pettersson, & Horn, 2014,
for a recent review) and in political science (Fowler & Dawes, 2008, 2013).
Generally, researchers have adopted the candidate gene approach or
conducted genome-wide association studies (GWAS). The candidate gene
approach is based on previous theories and findings on specific genetic
markers (Plomin et al., 2013). The rationale is that we base our examination of specific genes on theories and previous exploratory research linking
such genes to individual characteristics (e.g., personality traits) and
behavioral outcomes. Then we derive a priori hypotheses and conduct our
empirical investigation. On the other hand, GWAS is more data driven
and its purpose is to test all possible genetic markers that may be related
to a variable of interest without any theoretical or empirical grounds
SCHYNS_9781785367274_t.indd 138
10/11/2017 15:19
Behavioral genetics and leadership research ­
139
(ibid.). Thus, it seems that the candidate gene approach fits better with the
theory-building and theory-testing tradition in organizational research,
and leadership research is not an exception. However, due to the complexity of the inquiry into specific genetic markers and its multidisciplinary
nature, a middle-ground approach that integrates theory building/testing
and empirical exploration is needed (Li et al., 2015).
Candidate Gene Approach
The candidate gene approach was the very first approach adopted in
molecular genetic research. Two seminal papers were published in 1996,
both in Nature (Cloninger, Adolfsson, & Svrakic, 1996; Ebstein et al.,
1996). They reported that one genetic marker on the dopamine receptor
gene D4 was significantly related to personality traits such as sensation
seeking and novelty seeking. The basic rationale of the candidate gene
approach is to correlate genetic variation on one specific genetic marker
with an observed variable of interest (e.g., leadership). The analysis often
used is regression, treating genetic variation as either continuous or categorical variables. Although researchers cautioned that the candidate gene
approach may produce findings that cannot be replicated in other samples
(Ebstein, Israel, Chew, Zhong, & Knafo, 2010), meta-analyses have shown
that for many candidate genes, their influences were indeed significant
across various samples, though sometimes small in magnitude (e.g., Li,
Sham, Owen, & He, 2006; Munafò, Yalcin, Willis-Owen, & Flint, 2008).
Research employing the candidate gene approach often reports small
effect sizes. This is typical and also understandable, given that most
outcome variables we study may be influenced by multiple genes and also
gene–environment interactions may play a crucial role (Turkheimer et al.,
2014). Considering the large number of possible genetic variations, when
such evidence on specific genes based on the candidate gene approach
accumulates, more variance in a variable can be accounted for. Indeed,
recently, researchers started to generate polygenetic scores based on multiple genetic markers and found polygenetic scores can indeed explain more
variance than a single genetic marker (e.g., McCrae, Scally, Terracciano,
Abecasis, & Costa Jr., 2010).
The candidate gene approach has been adopted by organizational
researchers in their search for genetic markers associated with job satisfaction (Song, Li, & Arvey, 2011), leadership (Li et al., 2015), and job
changes (Chi, Li, Wang, & Song, 2016). Using the study on leadership as
one example, Li et al. (2015) found that the number of 10-repeat alleles on
a dopamine transporter gene, DAT1, was negatively related to proactive
personality, but positively related to moderate rule-breaking behaviors;
SCHYNS_9781785367274_t.indd 139
10/11/2017 15:19
140 Handbook of methods in leadership research
both proactive personality and rule breaking were positively related to
leadership role occupancy. Thus, the DTA1 gene had both positive and
negative indirect influences on leadership role occupancy, rendering the
total influences non-significant. The authors stated that having this gene
might be a “mixed blessing” (p. 671). This study has important implications for organizational genetics research. First, it shows that influences
of specific genes on work-related outcomes may be more complex than
expected. Researchers and lay people alike want to believe that there
might be a gene, or some genes, for leadership or entrepreneurship.
However, this research demonstrated that the influence of one gene may
be both positive and negative. In evolutionary genetics research, this is
called “stabilizing selection,” a natural selection process that “maintains
different alleles rather than favoring one allele over another” (Plomin et
al., 2013, p. 336). Second, it also has important implications for genetics
researchers looking for specific genes related to personality traits or other
behavioral outcomes. It underscores the importance of examining indirect mediating effects based on theories and previous empirical findings.
With the decreasing cost of gene sequencing, we urge more organizational
researchers to take a candidate gene approach, at least as a starting point,
to examine intriguing interplays between genes and environmental factors
on important leadership variables.
Genome-wide Association Studies (GWAS)
Although the candidate gene approach is capable of examining the genetic
effects of specific genes, this approach typically requires prior knowledge
or assumptions regarding which specific genes are potentially responsible
for a variable of research interest. In addition, this approach only focuses
on one or a very small number of genes. Recently the genome-wide association study (GWAS) method has been developed and has rapidly become
widely adopted in genetic research.
The GWAS method takes a bottom-up approach. With no prior
hypotheses required regarding which genes may be at work, this method
systematically scans thousands of genetic markers and even entire genomes
to examine the genetic variations, analyse the association strength between
various genetic variants and observed variables, and identify genetic variants that display statistically significant associations. The typical genetic
variants focused on in GWAS are single nucleotide polymorphisms (SNPs),
which are single base-pair variations with high occurring frequency in
a DNA sequence in the human genome (Genomes Project Consortium,
2010). An SNP has two alleles occurring in a pair basis and can be quantitatively characterized by the frequency of the less common allele (i.e.,
SCHYNS_9781785367274_t.indd 140
10/11/2017 15:19
Behavioral genetics and leadership research ­
141
the minor allele). Although SNPs exist in a molecular form, they can have
fundamental consequences for biological functions.
Furthermore, the recent advancement of genotyping technology –
­specifically the chip-based microarray technology – has made the GWAS
method easily accessible for many researchers from a wide array of
disciplines (Turkheimer, 2012). This new technology provides chipbased platforms that can measure and test genetic variation for millions
(recently up to five million) of SNPs with a relatively low-cost and easily
readable results. We encourage leadership researchers to take advantage
of such state-of-the-art approaches and technology to examine nuanced
interplays between the person and the environment in their inquiries. For
instance, research published in Science (Rietveld et al., 2013) found that
polygenetic scores based on various single genes explained approximately
2 percent (this is a big effect size in molecular genetics research) of the
variance in educational achievement. A recent study (Belsky et al., 2016)
using a similar approach revealed that polygenetic scores were significantly associated with social mobility and economic success, which were
mediated by individual difference variables such as intelligence and self
control.
REFERENCES
Arvey, R.D., Bouchard, T.J., Segal, N.L., & Abraham, L.M. (1989). Job satisfaction:
Environmental and genetic components. Journal of Applied Psychology, 74(2), 187–192.
Arvey, R.D., Li, W.D., & Wang, N. (2016). Genetics and organizational behavior. Annual
Review of Organizational Psychology and Organizational Behavior, 3, 167–190.
Arvey, R.D., Rotundo, M., Johnson, W., Zhang, Z., & McGue, M. (2006). The determinants
of leadership role occupancy: Genetic and personality factors. The Leadership Quarterly,
17(1), 1–20.
Arvey, R.D., Wang, N., Song, Z., Li, W., & Day, D. (2014). The biology of leadership. In
D. Day (Ed.), Oxford handbook of leadership and organizations (pp. 75–92). New York:
Oxford University Press.
Arvey, R.D., Zhang, Z., Avolio, B.J., & Krueger, R.F. (2007). Developmental and genetic
determinants of leadership role occupancy among women. Journal of Applied Psychology,
92(3), 693–706.
Bass, B.M., & Bass, R. (2008). The Bass handbook of leadership: Theory, research, and managerial applications (4th ed.). New York: Free Press.
Belsky, D.W., Moffitt, T.E., Corcoran, D.L., Domingue, B., Harrington, H., Hogan,
S.,. . .Caspi, A. (2016). The genetics of success: How single-nucleotide polymorphisms
associated with educational attainment relate to life-course development. Psychological
Science, 27(7), 957–972.
Bindl, U.K., & Parker, S.K. (2010). Proactive work behavior: Forward-thinking and changeoriented action in organizations. In S. Zedeck (Ed.), APA handbook of industrial and
organizational psychology (Vol. 2, pp. 567–598). Washington, DC: American Psychological
Association.
Bouchard, T.J. (2004). Genetic influence on human psychological traits. Current Directions
in Psychological Science, 13(4), 148–151.
SCHYNS_9781785367274_t.indd 141
10/11/2017 15:19
142 Handbook of methods in leadership research
Bouchard, T.J., & Propping, P. (1993). Twins as a tool for behavioral genetics. Chichester,
UK: John Wiley & Sons.
Bouchard, T.J., Lykken, D.T., McGue, M., Segal, N.L., & Tellegen, A. (1990). Sources of
human psychological differences: The Minnesota study of twins reared apart. Science,
250(4978), 223–228.
Cesarini, D., Johannesson, M., Lichtenstein, P., Sandewall, Ö., & Wallace, B. (2008). Is
financial risk-taking behavior genetically transmitted? IFN Working Paper No. 765.
Stockholm: Research Institute of Industrial Economics.
Chi, W., Li, W.D., Wang, N., & Song, Z. (2016). Can genes play a role in explaining turnover? An examination of gene–environment interaction from human capital theory. Journal
of Applied Psychology, 101(7), 1030–1044.
Cloninger, C.R., Adolfsson, R., & Svrakic, N.M. (1996). Mapping genes for human personality. Nature Genetics, 12(1), 3–4.
Cropanzano, R., & Becker, W.J. (2013). The promise and peril of organizational neuroscience today and tomorrow. Journal of Management Inquiry, 22(3), 306–310.
Day, D.V., Fleenor, J.W., Atwater, L.E., Sturm, R.E., & McKee, R.A. (2014). Advances
in leader and leadership development: A review of 25 years of research and theory. The
Leadership Quarterly, 25(1), 63–82.
Day, D.V., Harrison, M.M., & Halpin, S.M. (2009) An integrative approach to leader
­development: Connecting adult development, identity, and expertise. New York: Psychology
Press.
DeRue, D.S., Nahrgang, J.D., Hollenbeck, J.R., & Workman, K. (2012). A quasi-­
experimental study of after-event reviews and leadership development. Journal of Applied
Psychology, 97(5), 997–1015.
DeRue, D.S., Nahrgang, J.D., Wellman, N., & Humphrey, S.E. (2011). Trait and behavioral theories of leadership: An integration and meta-analytic test of their relative validity.
Personnel Psychology, 64(1), 7–52.
Dragoni, L., Tesluk, P.E., Russell, J.E.A., & Oh, I.S. (2009). Understanding managerial
development: Integrating developmental assignments, learning orientation, and access
to developmental opportunities in predicting managerial competencies. Academy of
Management Journal, 52(4), 731–743.
Eaves, L.J., Silberg, J.L., Meyer, J.M., Maes, H.H., Simonoff, E., Pickles, A.,. . .Hewitt,
J.K. (1997). Genetics and developmental psychopathology: 2. The main effects
of genes and environment on behavioral problems in the Virginia Twin Study of
Adolescent Behavioral Development. Journal of Child Psychology and Psychiatry, 38(8),
965–980.
Ebstein, R.P., Israel, S., Chew, S.H., Zhong, S., & Knafo, A. (2010). Genetics of human
social behavior. Neuron, 65(6), 831–844.
Ebstein, R.P., Novick, O., Umansky, R., Priel, B., Osher, Y., Blaine, D.,. . .Belmaker,
R. (1996). Dopamine D4 receptor (D4DR) exon III polymorphism associated with the
human personality trait of novelty seeking. Nature Genetics, 12(1), 78–80.
Edwards, J.R. (2008). Person–environment fit in organizations: An assessment of theoretical
progress. The Academy of Management Annals, 2(1), 167–230.
Fowler, J.H., & Dawes, C.T. (2008). Two genes predict voter turnout. The Journal of Politics,
70(3), 579–594.
Fowler, J.H., & Dawes, C.T. (2013). In defense of genopolitics. American Political Science
Review, 107(2), 362–374.
Fowler, J.H., Baker, L.A., & Dawes, C.T. (2008). Genetic variation in political participation.
American Political Science Review, 102(02), 233–248.
Freese, J. (2008). Genetics and the social science explanation of individual outcomes.
American Journal of Sociology, 114(S1), 1–35.
Frese, M., & Fay, D. (2001). Personal initiative: An active performance concept for work in
the 21st century. Research in Organizational Behavior, 23, 133–187.
Galton, F. (1869). Hereditary genius: An inquiry into its laws and consequences. London:
Macmillan.
SCHYNS_9781785367274_t.indd 142
10/11/2017 15:19
Behavioral genetics and leadership research ­
143
Genomes Project Consortium. (2010). A map of human genome variation from populationscale sequencing. Nature, 467(7319), 1061–1073.
Gottesman, I.I., & Shields, J. (1972). Schizophrenia and genetics: A twin study vantage point.
New York: Academic Press.
Grant, A.M., & Ashford, S.J. (2008). The dynamics of proactivity at work. Research in
Organizational Behavior, 28, 3–34.
Grant, A.M., Fried, Y., & Juillerat, T. (2010). Work matters: Job design in classic and contemporary perspectives. In S. Zedeck (Ed.), APA handbook of industrial and organizational
psychology (Vol. 1, pp. 417–453). Washington, DC: American Psychological Association.
Heaphy, E.D., & Dutton, J.E. (2008). Positive social interactions and the human body at
work: Linking organizations and physiology. The Academy of Management Review, 33(1),
137–162.
Hoffman, L.W. (1991). The influence of the family environment on personality: Accounting
for sibling differences. Psychological Bulletin, 110(2), 187–203.
Hopwood, C.J., Donnellan, M.B., Blonigen, D.M., Krueger, R.F., McGue, M., Iacono,
W.G.,. . .Burt, S.A. (2011). Genetic and environmental influences on personality trait
stability and growth during the transition to adulthood: A three-wave longitudinal study.
Journal of Personality and Social Psychology, 100(3), 545–556.
Ilies, R., & Judge, T.A. (2003). On the heritability of job satisfaction: The mediating role of
personality. Journal of Applied Psychology, 88(4), 750–759.
Ilies, R., Arvey, R.D., & Bouchard, T.J. (2006). Darwinism, behavioral genetics, and organizational behavior: A review and agenda for future research. Journal of Organizational
Behavior, 27(2), 121–141.
Johnson, W., McGue, M., & Iacono, W.G. (2006). Genetic and environmental influences on
academic achievement trajectories during adolescence. Developmental Psychology, 42(3),
514–532.
Johnson, W., Turkheimer, E., Gottesman, I.I., & Bouchard, T.J. (2009). Beyond heritability:
Twin studies in behavioral research. Current Directions in Psychological Science, 18(4),
217–220.
Judge, T.A., & Cable, D.M. (2004). The effect of physical height on workplace success and
income: Preliminary test of a theoretical model. Journal of Applied Psychology, 89(3),
428–441.
Judge, T.A., Bono, J.E., Ilies, R., & Gerhardt, M.W. (2002). Personality and leadership: A
qualitative and quantitative review. Journal of Applied Psychology, 87(4), 765–779.
Judge, T.A., Colbert, A.E., & Ilies, R. (2004). Intelligence and leadership: A quantitative
review and test of theoretical propositions. Journal of Applied Psychology, 89(3), 542–552.
Judge, T.A., Ilies, R., & Zhang, Z. (2012). Genetic influences on core self-evaluations, job
satisfaction, and work stress: A behavioral genetics mediated model. Organizational
Behavior & Human Decision Processes, 117(1), 208–220.
Kendler, K.S., & Baker, J.H. (2007). Genetic influences on measures of the environment: A
systematic review. Psychological Medicine, 37(5), 615–626.
Kendler, K.S., Neale, M.C., Kessler, R.C., Heath, A.C., & Eaves, L.J. (1994). Parental
treatment and the equal environment assumption in twin studies of psychiatric illness.
Psychological Medicine, 24(03), 579–590.
Kristof-Brown, A.L., & Guay, R.P. (2010). Person–environment fit. In S. Zedeck (Ed.), APA
handbook of industrial and organizational psychology (Vol. 3, pp. 3–50). Washington, DC:
American Psychological Association.
Lewin, K. (1935). A dynamic theory of personality. New York: McGraw-Hill.
Li, D., Sham, P.C., Owen, M.J., & He, L. (2006). Meta-analysis shows significant association between dopamine system genes and attention deficit hyperactivity disorder (ADHD).
Human Molecular Genetics, 15(14), 2276–2284.
Li, W.D. (2011). Proactive personality and work success: Disentangling genetic and environmental influences. Paper presented at the 2011 Annual Academy of Management Meeting,
San Antonio, Texas.
Li, W.D., Arvey, R.D., Zhang, Z., & Song, Z. (2012). Do leadership role occupancy and
SCHYNS_9781785367274_t.indd 143
10/11/2017 15:20
144 Handbook of methods in leadership research
transformational leadership share the same genetic and environmental influences? The
Leadership Quarterly, 23(2), 233–243.
Li, W.D., Fay, D., Frese, M., Harms, P.D., & Gao, X. (2014). Reciprocal relationships
between proactive personality and work characteristics: A latent change score approach.
Journal of Applied Psychology, 99(5), 948–965.
Li, W.D., Song, Z., & Arvey, R.D. (2011). The influence of general mental ability, self-esteem
and family socioeconomic status on leadership role occupancy and leader advancement:
The moderating role of gender. The Leadership Quarterly, 22(3), 520–534.
Li, W.D., Stanek, K., Zhang, Z., Ones, D.S., & McGue, M. (2016). Are genetic and environmental influences on job satisfaction stable over time? A three-wave longitudinal twin
study. Journal of Applied Psychology, 101(11), 1598–1619.
Li, W.D., Wang, N., Arvey, R., Soong, R., Saw, S.M., & Song, Z. (2015). A mixed blessing?
Dual mediating mechanisms in the relationship between dopamine transporter gene DAT1
and leadership role occupancy. The Leadership Quarterly, 26(5), 671–686.
Li, W.D., Zhang, Z., Song, Z., & Arvey, R. (2016). It is also in our nature: Genetic influences
on work characteristics and in explaining their relationships with well-being. Journal of
Organizational Behavior, 37(6), 868–888.
Loehlin, J.C. (2007). The strange case of c2 5 0: What does it imply for views of human development? Research in Human Development, 4(3–4), 151–162.
McCrae, R.R., Scally, M., Terracciano, A., Abecasis, G.R., & Costa, P.T., Jr. (2010).
An alternative to the search for single polymorphisms: Toward molecular personality
scales for the five-factor model. Journal of Personality and Social Psychology, 99(6),
1014–1024.
McGue, M., Bacon, S., & Lykken, D.T. (1993). Personality stability and change in early
adulthood: A behavioral genetic analysis. Developmental Psychology, 29(1), 96–109.
Miller, P., Mulvey, C., & Martin, N. (1995). What do twins studies reveal about the economic returns to education? A comparison of Australian and US findings. The American
Economic Review, 85(3), 586–599.
Munafò, M.R., Yalcin, B., Willis-Owen, S.A., & Flint, J. (2008). Association of the dopamine D4 receptor (DRD4) gene and approach-related personality traits: Meta-analysis
and new data. Biological Psychiatry, 63(2), 197–206.
Neale, M.C., & Cardon, L.R. (1992). Methodology for genetic studies of twins and families.
Dordrecht: Kluwer Academic.
Nicolaou, N., Shane, S., Cherkas, L., Hunkin, J., & Spector, T.D. (2008). Is the tendency to
engage in entrepreneurship genetic? Management Science, 54(1), 167–179.
Oldham, G.R., & Hackman, J.R. (2010). Not what it was and not what it will be: The future
of job design research. Journal of Organizational Behavior, 31(2–3), 463–479.
Parker, S.K., Andrei, D., & Li, W.D. (2014). An overdue overhaul: Revamping work design
theory from a time perspective. In A.J. Shipp & Y. Fried (Eds.), Time and work: How time
impacts individuals (Vol. 65, pp. 191–228). New York: Psychology Press.
Plomin, R., & Spinath, F.M. (2002). Genetics and general cognitive ability (g). Trends in
Cognitive Sciences, 6(4), 169–176.
Plomin, R., DeFries, J.C., Knopic, V.S., & Neiderhiser, J.M. (2013). Behavioral genetics (6th
ed.). New York: Worth.
Plomin, R., DeFries, J., & Loehlin, J.C. (1977). Genotype–environment interaction and correlation in the analysis of human behavior. Psychological Bulletin, 84(2), 309–322.
Plomin, R., Owen, M.J., & McGuffin, P. (1994). The genetic basis of complex human behaviors. Science, 264(5166), 1733–1739.
Ployhart, R.E., & Vandenberg, R.J. (2010). Longitudinal research: The theory, design, and
analysis of change. Journal of Management, 36(1), 94–120.
Preacher, K.J., Briggs, N.E., Wichman, A.L., & MacCallum, R.C. (2008). Latent growth
curve modeling. Los Angeles, CA: Sage.
Rietveld, C.A., Medland, S.E., Derringer, J., Yang, J., Esko, T., Martin, N.W.,. . .Koellinger,
P.D. (2013). GWAS of 126,559 individuals identifies genetic variants associated with educational attainment. Science, 340(6139), 1467–1471.
SCHYNS_9781785367274_t.indd 144
10/11/2017 15:20
Behavioral genetics and leadership research ­
145
Scarr, S., & Carter-Saltzman, L. (1979). Twin method: Defense of a critical assumption.
Behavior Genetics, 9(6), 527–542.
Scarr, S., & McCartney, K. (1983). How people make their own environments: A theory of
genotype → environment effects. Child Development, 54(2), 424–435.
Schneider, B. (1987). The people make the place. Personnel Psychology, 40(3), 437–453.
Schneider, B., Goldstein, H.W., & Smith, D.B. (1995). The ASA framework: An update.
Personnel Psychology, 48(4), 747–773.
Senior, C., Lee, N., & Butler, M.J.R. (2011). Organizational cognitive neuroscience.
Organization Science, 22(3), 804–815.
Shane, S., Nicolaou, N., Cherkas, L., & Spector, T.D. (2010). Genetics, the big five, and the
tendency to be self-employed. Journal of Applied Psychology, 95(6), 1154–1162.
Song, Z., Li, W.D., & Arvey, R.D. (2011). Associations between dopamine and serotonin
genes and job satisfaction: Preliminary evidence from the Add Health Study. Journal of
Applied Psychology, 96(6), 1223–1233.
Stogdill, R.M. (1948). Personal factors associated with leadership: A survey of the literature.
The Journal of Psychology, 25(1), 35–71.
Turkheimer, E. (2000). Three laws of behavior genetics and what they mean. Current
Directions in Psychological Science, 9(5) 160–164.
Turkheimer, E. (2012). Genome-wide association studies of behavior are social science. In
K. Plaisance & T. Reydon (Eds.), Philosophy of behavioral biology (pp. 43–64). New York:
Springer.
Turkheimer, E., Pettersson, E., & Horn, E.E. (2014). A phenotypic null hypothesis for the
genetics of personality. Annual Review of Psychology, 65, 515–540.
Waldman, D.A., Balthazard, P.A., & Peterson, S.J. (2011). Leadership and neuroscience:
Can we revolutionize the way that inspirational leaders are identified and developed? The
Academy of Management Perspectives, 25(1), 60–74.
Zhang, Z., Ilies, R., & Arvey, R.D. (2009). Beyond genetic explanations for leadership: The
moderating role of the social environment. Organizational Behavior and Human Decision
Processes, 110(2), 118–128.
Zyphur, M.J., Narayanan, J., Arvey, R.D., & Alexander, G.J. (2009). The genetics of economic risk preferences. Journal of Behavioral Decision Making, 22(4), 367–377.
SCHYNS_9781785367274_t.indd 145
10/11/2017 15:20
7.
Biosensor approaches to studying
leadership
Aurora J. Dixon, Jessica M. Webb and
Chu-Hsiang (Daisy) Chang
In spite of over 70 years of research conducted on the topic of leadership, it is a field that continues to grow and generate additional interest
(Day, 2012). For the purposes of this chapter, we define leadership as the
process through which one individual exerts influence on one or more
other individuals (Day & Antonakis, 2012). As an organizing framework
in this review, we consider the applicability of biosensor techniques for
research using all three of Bass’s (2008) general approaches for understanding leadership, namely: a leader-centric approach, leadership as
an effect, and the interaction between leaders and their followers. First
and most common is the leader-centric approach that focuses on stable
individual differences of leaders (e.g., conscientiousness), the behaviors
they exhibit (e.g., assigning roles; providing feedback), or on how leaders
influence the external environment and the individuals who follow them
(Bass, 2008; Day & Antonakis, 2012). Second, leadership has been conceptualized as an effect, such that followers’ actions (e.g., goal pursuit)
and reactions (e.g., satisfaction) are considered outcomes of the leader’s
influences (Day & Antonakis, 2012). Finally, Bass (2008) argues that
leadership can be defined by the relations between leaders and their followers (e.g., leader–member exchange; see Graen & Uhl-Bien, 1995 for a
review). These leader–member exchange (LMX) relationships can range
from high quality (e.g., characterized by trust and social support) to low
quality (e.g., characterized by distrust and social distance) depending on
the interactions within each leader–follower dyad. Together, these three
approaches have been valuable to the study of leadership because they
yield unique and complementary insights for understanding leadership as
both a process of influence (e.g., leader behaviors, LMX) and an outcome
of that influence (e.g., leader emergence, leader effectiveness).
The goal of this chapter is to review emerging biosensor methodologies
for studying leadership and examine how these methods have been applied
to “leader-centric”, “leadership as an effect”, and “leader–follower relations” conceptualizations of leadership. Biosensors are methods that
combine biology, chemistry, and technology to measure some electrical
146
SCHYNS_9781785367274_t.indd 146
10/11/2017 15:20
Biosensor approaches to studying leadership ­147
or chemical response in the body (Wang, 2012), thus we include methodologies used to study leadership as a process or outcome that are rooted
in biology, chemistry, neuroscience, and evolutionary psychology. These
methods include collecting saliva samples, drawing blood to examine
chemical markers, neuroimaging techniques (e.g., functional magnetic
resonance imaging – fMRI) used to examine the brain, and genetic information from DNA. The key value in using these methods is that researchers can identify physical explanations and physiological mechanisms
underlying leader traits and behaviors, follower actions and reactions in
response to leaders, and processes through which leader–member relationships develop. Because objective data are collected using biosensors,
these methods not only provide additional explanations to elucidate
leadership processes and effects, but also generate data that are free of the
biases associated with self-report questionnaires. For example, self-report
methods allow the potential for respondents to provide socially desirable
responses, which may not reflect the true state of the construct being measured (Podsakoff & Organ, 1986). In addition, when self-report methods
are used to measure multiple constructs, the relationships between those
constructs are subject to common method variance, which can lead to
erroneous estimates of the strength of those relationships (ibid.). Objective
biosensors can provide alternative data that may be used to corroborate
or revise prior findings that are based on more traditional methods such
as self-reported and observational data. In this chapter, other advantages
and disadvantages of biosensor approaches will also be discussed and recommendations for overcoming their limitations will be introduced.
BRIEF HISTORY OF BIOSENSOR APPROACHES IN
APPLIED SETTINGS
Biosensor methods are relatively recent additions to the study of workplace phenomena. A number of these methods are born out of the neuroscience movement in cognitive and social psychology, and have already
been shown to further illuminate general underlying mechanisms for social
and psychological phenomena. For example, Ochsner and Lieberman’s
(2001) review of the emerging neuroscience discussed the linkages of
various social and psychological phenomena to activation in the brain.
They describe studies that used fMRI to examine the association between
activation in the amygdala (i.e., the part of the brain responsible for emotions, decision-making, affective judgments, and memory) and reliance on
stereotypes (i.e., cognitive representations of a social group’s attributes or
behaviors that impact expectations of the social group) when ­perceiving
SCHYNS_9781785367274_t.indd 147
10/11/2017 15:20
148 Handbook of methods in leadership research
facial stimuli (e.g., Hart et al., 2000; Phelps et al., 2000). By using the
neuroimaging technique, the researchers expanded their understanding of
the underlying mechanisms by which individuals use stereotypes to inform
perceptions and judgments. Since the fMRI detected that stereotypical
perceptions and amygdala activation were associated with one another,
they reasoned that stereotypical perceptions were likely linked to positive or negative emotions about a specific social group. This then affected
decisions about how to interact with that social group. Thus, this biosensor method enabled researchers to further develop theory to explain how
stereotypes may influence perceptions, and to use objective evidence to
directly test the theoretical proposition.
Turning our attention more specifically to the workplace, social cognitive neuroscience research has been highlighted as a special topic and a
new research stream for the study of organizational behavior. Becker and
Cropanzano (2010) discussed three ways in which neuroscience techniques
can be used to understand behaviors at work, such as employees’ goal
selection, group climate perceptions, and resistance to organizational
change. In addition, Waldman, Balthazard, and Peterson (2011) advocated for the use of neuroscientific methods to understand inspirational
leadership. Scientists have also started using other biosensor approaches,
such as cardiovascular health indicators (e.g., Nyberg et al., 2009), to
understand the effects of the leadership process and the environmental
context (e.g., a stressful workplace) on individuals. Below, we review
various biosensor-based research methods that researchers have applied
to investigate the mechanisms underlying leadership. We will also identify
the specific biological characteristics or reactions (e.g., hormone concentration in the blood, activation in brain regions, genetic information) captured by these methods and the types of inferences that can be drawn from
such indicators within the context of leadership research.
BIOSENSOR-BASED RESEARCH METHODS IN
LEADERSHIP
Four classes of biosensor-based methods have been prominent in leadership research. The first category involves the collection of body fluids,
such as blood, saliva, and urine samples (e.g., Hansen, Larsen, Rugulies,
Garde, & Knudsen, 2009; Sherman et al., 2012). These samples are used to
test for specific biomarkers relevant to the research questions of interest.
For example, saliva and blood samples can be collected in order to test
for hormones that are associated with dominance and status relationships
(e.g., Hamilton, Carré, Mehta, Olmstead, & Whitaker, 2015). The hor-
SCHYNS_9781785367274_t.indd 148
10/11/2017 15:20
Biosensor approaches to studying leadership ­149
mones most typically studied in the leadership context are cortisol, a stress
hormone, (e.g., Sherman et al., 2012) and testosterone, a hormone linked
to aggression (e.g., Sellers, 2006). Alternative hormones that might be
relevant to leadership include estradiol and oxytocin, which are hormones
linked to nurturing (e.g., Knight & Mehta, 2014), but these have been less
commonly studied. These biomarkers may be particularly relevant for
understanding leader emergence, because they can elucidate the physiological explanations for why certain personality traits (e.g., extraversion)
are predictive of leader emergence (e.g., Judge, Bono, Ilies, & Gerhardt,
2002). Moreover, these biomarkers may also inform the physiological
basis for leader behaviors or leadership styles, such that testosterone may
be associated with leaders’ likelihood to adopt an authoritarian style
whereas oxytocin may be linked to leaders’ person-oriented or consideration behaviors.
Also in this category is the collection of blood and urine samples in
order to study biomarkers other than hormones. Research using these
bodily fluids can provide information about non-hormone markers, such
as cholesterol and glycated hemoglobin, which indicate responses to environmental conditions, such as the physiological effects of stressful work
environments that might be triggered by leader behaviors (e.g., Hansen et
al., 2009) or might be experienced by leaders in certain kinds of situations.
The second biosensor-based methodology that has been used to study
leadership is the monitoring of cardiovascular activity. Cardiovascular
monitoring devices include galvanic skin response (GSR) measurement
devices and heart rate monitors. Devices that measure GSR capture the
conductivity of the skin (e.g., Seemann, 1982), which indicates variations
in levels of cardiovascular activity (e.g., higher or lower heart rate). Heart
rate monitors directly measure heart rate, and can do so both at specific
time points and over longer periods of time (Smith & Jordan, 2015). In
leadership research, cardiovascular activity is typically used to assess
physiological effects of leadership on followers, such as follower wellbeing (e.g., Nyberg et al., 2009).
Third, neuroimaging methodologies are used to determine patterns
of brain activity. The first of these methods is electroencephalography
(EEG) (e.g., Waldman et al., 2013). In EEG, multiple electrodes are placed
at specific positions on a participant’s scalp to measure gross electrical
activity that originates in the neurons of the brain. The placement of the
electrodes indicates which areas of the brain have greater electrical activity
at any given time, indicating the activation patterns of the brain at different times and in different scenarios (Mayo Clinic, 2016a). These data
are analysed using calculations that look at metrics such as how powerful
the EEG signal is in different areas of the brain and how similar patterns
SCHYNS_9781785367274_t.indd 149
10/11/2017 15:20
150 Handbook of methods in leadership research
of electrical activity are across different areas of the brain (e.g., Harung,
Travis, Blank, & Heaton, 2009). EEG is used in a number of different
research areas, including leader performance (ibid.) and follower reactions
to leader behaviors (Dinh, 2014).
The second neuroimaging method used in leadership research is fMRI
(e.g., Fairhurst et al., 2014). fMRI involves placing a participant into a
large magnet, which is used to track blood flow in the brain over time.
During this time, participants can be presented with different scenarios or
stimuli if desired. Increased blood flow to an area of the brain indicates
its activation, thereby mapping the brain activities of the participant (UC
San Diego, 2016). These maps are used to calculate percentages of signal
change in brain areas (e.g., Molenberghs, Prochilo, Steffens, Zacher, &
Haslam, 2015) or changes in brain area activation over time (e.g., Boyatzis
et al., 2012). fMRI has been used to study a variety of leadership outcomes, including leader performance (e.g., Gilkey, Caceda, & Kilts, 2010)
and reactions of followers to different forms of leadership (e.g., Boyatzis
et al., 2012).
Finally, biosensor research on leadership can also be broadly defined
to include research on genetics and evolutionary characteristics (see also
Chapter 6 in this volume). This methodology considers leader behaviors that may have developed as a result of genetics and evolution, and
focuses on investigating hereditary traits of leaders and followers (e.g.,
Ilies, Gerhardt, & Le, 2004). For example, Li et al. (2015) look at genetic
differences among leaders and followers. Neurotransmitter levels are
sometimes collected in conjunction with genetic information when studying leadership. Neurotransmitter research typically focuses on the specific
neurotransmitters of serotonin (e.g., Summers & Winberg, 2006), which
is linked to the inhibition of aggression, and dopamine, which is linked to
personality through its relationship to positive emotionality (e.g., Depue,
Luciana, Arbisi, Collins, & Leon, 1994). These neurotransmitters are
often studied as the more proximal predictors for explaining how genes
may influence individuals’ behaviors or characteristics. Animal studies
have also provided evidence for linking neurotransmitters to various
behavior patterns similar to humans’ (e.g., Schneirla, 1959).
Advantages and Disadvantages of Biosensor Methods
While each of the tools used to study bioindicators is unique and provides
a different type of data for researchers to analyse, they share a number
of advantages and disadvantages, with some variations depending on
the method. First, biosensor methods share the benefits of being generally unobtrusive. Especially important, they do not require interruption
SCHYNS_9781785367274_t.indd 150
10/11/2017 15:20
Biosensor approaches to studying leadership ­151
of the leader–follower interaction in order to collect data when repeated
measurement or continuous monitoring is of interest. For instance, heart
rate monitors can be worn throughout the day, and fMRI and EEG can
be used to collect data during an experimental task simulating leader–­
follower interaction. This ability to collect data at exactly the same time
the participant is experiencing a stimulus, engaging in a task, or interacting
with one or more individuals gives biosensor methods a strong advantage
over techniques such as self-report surveys that either require disrupting
the task or retrospective reporting.
A second advantage of biosensor methods is that all of them can generate objective data (Becker & Menges, 2013). Participants cannot easily
consciously or unconsciously manipulate them. Biosensors minimize or
eliminate concerns such as social desirability, impression management,
inattention, or other human information processing biases that are
common with self-reported data. Thus, biosensor data can provide additional information beyond the survey ratings. Moreover, it may be possible to verify the reliability and robustness of the raw data by reanalysing
the original samples collected.
Additionally, biosensor methods allow researchers to gain insight into
implicit processes that people have no ability to observe themselves (ibid.).
These methods allow researchers to understand participants’ physiological reactions to leadership-related stimuli that would be unobservable
using self-report methods. For instance, fMRI has been used to identify
whether people emerge as leaders or followers in tasks in which they
have to coordinate their behaviors with other team members’ behaviors
(Fairhurst et al., 2014). In such interactions, a person may not be able
to identify whether they were leading or following through self-report.
However, their brain activities can provide insights into the leader emergence process. Such observations are valuable in their own right, and may
be even more of interest when paired with more commonly used measurement techniques.
Nevertheless, there are disadvantages to overcome with biosensor
methods. While biosensor approaches are more objective than self-report
surveys, some data collection procedures may be perceived by participants
as intrusive. Collection of biological markers may require participants to
provide saliva by holding cotton swabs or rope in their mouths, collect
urine in a receptacle, or draw blood samples with a needle. Similarly, individuals are required to wear potentially uncomfortable electrodes on their
head with an EEG or to stay completely still in a tight space for long durations in an fMRI. These methods require much more physical involvement
on the part of the participants than traditional survey research. And,
while none of the techniques typically involves more than discomfort at
SCHYNS_9781785367274_t.indd 151
10/11/2017 15:20
152 Handbook of methods in leadership research
worst, some participants might find them anxiety provoking, distasteful
or experience them as invading personal privacy. As such, it is important
for researchers to clearly explain what is expected of participants to institutional review boards (IRBs), which review ethical conduct of research
based on human subjects to ensure that the research upholds ethical
standards. IRBs have increased requirements for reporting when using
biosensor methodologies. Because many of these methods (e.g., drawing
blood) can be considered medical procedures, there are more constraints
on this research. Medical IRBs must approve these methods, and the data
must be handled especially carefully due to its sensitive nature (U.S. Food
and Drug Administration, 2016). It is also important to explain these
expectations directly to participants during an informed consent procedure to ensure that they understand what they are agreeing to do, and also
so that they fully understand the difference between a procedure engaged
in for research purposes versus something done for purposes of medical
diagnosis.
Researchers should be fully aware of the potential discomforts or risks
to participants with the methods that were just described. Moreover, they
should be knowledgeable about the practical risks and requirements for
them to collect these data. For example, when biomarkers involve the
collection of bodily fluids (especially for blood draws requiring venipuncture), there may be limits on the amount that can be taken over a period
of time, restrictions on the personnel who are qualified to do the collection
and the setting in which it can occur, emergency procedures that must be
in place, and considerations related to avoiding contact with blood and
other bodily fluids, such as use of rubber gloves, disinfectants and biohazard waste disposal. And, especially if biologic specimens must be collected
over an appreciable length of time, there may be requirements for transporting and preserving samples prior to chemical analysis, such as special
containers and low-temperature freezers. Researchers should also have
clearly worked out what their responsibilities and protocols are in case a
measurement taken as part of a research study is suggestive of a hidden
medical condition in a participant.
Because of the complications related to data collection, gaining access
to biosensor data may require additional resources from the researchers.
Collecting biosensor data requires financial resources, equipment and
skills that may not be readily available to many leadership researchers.
For example, leadership researchers may need to rely on external laboratory services (e.g., Salimetrics) to process their saliva, blood, or urine
samples for hormones and genetic markers. They may need to purchase
equipment (e.g., headsets) and processing software to make sense of the
EEG data, since the data require additional processing above and beyond
SCHYNS_9781785367274_t.indd 152
10/11/2017 15:20
Biosensor approaches to studying leadership ­153
its immediate output. Finally, leadership researchers may rent out MRI
scanning time from local laboratories or medical centers to conduct their
research, which can be quite expensive (the current hourly rate for renting
an fMRI scan is approximately $600). After this rental, the processing of
fMRI data can also be time intensive. Because of the requirement for additional resources, external funding may be crucial for successful leadership
research involving biosensors. Moreover, researchers with expertise in the
leadership literature would do well to partner with other researchers who
are familiar with the relevant physiological or biochemical knowledge and
biosensor techniques if they wish to conduct studies using biosensor technology. Such interdisciplinary research teams are more likely to generate
high-quality research findings.
There can also be limitations on the interpretation of results from
biosensor methods once the researcher has them. As noted previously,
biosensor approaches assess physiological indicators, which are used to
indirectly infer participants’ psychological processes and reactions. While
these approaches provide direct, objective comparisons of participants’
hormones, brain activities, and so on, they do not provide a direct indication of specific psychological processes experienced by the participant,
nor do they test the inferred associations between bioindicators and the
psychological phenomenon (e.g., Waldman et al., 2011). Lindebaum and
Zundel (2013) argue that cognitive neuroscience approaches to leadership
(e.g., fMRI, EEG) risk oversimplifying complex behaviors by reducing
social phenomena to activation in select areas of the brain. They argue
that the reduction of these phenomena to singular biological processes
does not provide the whole picture of individuals’ rich experiences. By
using biosensor approaches, researchers may not be able to capture the
full context in which the leadership phenomenon is embedded.
Additionally, researchers have noted that current biosensor approaches
also lack sophistication when it comes to providing a full understanding
even of underlying physiological mechanisms (Waldman et al., 2011).
For example, EEG provides an overall picture of how electrical current
proceeds through the brain, but it is not perfectly precise, as the spatial
resolution of EEG is low (Becker & Menges, 2013). Similarly, fMRI provides information concerning where blood is flowing in the brain and gives
a general idea of which areas of the brain are activated. However, it does
not pinpoint exactly where the blood is flowing or why it is going there.
Thus, leadership researchers will need to make inferential leaps based
on a combination of leadership theory and knowledge of physiological
functioning to reason why blood is flowing to an activated area based on
leadership theory and what is known about the characteristics of that area
of the brain.
SCHYNS_9781785367274_t.indd 153
10/11/2017 15:20
154 Handbook of methods in leadership research
Ochsner and Lieberman’s (2001) review of a link between stereotyped
perceptions and activation in the amygdala provides a good example of
such inferential leaps. While there is evidence that the amygdala is linked
to emotions, affective judgments, and memory, researchers cannot say
definitively that the amygdala causes stereotyped perceptions. Researchers
can infer that perception and the functions of the amygdala are related
because the amygdala is associated with affective judgments (e.g., making
a snap judgment about a social group) and blood was shown flowing to
this general area when stereotypical perceptions occurred. With this in
mind, researchers using biosensor approaches should consider using these
methods in conjunction with traditional psychological research methods
(e.g., surveys, observation) to capitalize on a multi-method approach to
studying leadership. Since these biosensor methods are relatively new in
the leadership domain, Becker and Cropanzano (2010) argue that they
should be used to complement traditional methods of observation. In this
way, the leadership field can establish the validity of biosensor methods by
comparing results generated from these different methods.
Finally, researchers using biosensor approaches should also take care
to ensure that these methods are grounded in theory. At the present time,
researchers using neuroimaging tend to look at where blood is generally
flowing in the brain and use this as evidence for a psychological phenomenon. Lee and Chamberlain (2007) state that there is often little a priori
theoretical reasoning as to why blood should flow to the specific areas
and why the blood flow is indicative of evidence of the phenomenon in
question. Thus, researchers who choose to use biosensor approaches in
leadership should evaluate theories from psychology, management, and
the domain from which their methodological approach is originated to
develop a priori hypotheses as to why certain physiological evidence
should support the existence of a leadership phenomenon. As biosensor approaches become more popular worldwide, caution should also
be taken to study the equivalence of these methods across cultures (e.g.,
individualistic versus collectivistic countries). According to Lowe and
Gardner (2000), researchers should seek to understand if leadership can
be measured using the same methodology and criteria across different
cultures, which would require more emic (e.g., within-culture) versus etic
(e.g., cross-culture) research studies.
Despite the limitations of biosensor approaches described above, the
potential benefits of using these techniques can outweigh the problems
with invasiveness, sophistication, and theoretical grounding, especially
if leadership researchers collaborate with biosensor experts and ground
their methods in theory. Below, we review the literature of biosensorbased research, providing examples of biosensor-based studies focused
SCHYNS_9781785367274_t.indd 154
10/11/2017 15:20
Biosensor approaches to studying leadership ­155
on leader-centric approaches, understanding the effect of leadership on
others, and understanding the interactions between leaders and followers.
BIOSENSOR METHODS APPLIED TO LEADERCENTRIC LEADERSHIP RESEARCH
Leader-centric research assesses leaders’ individual traits, their behaviors, and their influence on others (Bass, 2008; Day & Antonakis, 2012).
Leader-centered research using biosensors has investigated the topics of
emergent leadership, abusive supervision, leader performance, and transformational leadership. These topics have been investigated by examining
a number of different biosensors, such as including hormones and genetics.
Biosensor-based Studies of Emergent Leadership
Hormone assessment is often used to compare leaders and non-leaders.
Sellers (2006) found that salivary testosterone is related to higher levels of
status seeking for both men and women. Moreover, she found that leaders
tend to have higher levels of testosterone than non-leaders. Because of the
known associations between testosterone and dominance and aggression,
these findings suggest that holding a leadership role is related to aggression, although the causal order of the two variables cannot be clearly
established. On the one hand, it is possible that aggressive and dominant
individuals are more likely to rise to leadership roles. On the other hand,
occupancy of leadership roles may result in leaders exerting more dominance and influence. As such, understanding variations in levels of salivary testosterone may help identify leaders, or those who may emerge as
leaders, for future research efforts.
Smith and Jordan (2015) found that individuals who are concerned with
status threat and acceptance threat have higher levels of salivary cortisol.
Status threat can be considered a threat to one’s competence or leadership potential, whereas acceptance threat describes a threat to one’s likeability or inclusion by others. As leaders tend to engage in status-seeking
behaviors, it is possible that leaders may also have higher levels of cortisol,
especially when they experience threats to their power.
Although individuals who are particularly concerned with their status
have higher testosterone and cortisol (ibid.), research has identified a more
complex pattern between these hormones and leader emergence. Multiple
researchers (Hamilton et al., 2015; Knight & Mehta, 2014; Sherman et
al., 2012; Sherman, Lerner, Josephs, Renshon, & Gross, 2016) have demonstrated that cortisol and testosterone interact and impact individuals’
SCHYNS_9781785367274_t.indd 155
10/11/2017 15:20
156 Handbook of methods in leadership research
status within organizations. These two hormones interact in such a way
that individuals who have high levels of testosterone and low levels of
cortisol are more likely to gain power and status and emerge as leaders
in organizations. While this result is inconsistent in some studies (e.g.,
Westendorp, 2012), the majority of research supports the association
between the hormone pattern of high testosterone and low cortisol and
leader status. Research on variations in levels of these hormones in leaders
may be able to help inform who among incoming employees may emerge
as leaders based on their hormone levels. Additionally, this research may
inform understanding in how leaders and followers experience and react
to status-related stressors.
Genetic and evolutionary characteristic research has focused on which
genes and hereditary traits are linked to leadership emergence. Zhang,
Ilies, and Arvey (2009) discussed twin research on leader role occupancy,
highlighting that genetic influences play a role in leadership positions,
especially for individuals from poorer social environments. Using twins as
their participants, De Neve, Mikhaylov, Dawes, Christakis, and Fowler
(2013) estimated a 24 percent heritability of leader role occupancy, and
demonstrated that leader role occupancy is linked to a specific gene,
CHRNB3. Similarly, Li, Arvey, Zhang, and Song (2012) and Chaturvedi,
Zyphur, Arvey, Avolio, and Larsson (2012) both studied twins, finding
that leader role occupancy was predicted by the genetic makeup of the
participants. Li et al. (2015) focused on DAT1, a dopamine transporter
gene, collected from saliva. They found that DAT1 10-repeat allele was
related to leader role occupancy, and this effect was mediated by personality trait manifestation. Specifically, DAT1 is negatively linked to proactive
personality, which is positively linked to leader role occupancy. However,
they also found that DAT1 is positively related to rule-breaking behavior,
which in turn is positively related to leader role occupancy. While these
results overall led to no effect of DAT1 on leader role occupancy, they
point to the complex role of genetics in leader emergence. By understanding the link between genetics and leader role occupancy, and the intermediate trait manifestation, we may be able to better predict who will emerge
as leaders based on individuals’ genetic makeups.
Finally, research on neurotransmitters has indicated that dopamine
levels also relate to different personality traits (Depue et al., 1994), similar
to the research by Li et al. (2015). High levels of dopamine have been
linked to behaviors in animals that are similar to positive emotionality
(i.e., desire, incentive-reward motivation) (Schneirla, 1959), including in
rats (McGinty, Lardeux, Taha, Kim, & Nicola, 2013). Based on findings
from animal-based research, it is expected that behaviors associated with
positive affect, and therefore positive emotionality, are related to high
SCHYNS_9781785367274_t.indd 156
10/11/2017 15:20
Biosensor approaches to studying leadership ­157
levels of dopamine in humans. Depue et al. (1994) argue that positive
emotionality is predictive of leadership emergence, and as such, dopamine
may be linked to leader emergence through its positive impact on positive
emotionality.
Leader Behaviors
In addition to examining leader role occupancy, researchers have also
applied biosensor methods to examine positive and negative leader behaviors. In terms of positive leader behaviors, neuroimaging research has been
applied to characterizing brain activity patterns among transformational
leaders who inspire their followers with idealized influence, intellectual
stimulation, inspirational motivation, and individualized consideration
(Balthazard, Waldman, Thatcher, & Hannah, 2012). Balthazard et al.
(2012) collected resting EEG data from civilian and military leaders, and
also collected follower ratings of those leaders’ transformational leadership style. They used patterns of brain activity to classify leaders as high
or low on transformational leadership. In a test sample, leaders were
categorized into high (1 SD above the mean) versus low (1 SD below the
mean) transformational leaders based on their followers’ ratings of their
transformational leadership. Leaders high in transformational leadership
consistently exhibited clearer and more distinct EEG patterns with more
differentiated brain activities compared to those rated lower on transformational behaviors. These variations in the patterns of brain activities
were then used to predict whether leaders were transformational versus
not using a second validation sample. This classification achieved 92.5
percent accuracy in categorization. Their results suggest that transformational leaders share a common neural activation pattern, whereas non-­
transformational leaders do not have a consistent pattern.
In addition to positive leader behaviors, biosensor approaches have
also been applied to understanding the mechanisms underlying negative
leader behaviors. For example, research on leaders’ abusive behaviors has
typically focused on identifying hormones that are associated with these
negative behaviors. In a lab study, Bendahan, Zehnder, Pralong, and
Antonakis (2015) found that salivary testosterone was positively related
to the level of corruption of participants placed in a simulated leadership position. In this case, corruption was operationalized by whether
participants would increase their own earnings at the expense of their
group’s earnings. In addition, they also found that participants’ power,
measured by number of followers, interacted with their testosterone to
predict corruption. Overall, like testosterone, power was positively related
to antisocial behavior. However, the positive relationship between power
SCHYNS_9781785367274_t.indd 157
10/11/2017 15:20
158 Handbook of methods in leadership research
and antisocial behavior was amplified in the high versus low testosterone
participants. These findings point to the role of testosterone in explaining how power may corrupt leader behaviors, such that when power and
testosterone were both at their highest, leaders were more likely to act in
an antisocial way.
Westendorp (2012) found that men who have high levels of salivary
testosterone are often higher in psychopathy as well. In turn, psychopathy
was positively related to eight factors of leadership: creative thinking,
empathy, charm, agreeableness, risk taking, need for achievement, need
for affiliation, and taking charge. These results suggest that testosterone
and psychopathy are both positively related to leadership among males.
Because psychopaths do not experience guilt or remorse, the positive association between psychopathy and leadership may explain why leaders may
engage in antisocial behaviors to exploit others (ibid.). This relationship
between psychopathy and antisocial behavior and exploitation points to
concerns about how these leaders may treat their followers.
Leader Performance
Research on leader performance has primarily used neuroimaging methodologies to characterize the brain activities of effective leaders. Gilkey et
al. (2010) found that individuals with high levels of emotional intelligence
have different brain activity patterns from those with low levels of emotional intelligence. Using fMRI, Gilkey et al. (2010) found that leaders
who self-reported high emotional intelligence tend to use areas of the brain
associated with empathy and intuitive responses when reacting to fictional
strategic and tactical management dilemmas. Harung and Travis (2012)
found that electrical brain activity patterns captured by EEG are related
to performance of leaders. More effective performers, operationalized
as managers with top-level positions in their organizations, have higher
integration of electrical brain activities than do average performers, or
managers at lower levels in organizations. Although Harung and Travis
used leaders’ level within the organization as a proxy for performance, this
operationalization of leader performance may be problematic. Leaders
of different levels within the organization may not only have different
performance, but may also be different in other factors such as tenure,
experience, and power levels. These additional differences may account for
different EEG patterns observed between high- versus low-level leaders.
Hannah, Balthazard, Waldman, Jennings, and Thatcher (2013) found
that adaptive leaders show different EEG brain activity patterns than
non-adaptive leaders when engaging in a military scenario designed to
assess adaptive behavior. These EEG patterns predicted leaders’ adaptive
SCHYNS_9781785367274_t.indd 158
10/11/2017 15:20
Biosensor approaches to studying leadership ­159
decision making above and beyond their self-reported complexity of selfconcepts. These findings point to a link between performance and brain
activities, and may be used both in predicting performance and determining which leaders will perform in more effective or less effective ways.
Summary
Biosensor research focusing on leader-centric topics has broadly explored
various facets of leadership. Research on emergent leadership has been
conducted using hormone and neurotransmitter assessments and genetic
methods. Leader behaviors, both positive and negative, have been studied
using neuroimaging, genetics, and hormones. Finally, leader performance
has been linked to different brain activity patterns using neuroimaging
methods. These results indicate that testosterone is an important hormone
for studying antisocial leader behaviors (e.g., Bendahan et al., 2015), and
that testosterone and cortisol can provide further understanding of leadership emergence (e.g., Hamilton et al., 2015). Additionally, genes such as
CHRNB3 (De Neve et al., 2013) and DAT1 (Li et al., 2015) can inform
leadership emergence, as can the neurotransmitter dopamine (Depue et
al., 1994). Finally, neuroimaging may be used to assess positive leader
behaviors (e.g., transformational leadership: Balthazard et al., 2012) and
leader effectiveness (e.g., Gilkey et al., 2010).
BIOSENSOR METHODS APPLIED TO THE IMPACT
OF LEADERSHIP ON OTHERS
Biosensor-based research studying the impact of leaders’ behaviors on
others has focused on indicators of follower well-being, such as cardiovascular activities, in relation to leader behaviors. Earlier research on leadership styles pointed to a limited impact of leadership on follower health.
For example, Seemann (1982) found similar galvanic skin responses
among followers of both democratic versus autocratic leaders. This suggests that the two types of leader behaviors are associated with similar
arousal levels for their followers, suggesting that there is no difference in
the stressfulness elicited by the two leadership styles. However, more recent
studies have painted a different picture, suggesting that leader behaviors
may have a significant impact on follower well-being, as assessed through
blood and urine markers (Hansen et al., 2009).
Specifically, cardiovascular research indicates that certain leadership
styles are better for followers’ cardiovascular health. For instance, managerial leadership is characterized by leaders’ concrete behaviors that
SCHYNS_9781785367274_t.indd 159
10/11/2017 15:20
160 Handbook of methods in leadership research
support a healthy psychosocial work environment, such as consideration
and transformational leadership behaviors (Nyberg et al., 2009). Nyberg
et al. (2009) found that the subordinate-rated level of managerial leadership was negatively related to the cardiovascular problems of followers, as
operationalized by hospital admissions for acute myocardial infarction or
unstable angina and deaths from ischemic heart disease or cardiac arrest.
In another study, Smith and Jordan (2015) found that individuals’ socialevaluative threat (i.e., concern that others will evaluate them negatively)
is positively related to blood pressure and heart rate. While this study did
not focus on followers specifically, social-evaluative threat may be related
to followers’ concerns about leaders’ judgments. This finding suggests that
followers’ heart health may suffer if their leaders are too critical, or are less
skilled in delivering negative feedback in a constructive manner.
Other follower-centric research uses biosensor methods to capture followers’ brain activities in response to leaders’ behaviors. Molenberghs et
al. (2015) found that individuals who were listening to inspirational versus
non-inspirational statements from in-group versus out-group political
leaders had different patterns of brain activity. There was a main effect
for statement type, such that individuals showed greater brain activities
in areas associated with negative or norm-breaking information, such as
the left lateral orbitofrontal cortex, the adjacent pars triangularis, and the
angular gyrus, when listening to non-inspirational statements compared to
when listening to inspirational statements. This main effect was qualified
by an interaction between statement type and leader status. When listening to inspirational statements from in-group leaders, individuals showed
activity in areas of the brain associated with semantic information processing, which indicates deeper processing and encoding of information.
When listening to non-inspirational statements from in-group leaders,
individuals showed activity in areas of the brain associated with reasoning
about mental states. The opposite was true when listening to out-group
leaders. These results identified cognitive mechanisms to explain important boundary conditions (i.e., follower identification with the leader) for
the effectiveness of leaders’ inspirational messages.
Beyond studying effective leader behaviors, researchers also explore
the brain activity of followers in response to leaders’ unethical behaviors.
Dinh (2014) measured electrical activity in the brains of followers using
EEG and found that followers made implicit self-evaluations about their
own ethicality in response to leaders’ dark leadership behaviors (e.g., ridiculing, invasion of privacy, and other abusive supervisory behaviors). Her
results showed that participants evaluated themselves as more associated
with dark attributes and acted less ethically when assigned to a supervisor
who acted unethically than when assigned to a supervisor who acted sup-
SCHYNS_9781785367274_t.indd 160
10/11/2017 15:20
Biosensor approaches to studying leadership ­161
portively toward a fictitious follower. These behaviors and evaluations
were related to differences in brain activity patterns between individuals
with differing self-evaluations of dark attributes, as captured by EEG
(measured during interactions with the leader and during the time participants were making self-evaluations). This suggests that leaders’ impact on
follower self-perceptions and their subsequent moral behaviors may be
explained by different brain activities in followers.
Summary
Biosensor methodologies have been used in research focusing on the
effects of leaders on followers. Specifically, research has examined followers’ well-being as an outcome of different leadership styles. Different
leader behaviors are associated with followers’ different brain activity patterns. This research provides a more complete understanding of the effects
of leaders on their followers, which complements prior research that is primarily based on self–reported measures to capture followers’ experiences.
BIOSENSOR METHODS APPLIED TO INTERACTIONBASED LEADERSHIP RESEARCH
Biosensor research on interactions between leaders and followers has
focused on the topics of power, status, and dominance, and leader–­
follower relationships. These topics have primarily been examined using
neuroimaging methodologies.
Neuroimaging research on social status has shown that the brain reacts
to changes in status. Beasly, Sabatinelli, and Obasi (2006) argued that
human brains are adjusted to navigating social systems based on status.
They reviewed neuroimaging research, which has demonstrated that neural
networks appear to be “wired” for navigating social situations based on
status and power. Specifically, research using fMRI indicates that brain
activity patterns vary depending on whether a person is interacting with a
person of higher versus lower status in a hierarchy. When interacting with
a person of higher status within a stable hierarchical structure, individuals
have higher levels of activities in various areas of the brain (e.g., the bilateral occipital/parietal cortex, striatum, parahippocampal cortex, and the
dorsolateral prefrontal cortex) than during other interactions. However,
when interacting with a person of lower status within a stable hierarchical structure, no unique activation pattern is observed. In unstable
hierarchies, interacting with a person of higher status involved even more
areas of the brain, and demanded more cognitive resources. During these
SCHYNS_9781785367274_t.indd 161
10/11/2017 15:20
162 Handbook of methods in leadership research
interactions, the bilateral thalamus, right amygdala, posterior cingulate,
medial prefrontal cortex, premotor cortex somatosensory cortex, and supplementary motor area were activated. These results indicate that social
status processing, specifically with higher-status individuals and in more
complex environments, demands more processing and energy.
In addition to activation patterns, there also appeared to be structural
differences across individuals of chronic high versus low status as identified by magnetic resonance imaging (MRI) to assess the structure of the
brain (Mayo Clinic, 2016b). Beasley et al. (2012) found that individuals
who perceived themselves as low status had less gray matter (i.e., brain
cells) in brain regions associated with coping with emotional and psychosocial stressors compared to those who perceived themselves as having
higher status. These results point to an impact of the perception of status
on individual outcomes. These results may relate to the brain structures
of leaders and followers, although conclusions are not clear, as there are
many environmental factors involved in brain development.
Biosensor research has also assessed the neurotransmitters involved
during interactions between individuals of different status. In interactions
between individuals of varying social status, such as leaders and followers,
serotonin and cortisol interact to reduce aggressive reactions by low-status
individuals toward high-status interaction partners (Summers & Winberg,
2006). Corticosteroids increase serotonin production, and serotonin inhibits aggressive responses. Reduction in aggressive tendencies benefits both
interaction partners, such that it creates a low-stress environment for both
high- and low-status individuals to facilitate interaction. Moreover, it
­benefits the high-status individuals by reinforcing their dominance.
Studies using fMRI and EEG have indicated that interactions between
leaders and followers can be studied using patterns of brain activity.
Jack, Boyatzis, Khawaja, Passarelli, and Leckie (2013) coached participants using positive emotional attractor (PEA) coaching given by an
interviewer, which uses positive motivational factors aimed at changing
a person’s behaviors. This type of coaching is designed to motivate followers to change their behaviors through inspiration. The interviewing
technique used in this study simulated a coaching relationship, such that
the interviewer acted as a leader and the participant as a follower. PEA
coaching is contrasted with the negative emotional attractor (NEA)
coaching. Rather than inspiration, NEA coaching focuses on exercising
willpower and training people to monitor themselves, which is expected
to deplete individuals’ executive resources. PEA coaching improved
interactions between the leader (interviewer) and the follower (participant), and produced different patterns of brain activity in the participants
(ibid.). Individuals coached using PEA showed greater activation of the
SCHYNS_9781785367274_t.indd 162
10/11/2017 15:20
Biosensor approaches to studying leadership ­163
parasympathetic nervous system, which is responsible for regulating the
body’s rest-and-digest functions, and areas associated with positive affect.
Participants coached with the NEA technique had greater activation of
the sympathetic nervous system, which is responsible for regulating the
body’s fight-or-flight responses, and areas associated with negative affect.
These results identify ways in which differences in brain activation are
associated with different types of interactions between leaders and followers, and how different interactions may be driven by different leader
behaviors.
Researchers have also used neuroimaging to assess the impact of the
quality of leader–follower relationships on the brain activity of followers.
The impact of resonance and dissonance between followers’ preferences
and leaders’ styles has been explored via fMRI. Resonance is defined as
“physiological attunement and interpersonal synchrony between a leader
and another individual” (Boyatzis et al., 2012, p. 261), and dissonance
is defined as the absence of those characteristics. Resonance between a
leader and a follower is a marker of a high-quality relationship, whereas
dissonance is a marker of a low-quality relationship. Boyatzis et al. (2012)
found that when followers reflected on resonant relationships with a
leader, brain areas associated with positive affect and mirror neurons were
activated. However, when followers reflected on dissonant relationships,
areas associated with lower attention and negative affect were activated.
Studies examining both leader and follower brain activities show differences in activity patterns between leaders and followers during their
interactions with one another. For example, Konvalinka et al. (2014)
collected EEG data from leader–follower pairs completing a joint task,
and found that only the leader, but not the follower, exhibited the key
pattern of brain activity associated with the task. Fairhurst et al. (2014)
found that when paired with virtual partners, certain individuals emerged
as leaders, and others as followers, as distinguished both by brain activity patterns captured by fMRI and task preferences of characteristics of
the tapping task they were assigned to. The researchers identified leaders
as the individuals who found it easier to synchronize with their virtual
partner when they perceived more control over the tempo of tapping in
the task. Followers, on the other hand, found it easier to synchronize
their tapping with their partner’s tapping when they perceived that the
virtual partner was more in control of the tempo. fMRI data found that
leaders showed greater brain activity than did followers in areas associated with self-­initiated actions, indicating that leaders were directing their
own behaviors while followers were not. Additionally, leaders focused less
on error correction and more on prioritizing maintaining tempo versus
matching their partner (ibid.).
SCHYNS_9781785367274_t.indd 163
10/11/2017 15:20
164 Handbook of methods in leadership research
Summary
Research on interactions between leaders and followers has focused
on two areas: status and relationships. Both of these research topics
have been investigated using neuroimaging methodologies. Overall, this
research points to neuroimaging as being a useful tool to identify different
brain activity patterns among leaders and followers during their interactions, which is useful for understanding their mutual experiences.
DISCUSSION
Biosensor approaches can be applied to understanding leadership from a
leader-centric perspective, to examine leadership as an outcome, and to
explore leadership as characterized by the interactions between leaders and
followers. We summarized ways that biosensor approaches have informed
leadership research using methods such as neuroimaging (e.g., fMRI,
EEG), saliva, blood, and urine markers (e.g., cortisol, testosterone), and
genetics (e.g., twin studies examining the role of specific genes). Biosensor
approaches offer unique advantages, such as providing objective assessment to understand leadership processes that can be used in tandem with
other data sources (e.g., survey ratings from leaders or followers).
Notably, using biosensor approaches will not solve all methodological
and theoretical problems within leadership research. We are not advocating that researchers abandon survey or observation methods when studying leadership. Rather, they may find that they gain by supplementing the
study of leadership with other methods, including biosensor approaches.
Hiller, DeChurch, Murase, and Doty (2011) have made a similar argument. Their review of leadership research from 1985 to 2010 identified that
approximately 63 percent of leadership research used surveys to study the
phenomenon of interest, and they recommended the use of other methods
(e.g., observation, interviews, experimental manipulations) to understand
the rich process of leadership. Like the other methods discussed by Hiller et
al., biosensor approaches can be used in conjunction with survey research.
Together, they can help isolate psychological mechanisms and their physiological roots of leadership in order to improve our understanding of the
leader emergence process and factors contributing to leader effectiveness.
As with much of the work conducted in 70 years of leadership research,
the majority of studies discussed in this review focused on leader traits
and behaviors. While these studies have been informative, other perspectives of leadership research can benefit from using biosensor approaches.
For instance, applying biosensor approaches to studying leadership as an
SCHYNS_9781785367274_t.indd 164
10/11/2017 15:20
Biosensor approaches to studying leadership ­165
effect, we may be able to better elucidate the effects leaders have on followers’ cognitive, affective, and conative processes. Similarly, relationshiporiented leadership studies can utilize biosensor approaches to further
understand how high- versus low-quality relationships may develop over
time among the same leader and her or his different followers. Collecting
bioindicators may help discover the physiological mechanisms underlying
these effects.
In addition to encouraging leadership scholars to utilize a variety of
research methods and designs to examine leadership, Hiller et al. (2011)
also advocated for more research that focuses on the temporal dynamics of leadership. For example, they concluded that the majority of the
leadership studies used cross-sectional data (59 percent of studies), with
only 29 percent of the existing studies collecting longitudinal data (ibid.).
As a result, we have limited understanding of how time may be a factor
that can impact the leader emergence, leader–follower relationship formation, and leadership development. By studying leadership phenomena
longitudinally, concerted efforts can be made to align the frequency of
measurement with fluctuations in leadership phenomena as they develop
and change over time. Notably, the studies using biosensor methods in
this review were mostly short-term laboratory experiments or observational studies. In line with Hiller and colleagues’ (2011) recommendation,
studies using biosensor approaches should also expand beyond short-term
laboratory work and measure leadership longitudinally and in the field to
capture changes in the phenomenon over time.
An important complication when designing longitudinal biosensor
research is to not only consider the temporal characteristics of the leadership phenomenon in question, but also pay attention to how the focal
biomarkers of interest may change over time. For instance, both cortisol
and testosterone follow a diurnal rhythm, such that typical individuals
have the highest levels of cortisol and testosterone right after waking up,
levels of these hormones gradually decrease over the course of the day,
and are at their lowest right before bedtime (Adam & Kumari, 2009;
Van Anders, Goldey, & Bell, 2014). The between-person differences
in these hormones reflect both the within-person fluctuations associated with times of the day, and the effects associated with the external
factors – the leadership phenomenon in this case. Moreover, it is not
uncommon for different hormones to have different time-based changes
in response to the same external factor. For example, when individuals
experience acute stress, their cortisol level will show a clear increase
about 20 minutes after the exposure to a stressor (Dickerson & Kemeny,
2004), whereas their alpha-amylase (an enzyme commonly used to
mark sympathetic nervous system activation) will spike right after the
SCHYNS_9781785367274_t.indd 165
10/11/2017 15:20
166 Handbook of methods in leadership research
exposure. Thus, when applied to leadership research, biomarker measurements need to correspond with the leadership phenomenon in question, and the timing of these measures may be critical. It is commonly
recommended that researchers standardize the time of the day for data
collection in order to account for the diurnal fluctuations (Van Anders
et al., 2014). Moreover, researchers need to calibrate the change rates
for both the leadership phenomenon and the bioindicators to guide the
measurement frequency.
Finally, although biosensor methods offer benefits of objective measures, their adaptation may come with a cost to the physical and psychological fidelity of the experimental setting, and even external or face validity.
For example, studies utilizing fMRI require participants to stay still in isolation, and can only employ relatively simple tasks, such as tapping in sync
with a partner (e.g., Fairhurst et al., 2014), or responding to agreement
or disagreement with presented statements (e.g., Jack et al., 2013). These
experimental tasks may have limited physical and psychological fidelity in
simulating the actual leader behavior and leadership processes. Moreover,
it is unclear whether results from the laboratory can generalize to the realworld contexts in which leadership is embedded, thereby creating concerns
for the external validity of these findings. Finally, it is often assumed that
the physiological reactions assessed by biosensor approaches are consistent across leaders in different occupational contexts (e.g., military versus
religious versus business settings) and societal cultures (e.g., cultures high
in individualism versus collectivism; masculinity versus femininity; House,
Hanges, Javidan, Dorfman, & Gupta, 2004). To alleviate these concerns,
more flexible biosensor approaches could be implemented in field settings
to capture changes in biomarkers in real contexts. For instance, participants may wear GSR devices, heart rate monitors, and portable EEGs at
their workplaces where the leadership process unfolds. By observing
leadership phenomena in context and assessing their impact with real-time
biosensor measures, the findings will likely be more generalizable than
studies performed in a laboratory.
Overall, there remain numerous areas of leadership research that can be
further explored and clarified with biosensor approaches. As the field continues to grow, there is much promise in using biosensor methodologies
to inform current leadership theories and create a more objective, multimethod based approach for answering important research questions in the
leadership domain.
SCHYNS_9781785367274_t.indd 166
10/11/2017 15:20
Biosensor approaches to studying leadership ­167
REFERENCES
Adam, E.K., & Kumari, M. (2009). Assessing salivary cortisol in large-scale, epidemiological
research. Psychoneuroendocrinology, 34(10), 1423–1436.
Balthazard, P.A., Waldman, D.A., Thatcher, R.W., & Hannah, S.T. (2012). Differentiating
transformational and non-transformational leaders on the basis of neurological imaging.
The Leadership Quarterly, 23(2), 244–258.
Bass, B.M. (2008). The Bass handbook of leadership: Theory, research and managerial applications. New York: Simon and Schuster.
Beasley, M., Sabatinelli, D., & Obasi, E. (2012). Neuroimaging evidence for social rank
theory. Frontiers in Human Neuroscience, 6. doi: 10.3389/fnhum.2012.00123
Becker, W.J., & Cropanzano, R. (2010). Organizational neuroscience: The promise and
prospects of an emerging discipline. Journal of Organizational Behavior, 31(7), 1055–1059.
Becker, W.J., & Menges, J.I. (2013). Biological implicit measures in HRM and OB: A question of how not if. Human Resource Management Review, 23(3), 219–228.
Bendahan, S., Zehnder, C., Pralong, F.P., & Antonakis, J. (2015). Leader corruption
depends on power and testosterone. The Leadership Quarterly, 26(2), 101–122.
Boyatzis, R.E., Passarelli, A.M., Koenig, K., Lowe, M., Mathew, B., Stoller, J.K., & Phillips,
M. (2012). Examination of the neural substrates activated in memories of experiences with
resonant and dissonant leaders. The Leadership Quarterly, 23(2), 259–272.
Chaturvedi, S., Zyphur, M.J., Arvey, R D., Avolio, B.J., & Larsson, G. (2012). The heritability of emergent leadership: Age and gender as moderating factors. The Leadership
Quarterly, 23(2), 219–232.
Day, D.V. (2012). Leadership. In S.W.J. Kozlowski (Ed.), The Oxford handbook of organizational psychology. New York: Oxford University Press.
Day, D.V., & Antonakis, J. (2012). Leadership: Past, present, and future. In D.V. Day &
J. Antonakis (Eds.), The nature of leadership. Los Angeles, CA: Sage.
De Neve, J.E., Mikhaylov, S., Dawes, C.T., Christakis, N.A., & Fowler, J.H. (2013). Born
to lead? A twin design and genetic association study of leadership role occupancy. The
Leadership Quarterly, 24(1), 45–60.
Depue, R.A., Luciana, M., Arbisi, P., Collins, P., & Leon, A. (1994). Dopamine and the
structure of personality: Relation of agonist-induced dopamine activity to positive emotionality. Journal of Personality and Social Psychology, 67(3), 485–498.
Dickerson, S.S., & Kemeny, M.E. (2004). Acute stressors and cortisol responses: A theoretical
integration and synthesis of laboratory research. Psychological Bulletin, 130(3), 355–391.
Dinh, J.E. (2014). A neurocognitive perspective on dark leadership and employee deviance:
Influences of moral sensitivity and the self-concept. (Doctoral dissertation). Retrieved from
https://etd.ohiolink.edu/pg_10?0::NO:10:P10_ACCESSION_NUM:akron1390927195
Fairhurst, M.T., Janata, P., & Keller, P.E. (2014). Leading the follower: An fMRI investigation of dynamic cooperativity and leader–follower strategies in synchronization with an
adaptive virtual partner. Neuroimage, 84, 688–697.
Gilkey, R., Caceda, R., & Kilts, C. (2010). When emotional reasoning trumps IQ. Harvard
Business Review, 88(9), 20–21.
Graen, G.B., & Uhl-Bien, M. (1995). Relationship-based approach to leadership:
Development of leader–member exchange (LMX) theory of leadership over 25 years:
Applying a multi-level multi-domain perspective. The Leadership Quarterly, 6(2), 219–247.
Hamilton, L.D., Carré, J.M., Mehta, P.H., Olmstead, N., & Whitaker, J.D. (2015). Social
neuroendocrinology of status: A review and future directions. Adaptive Human Behavior
and Physiology, 1(2), 202–230.
Hannah, S.T., Balthazard, P.A., Waldman, D.A., Jennings, P.L., & Thatcher, R.W. (2013).
The psychological and neurological bases of leader self-complexity and effects on adaptive
decision-making. Journal of Applied Psychology, 98(3), 393–411.
Hansen, Å.M., Larsen, A.D., Rugulies, R., Garde, A.H., & Knudsen, L.E. (2009). A review
of the effect of the psychosocial working environment on physiological changes in blood
and urine. Basic & Clinical Pharmacology & Toxicology, 105(2), 73–83.
SCHYNS_9781785367274_t.indd 167
10/11/2017 15:20
168 Handbook of methods in leadership research
Hart, A.J., Whalen, P.J., Shin, L.M., McInerney, S.C., Fischer, H., & Rauch, S.L. (2000).
Differential response in the human amygdala to racial outgroup vs. ingroup face stimuli.
Neuroreport, 11(11), 2351–2355.
Harung, H.S., & Travis, F. (2012). Higher mind-brain development in successful leaders:
Testing a unified theory of performance. Cognitive Processing, 13(2), 171–181.
Harung, H., Travis, F., Blank, W., & Heaton, D. (2009). Higher development, brain integration, and excellence in leadership. Management Decision, 47(6), 872–894.
Hiller, N.J., DeChurch, L.A., Murase, T., & Doty, D. (2011). Searching for outcomes of
leadership: A 25-year review. Journal of Management, 37(4), 1137–1177.
House, R.J., Hanges, P.J., Javidan, M., Dorfman, P.W., & Gupta, V. (2004). Culture, leadership, and organizations: The GLOBE study of 62 societies. Thousand Oaks, CA: Sage.
Ilies, R., Gerhardt, M.W., & Le, H. (2004). Individual differences in leadership emergence:
Integrating meta-analytic findings and behavioral genetics estimates. International Journal
of Selection and Assessment, 12(3), 207–219.
Jack, A.I., Boyatzis, R E., Khawaja, M.S., Passarelli, A.M., & Leckie, R.L. (2013). Visioning
in the brain: An fMRI study of inspirational coaching and mentoring. Social Neuroscience,
8(4), 369–384.
Judge, T.A., Bono, J.E., Ilies, R., & Gerhardt, M. (2002). Personality and leadership: A
qualitative and quantitative review. Journal of Applied Psychology, 87(4), 765–780.
Konvalinka, I., Bauer, M., Stahlhut, C., Hansen, L.K., Roepstorff, A., & Frith, C.D. (2014).
Frontal alpha oscillations distinguish leaders from followers: Multivariate decoding of
mutually interacting brains. Neuroimage, 94, 79–88.
Knight, E.L., & Mehta, P.H. (2014). Hormones and hierarchies. In J.T. Cheng, J.L. Tracy,
& C. Anderson (Eds.), The psychology of social status (pp. 269–301). New York: Springer.
Lee, N., & Chamberlain, L. (2007). Neuroimaging and psychophysiological measurement in
organizational research: An agenda for research in organizational cognitive neuroscience.
Annals of the New York Academy of Sciences, 1118, 18–42.
Li, W.D., Arvey, R.D., Zhang, Z., & Song, Z. (2012). Do leadership role occupancy and
transformational leadership share the same genetic and environmental influences? The
Leadership Quarterly, 23(2), 233–243.
Li, W.D., Wang, N., Arvey, R.D., Soong, R., Saw, S.M., & Song, Z. (2015). A mixed blessing? Dual mediating mechanisms in the relationship between dopamine transporter gene
DAT1 and leadership role occupancy. The Leadership Quarterly, 26(5), 671–686.
Lindebaum, D., & Zundel, M. (2013). Not quite a revolution: Scrutinizing organizational
neuroscience in leadership studies. Human Relations, 66(6), 857–877.
Lowe, K.B., & Gardner, W.L. (2000). Ten years of the leadership quarterly: Contributions
and challenges for the future. The Leadership Quarterly, 11(4), 459–514.
Mayo Clinic. (2016a). Tests and procedures: EEG (electroencephalogram). Retrieved from
http://www.mayoclinic.org/tests-procedures/eeg/basics/definition/prc-20014093
Mayo Clinic. (2016b). Tests and procedures: MRI. Retrieved from http://www.mayoclinic.
org/tests-procedures/mri/home/ovc-20235698
McGinty, V.B., Lardeux, S., Taha, S.A., Kim, J.J., & Nicola, S.M. (2013). Invigoration of
reward seeking by cue and proximity encoding in the nucleus accumbens. Neuron, 78(5),
910–922.
Molenberghs, P., Prochilo, G., Steffens, N.K., Zacher, H., & Haslam, S.A. (2015). The
neuroscience of inspirational leadership: The importance of collective-oriented language
and shared group membership. Journal of Management. doi: 10.1177/0149206314565242
Nyberg, A., Alfredsson, L., Theorell, T., Westerlund, H., Vahtera, J., & Kivimäki, M.
(2009). Managerial leadership and ischaemic heart disease among employees: The Swedish
WOLF study. Occupational and Environmental Medicine, 66(1), 51–55.
Ochsner, K.N., & Lieberman, M.D. (2001). The emergence of social cognitive neuroscience.
American Psychologist, 56(9), 717–734.
Phelps, E.A., O’Connor, K.J., Cunningham, W.A., Funayama, E.S., Gatenby, J.C., Gore,
J.C., & Banaji, M. (2000). Performance on indirect measures of race evaluation predicts
amygdala activation. Journal of Cognitive Neuroscience, 12(5), 729–738.
SCHYNS_9781785367274_t.indd 168
10/11/2017 15:20
Biosensor approaches to studying leadership ­169
Podsakoff, P.M., & Organ, D.W. (1986). Self–reports in organizational research: Problems
and prospects. Journal of Management, 12(4), 531–544.
Schneirla, T.C. (1959). An evolutionary and developmental theory of biphasic processes
underlying approach and withdrawal. In R. Dienstbier (Ed.), Nebraska Symposium on
Motivation (pp. 27–58). Lincoln, NE: University of Nebraska Press.
Seemann, D.C. (1982). Leader style and anxiety level: Their relation to autonomic response.
Small Group Behavior, 13(2), 192–203.
Sellers, J.G. (2006). Testosterone and status seeking. (Doctoral dissertation). Retrieved from
https://repositories.lib.utexas.edu/handle/2152/2642
Sherman, G.D., Lee, J.J., Cuddy, A.J., Renshon, J., Oveis, C., Gross, J.J., & Lerner, J.S.
(2012). Leadership is associated with lower levels of stress. Proceedings of the National
Academy of Sciences, 109(44), 17903–17907.
Sherman, G.D., Lerner, J.S., Josephs, R.A., Renshon, J., & Gross, J.J. (2016). The interaction of testosterone and cortisol is associated with attained status in male executives.
Journal of Personality and Social Psychology, 110(6), 921–929.
Smith, T.W., & Jordan, K.D. (2015). Interpersonal motives and social-evaluative threat:
Effects of acceptance and status stressors on cardiovascular reactivity and salivary cortisol
response. Psychophysiology, 52(2), 269–276.
Summers, C.H., & Winberg, S. (2006). Interactions between the neural regulation of stress
and aggression. Journal of Experimental Biology, 209(23), 4581–4589.
UC San Diego School of Medicine. (2016). What is fMRI? Retrieved from http://fmri.ucsd.
edu/Research/whatisfmri.html
U.S. Food and Drug Administration. (2016). Radiation-emitting products – uses. Retrieved
from https://www.fda.gov/Radiation-EmittingProducts/RadiationEmittingProductsandPro​
cedures/MedicalImaging/MRI/ucm482763.htm
Van Anders, S.M., Goldey, K.L., & Bell, S.N. (2014). Measurement of testosterone in human
sexuality research: Methodological considerations. Archives of Sexual Behavior, 43(2),
231–250.
Waldman, D.A., Balthazard, P.A., & Peterson, S.J. (2011). Leadership and neuroscience:
Can we revolutionize the way that inspirational leaders are identified and developed? The
Academy of Management Perspectives, 25(1), 60–74.
Waldman, D.A., Wang, D., Stikic, M., Berka, C., Balthazard, P.A., Richardson, T.,. . .Maak,
T. (2013). Emergent leadership and team engagement: An application of neuroscience technology and methods. Academy of Management Proceedings. doi: 10.5465/AMBPP.2013.63
Wang, M.Y. (2012). Exploring potential R&D collaborators with complementary technologies: The case of biosensors. Technological Forecasting and Social Change, 79(5), 862–874.
Westendorp, R. (2012). The biology of leadership: The relation between leadership, psychopathy, and hormones. (Unpublished Master’s thesis). The Erasmus University, Rotterdam.
Zhang, Z., Ilies, R., & Arvey, R.D. (2009). Beyond genetic explanations for leadership: The
moderating role of the social environment. Organizational Behavior and Human Decision
Processes, 110(2), 118–128.
SCHYNS_9781785367274_t.indd 169
10/11/2017 15:20
SCHYNS_9781785367274_t.indd 170
10/11/2017 15:20
PART III
QUANTITATIVE
METHODS AND
ANALYTIC APPROACHES
SCHYNS_9781785367274_t.indd 171
10/11/2017 15:20
SCHYNS_9781785367274_t.indd 172
10/11/2017 15:20
8.
Mediation analysis in leadership studies:
new developments and perspectives
Rex B. Kline*
Leadership has been defined as a process of goal-directed influence that
mobilizes organizations toward desired goals; that is, leaders work with
people and structures within organizations to establish common meanings, goals, and outcomes (Riehl, 2012). This process is hypothetically
amenable to causal analysis if (1) relevant variables are specified and
measured, and (2) there is theory about direct or indirect effects among
these variables. Such variables could include attitudes or behaviors of
leaders, characteristics of subordinates or organizations, and outcomes.
Indirect causal effects involve the distinction between distal versus proximate causes such that a distal cause is expected to affect an outcome
through an intervening variable, or through a proximate cause. A related
concept is that of mediation, or the causal hypothesis that changes in one
variable lead to changes in another variable, or mediator, which in turn
lead to changes in outcome (Little, 2013). Indirect effects do not necessarily involve changes in the variables along the chain of effects from causes
to outcomes, but mediation always involves changes in these variables.
The distinction between indirect effect and mediation also depends on
the research design, a point elaborated later. Given a particular theory, a
presumed cause may have: (a) a direct effect on an outcome but no indirect effects, (b) no direct effect but at least one indirect effect, or (c) both
direct and indirect effects. Hallinger and Heck (1998) referred to the sets of
hypotheses just mentioned as, respectively, direct effects models, mediated
effects models, and mediated effects with antecedents models.
Distal causes in leadership processes are usually described as relatively
stable or enduring variables. Some examples include leader personality
characteristics, such as self-efficacy, emotional temperament, and social
potency, or the need for interpersonal power and desire to make an
impact on others (Baker, Larson, & Surapaneni, 2015). Another example
is quality of the leader–subordinate exchange relationship (e.g., LMX or
vertical dyad linkage theory), which can vary across dyads with the same
leader, but is generally viewed as relatively stable within a dyad (Wang,
Law, Hackett, Wang, & Chen, 2005). Intervening variables should be
amenable to influence by the leader, and thus are not generally understood
173
SCHYNS_9781785367274_t.indd 173
10/11/2017 15:20
174 Handbook of methods in leadership research
as persistent qualities. Examples of potentially malleable intervening variables include organizational structures, such as human resource policies,
or characteristics of subordinates, such as levels of organizational commitment or work assignments that can be modified through the actions
of a leader. The expectation that leaders affect outcomes at least in part
through indirect effects on other people, events, or organization factors
corresponds to the hypothesis of mediation (Hallinger & Heck, 1998).
There are many examples of analyses about direct or indirect effects of
leadership qualities on organizational outcomes. A few recent studies are
briefly described next:
●
Within a sample of employees and their respective supervisors from
●
●
about 20 different companies, Neves and Story (2015) reported
evidence that ethical leadership increases employees’ affective commitment to the organization, which in turn reduces deviance, or
intentional behavior that violates organizational norms or policies.
Also, the researchers found that the indirect effect just mentioned is
stronger when the leader’s personal reputation for moral standards
and effectiveness is greater. This result concerns moderation, or
a conditional causal effect. Moderation, or interaction, is not the
same thing as mediation, or an indirect effect, although the two are
sometimes confused – see Little (2013, Ch. 9) for a clear differentiation of the two.
Within a longitudinal sample of municipal workers, Tafvelin,
Armelius, and Westerberg (2011) analysed direct effects of
transformational leadership on employee affective well-being.
Transformational effects occur when leaders raise awareness of
moral values that encourage followers to sublimate their own personal goals for the collective good. These authors also reported evidence that transformational leadership affects well-being indirectly
through an organizational climate for innovation that is encouraged
by leaders.
Within a sample of young women enrolled in university, Baker et al.
(2015) reported evidence for the hypothesis that social potency indirectly affects the intention to engage in leadership activities, such as
in a group setting, through the intervening variables of leadership
self-efficacy and interest in leadership opportunities.
Mediation analysis has become “popular” in many research areas
besides the study of leadership. For example, Baron and Kenny (1986),
a classical work about the estimation of mediation effects among continuous variables with no interactions, has been cited over 50 000 times
SCHYNS_9781785367274_t.indd 174
10/11/2017 15:20
Mediation analysis in leadership studies ­
175
(Kenny, 2015). It is the single most widely cited article in the Journal of
Personality and Social Psychology (MacKinnon & Pirlott, 2015). There
are now hundreds, if not thousands, of published reports of mediation analyses. The reasons for this interest are not hard to understand.
Mediation analysis promises a look inside the “black box” of leadership or
other causal variables of interest in terms of how effects of such variables
operate. Pearl (2014) described mediation analysis as telling us how nature
works through the analysis of direct or indirect effects on outcomes at the
end of a causal chain. The directions and magnitudes of such effects may
have implications for human resource practices, organizational culture, or
other variables that connect leadership with outcomes.
But there is also evidence that the collective enthusiasm for mediation analysis has outstripped better judgment. Specifically, the typical
published mediation study may have so many flaws due to an inadequate
design, improper statistical analysis, or lack of attention to assumptions
that the results have little or no substantive value (Kline, 2015). Many of
these problems are addressed in a recent special issue of Basic and Applied
Social Psychology about problems with mediation analysis (Trafimow,
2015). These shortcomings are outlined next.
ASSUMPTIONS UNDERLYING MEDIATION
MODELS
Presented in Figure 8.1(a) is a basic “mediated effects with antecedents”
path model in which the presumed cause, X, has a direct effect on the
outcome, Y, and also an indirect effect on Y through a presumed mediator, M. The arrows at 45° angles that point to M and Y represent error
(unexplained) variance and also designate these two variables as dependent (endogenous) within this particular model. The specification that the
coefficient for the direct effect of X on Y equals zero (c 5 0) would change
Figure 8.1(a) to a mediated effects model with the sole causal pathway:
X→M→Y,
which indicates that the effect of X on Y is purely indirect, and through
a single mediator. For now we suppose that X, M, and Y are continuous
variables.
Implied in Figure 8.1(a) are several assumptions that are rarely acknowledged in the typical mediation study. Some of these assumptions are
also untestable in that the data can tell us nothing about whether the
assumption is plausible or not plausible (Bullock, Green, & Ha, 2010).
SCHYNS_9781785367274_t.indd 175
10/11/2017 15:20
176 Handbook of methods in leadership research
X
a
c
M
M
X
M
b
Y
(a) Basic mediation
Note:
X
Y
Y
(b) Reciprocal effects
(c) Correlated errors
X, cause; M, mediator; Y, outcome. All three models are equivalent.
Figure 8.1
( a) Basic mediation with antecedents model; (b) mediation
model with reciprocal effects; (c) correlated errors model with
no mediation
An example is the assumption of modularity, which means that the
causal process consists of components that are potentially isolatable,
and thus can be analysed as separate entities (e.g., the trivariate model in
the figure). A causal process that is organic or holistic is inseparable into
parts, and thus is not amenable to standard mediation analysis (Knight &
Winship, 2013). There is no empirical check for modularity, so it must be
assumed, but authors of mediation studies rarely even mention the issue.
Other assumptions implied in Figure 8.1(a) that generally cannot be tested
with the data are listed next.
It is assumed that all directionality specifications are correct, including
X→M, X → Y, and M → Y in the figure. If the specifications just stated
are incorrect, then the results may have no meaningful interpretation.
Some researchers incorrectly believe that directionality specifications are
actually tested in mediation analysis, but the truth is that such specifications are simply assumed to be true at the beginning of the empirical
analysis. The reason for concern with directionality has to do with whether
equivalent models with alternative causal directions (considered momentarily) are equally plausible, but directionality is generally assumed, not
tested, in mediation analysis.
It is also assumed for Figure 8.1(a) that there are no unmeasured
common causes, or confounders, for any pair of variables among X, M,
or Y. This includes the assumption of independent errors for dependent
variables M and Y. It is also assumed that no omitted confounder of the
association between M and Y is caused by X. These are strict requirements. Estimation of mediation using data from strong research designs
that make these assumptions more tenable helps. An example of a strong
design that supports directionality assumptions are longitudinal designs,
SCHYNS_9781785367274_t.indd 176
10/11/2017 15:20
Mediation analysis in leadership studies ­
177
in which presumed causes are measured before their outcomes. Another
example is an experimental design where direct manipulation of causal
variables helps to eliminate explanations due to confounders. But crosssectional (non-experimental) designs where all variables are concurrently
measured have no inherent support for causal inference, and thus causal
analysis in such designs may be intractable.
There are additional assumptions accompanying the models presented
in this chapter, but satisfying them is more under the control of the
researcher. For example, the model in Figure 8.1(a) implies that scores on
variable X are perfectly reliable, that is, rXX 5 1.0, where rXX is a reliability
coefficient. This is because independent (exogenous) variables in path
models have no error terms; thus, there is no allowance for imperfect reliability (i.e., rXX < 1.0) in exogenous variables. Instead, measurement error
in X tends to show up “downstream” in the model, or in values of path
coefficients for variables affected by X (e.g., M and Y in Figure 8.1(a)) or
in their error terms. If X were the sole cause (e.g., M in the figure), then
measurement error in X tends to reduce the absolute value of the coefficient. But measurement error in multiple causes of the same outcome (e.g.,
Y in the figure) can bias coefficients by either increasing or decreasing
their absolute values. If X were an experimental variable that represented
the random assignment of cases to conditions, the assumption of no
measurement error would be plausible, but this is not generally true if X
is a measured (non-experimental) variable, such as an attribute assessed
with a self-report questionnaire. However, because measurement error in
endogenous variables is manifested in their error terms, it is not assumed
for M and Y in the figure that:
rMM 5 rYY 5 1.0
There are ways to explicitly control for measurement error in each and
every observed variable in a path model (Hayduk & Littvay, 2012), but
such methods are rarely used in mediation studies. Cole and Preacher
(2014) outline the negative consequences of ignoring error in path analysis.
IMPORTANCE OF RESEARCH DESIGN
Most mediation studies are based on cross-sectional designs, but such
designs have no formal elements that directly support casual inference.
One reason is the absence of time precedence. For example, if variables
X, M, and Y in Figure 8.1(a) are all measured at the same occasion, there
may be no way to establish which of two variables, a presumed cause and
SCHYNS_9781785367274_t.indd 177
10/11/2017 15:20
178 Handbook of methods in leadership research
its presumed effect, occurred first. Therefore, the only grounds for causal
inference in cross-sectional designs is assumption, one supported by a
clear, convincing rationale for specifying particular directions of causal
effects (e.g., X → Y in Figure 8.1(a)). Making directionality specifications
with any degree of confidence in cross-sectional designs depends a great
deal on the researcher to rule out competing explanations of the association between X and Y. Without stating such a rationale, little confidence
is warranted about interpreting path coefficients as evidence for direct
causal effects in cross-sectional designs.
Tate (2015) reminded us that mediation involves time-ordered relations among variables such that the mediator must always intervene in
time between the cause and outcome (respectively, M, X, and Y in Figure
8.1(a)). It also requires a potentially changeable mediator; that is, not just
any variable can be specified as a mediator. This explains why the specification of personality trait variables as mediators is generally inappropriate. This is because stable traits cannot mediate a cause–outcome relation.
The same concern applies to other kinds of individual difference variables
considered to be stable characteristics, such as self-esteem, self-efficacy,
and advancement (promotion) versus security (prevention) motivational
orientations, among others. But it may be reasonable to specify state
variables as mediators, if such variables are hypothesized as outcomes of
prior causes and thus are conceptually malleable. States are temporary
behaviors, perceptions, or emotions that depend on a particular context
or are affected by enduring traits at a particular time. An example is the
distinction between trait anxiety as a general propensity to feel apprehensive versus state anxiety as the degree of unease at a particular moment.
State anxiety is generally seen as malleable, but trait anxiety may be more
enduring.
The Problem of Equivalent Models
With no actual or conceptual time ordering in a cross-sectional design,
it is possible to generate equivalent versions of a path model that explain
the data just as well as the original model. Consider Figure 8.1(a). Any
­rearrangement of the paths that does not result in a causal loop (i.e., the
model remains recursive) would fit the same data just as well as the model of
Figure 8.1(a). For instance, the alternative model with the paths listed next:
X → Y, X → M, but Y → M
is equivalent to Figure 8.1(a) in that both models would explain the same
data equally well. Altogether there are a total of six equivalent versions of
SCHYNS_9781785367274_t.indd 178
10/11/2017 15:20
Mediation analysis in leadership studies ­
179
Figure 8.1(a) (including this figure), where each equivalent model corresponds to a different causal ordering among X, M, and Y. In this example,
all of these models would have perfect fit because their degrees of freedom
are zero, but the point is that analysis cannot determine which (if any) of
these equivalent models is correct.
The problem of equivalent models is even worse than just described, if
there is no time ordering of any kind. For example, Figure 8.1(b) depicts
a non-recursive model with a causal (direct feedback) loop between
variables M and Y. In this model, M and Y each mediate the effects of the
independent variable on each other, thus it can be described as a reciprocal effects model. In order to be identified, this model must include the
assumption that the two direct effects that make up the causal loop are
equal. There are two equivalent versions of Figure 8.1(b), for example, the
model with the specifications:
Y → M, Y → X, but M → X and X → M,
where the last two direct effects are constrained to be equal. Including
Figure 8.1(b), there are three equivalent versions of reciprocal effects
models, and each of these variations is also equivalent to each of the six
equivalent versions of Figure 8.1(a).
Finally, Figure 8.1(c) is recursive but not does specify mediation because
there is no direct link between M and Y. Instead, it features a correlation
of the error for variable M with the error for variable Y. Nevertheless,
Figure 8.1(c) and its two equivalent versions, such as the non-mediational
model:
Y → M, Y → X, but M → X,
where the symbol ↔ designates correlated errors, are all equivalent to
the six equivalent versions of Figures 8.1(a) and the three equivalent
versions of Figure 8.1(b). Thus, there are a total of 12 equivalent versions of any model taking the form shown in Figure 8.1(a), including
the original model. Without some strong evidence that can eliminate the
causal directions implied by these equivalent models, they are equally
plausible explanations of the data that fail to support the original conception of the mediated effect. Unfortunately, too many researchers fail
even to mention the issue of equivalent models when analysing mediation
in cross-sectional designs, much less garner evidence to rule out those
models.
SCHYNS_9781785367274_t.indd 179
10/11/2017 15:20
180 Handbook of methods in leadership research
Conceptual Bases for Inferring Causal Direction
Sometimes alternative directionalities in cross-sectional designs can be
ruled out by the nature of the variables. Suppose that variable X in Figure
8.1(a) is gender and that both M and Y are individual difference variables. It would be illogical to assume that M or Y could change a person’s
gender, so the alternative specifications:
M → X or Y → X
would be indefensible. Another basis for specifying directionality in crosssectional designs is the Hyman–Tate conceptual timing criterion (Hyman,
1955; Tate, 2015), which requires a theoretical time-ordering of the cause,
mediator, and outcome. From this perspective, variables are not required
to have a strict temporal order in their measurement – cause first, mediator second, and outcome last – if there is a strong rationale about causal
order. In leadership studies, for example, it may be reasonable to assume
that managers have stronger effects on subordinates than the reverse.
Baker et al. (2015) argued that social potency is a general personality trait
that could theoretically affect leadership qualities, and thus they specified
social potency as causal and leadership variables as mediators even though
all variables were concurrently measured. If the rationale for a conceptual
timing order is not convincing, though, causal inference is unwarranted.
For instance, in some theoretical models, leadership is defined conditional
on outcomes; that is, people in leadership positions who have performed
well in in the past may be viewed as more leader-like because the successful performance is attributed to them (and they also may be more likely
to perform well in the future). Thus, if X is a rating of leadership and Y is
outcome, it would be difficult to make a strong argument for causal order
in a cross-sectional design (R. Hall, personal communication, September
12, 2016).
Designs that Incorporate Time Lags
Other types of research designs feature actual, not conceptual, time precedence. A measurement-of-mediation design features random assignment
of cases to the levels of a causal variable, for example, a two-group design
in which X 5 0 for control and X 5 1 for an intervention intended to boost
leadership skills. In this design, the mediator M is an individual difference
variable that is measured, not manipulated, at a later time but before the
outcome Y is assessed (Bullock et al., 2010). Selecting the appropriate
measurement schedule, or time lags between measurement occasions, is
SCHYNS_9781785367274_t.indd 180
10/11/2017 15:20
Mediation analysis in leadership studies ­
181
critical. For example, measuring M too soon may not give the intervention enough time to have its effect, but measuring M too late can miss
temporary effects that have dissipated. An advantage is that because X is
an experimental variable, over replication samples it will be isolated from
confounders that also affect M or Y. But because M is not manipulated,
it may be plausible that M and Y share at least one unmeasured cause.
In this case, the absence of randomization on M means that other causes
of Y, or confounders, cannot be ruled out; thus, the measurement-ofmediation design does not completely eliminate potential spurious causal
inferences.
As a possible solution to this problem, Antonakis, Bendahan, Jacquart,
and Lalive (2010) described the use of instrumental variables, or instruments, in leadership studies as a way to control for omitted causes of M
and Y. An instrument is a variable that should have a direct effect on M,
no direct effect on Y, and an indirect effect on Y only through M. The
mediator M is regressed on the instrument, and the predicted mediator in
this analysis replaces M in the analysis where the outcome Y is regressed
on both M and X for Figure 8.1(a). This tactic removes the influence of an
unmeasured common cause of M and Y from the analysis just described.
MacKinnon and Pirlott (2015) describe additional statistical methods for
enhancing the causal interpretation of the effect of M on Y in mediation
analysis. See also Bullock et al.’s (2010) description of manipulation-ofmediation designs, where the mediator is also an experimental variable.
Challenges of manipulating mediators are considerable, but successfully
doing so for both variables X and M isolates confounders for any pair of
variables among X, M, and Y, greatly strengthening the support for the
proposed causal direction.
Longitudinal designs also offer time precedence. There are specific
longitudinal designs for mediation analysis, such as those described by
Cole and Maxwell (2003), Maxwell and Cole (2007), Selig and Preacher
(2009), Little (2013), and others. Such designs feature causes, mediators, and outcomes, each measured at multiple points in time. Also, the
path models associated with such designs allow no direct effects between
variables measured at the same occasion. The latter specification avoids
violating the requirement for time ordering. Thus, longitudinal designs
for mediation analyses guarantee that estimates of X → M relations are
only made from causes that are measured before mediators and that, in
turn, estimates of M → Y relations are only made from mediators that are
measured before outcomes. These specifications are demonstrated next
with an example.
Tafvelin et al. (2011) measured transformational leadership, (X), climate
for innovation (M), and affective well-being (Y) on two occasions within
SCHYNS_9781785367274_t.indd 181
10/11/2017 15:20
182 Handbook of methods in leadership research
X1
X2
X
X
M
a
M1
M2
M
b
Y1
Y2
(a) Tafvelin et al.
W
Y
(b) Neves & Story
Y
(c) De Vries et al.
Notes:
a. Subscripts indicate time of measurement; X, transformational leadership; M, climate
for innovation; Y, affective well-being.
b. X, ethical leadership; W, reputation for performance; M, affective commitment; Y,
organizational deviance.
c. X, leadership support; M, need for leadership; Y, job satisfaction.
Black dots on the diagram represent interactive effects.
Figure 8.2
( a) Tafvelin et al. (2011) model;a (b) Neves and Story
(2015) conceptual model with conditional mediation;b (c)
example for CMA approach based on variables analysed by
De Vries et al. (2002)c
a 12-month interval in a large sample of municipal workers. Their model
is presented in Figure 8.2(a), where subscripts represent the two measurement occasions. Because there is no direct effect from X1 to Y2, the hypothesis of a fully mediated effect of X on Y is represented in Figure 8.2(a).
Adding the direct effect X1 → Y2 to the figure would represent the hypothesis of partial mediation, where X has both direct and indirect effects on
Y. The model in Figure 8.2(a) features the two cross-lagged direct effects
listed next:
X1 → M2 and M1 → Y2,
where coefficients a and b estimate these two direct effects. Coefficients a
and b are proper longitudinal estimates of the two component mediating
paths because there is time precedence in the measurement of each pair
of variables connected with a direct effect. Also, the value of coefficient
a controls for the prior effect of the mediator on itself (M1 → M2), and
the value of coefficient b does the same thing for the outcome (Y1 → Y2).
Figure 8.2(a) also features the four cross-sectional paths listed next:
SCHYNS_9781785367274_t.indd 182
10/11/2017 15:20
Mediation analysis in leadership studies ­
183
X1 → M1, M1 → Y1, X2 → M2, and M2 → Y2
Coefficients associated with the paths just listed are not proper estimates of
mediation because there is no time precedence among variables measured
on the same occasion. Figure 8.2(a) but without the four cross-sectional
paths is similar to a half-longitudinal design, where the mediator and
outcome (M, Y) are each measured on two occasions, but the cause (X) is
measured only at time 1. At time 2, the disturbances of the mediator and
outcome are assumed to covary, or M2 ↔ Y2. In a full-longitudinal design,
X, M, and Y are each measured at three different occasions. Mediation in
this design is estimated from the coefficients from paths of the single contiguous pathway through which the cause can affect the outcome through
the mediator over time, or:
X1 → M2 → Y3,
where the subscripts indicate the measurement occasion. There are also
proxy estimators of mediated effects that are available from coefficients
for noncontiguous pathways, such as:
X2 → M3 and M2 → Y3
See Little (2013) for more information about full-longitudinal designs for
mediation analysis.
To summarize, a proper design for estimating mediation offers
either actual temporal precedence or conceptual temporal precedence
with an ironclad rationale. This requirement is consistent with Little’s
(2013) definition of mediation given at the beginning of this chapter
that emphasizes the transmission of changes from cause to mediator
and then to outcome. It also explains the difference between the terms
“indirect effect” versus “mediation.” Specifically, mediation always
involves indirect effects, but not all indirect effects automatically signal
mediation. This is especially true in cross-sectional designs with no
time precedence of any kind. Such designs do not allow or control for
changes in any variable, cause, mediator, or outcome. Thus, without
a proper research design, use of the term “mediation” may be unwarranted. The term “indirect effect” would still apply, but mediation refers
to a strong causal hypothesis, and suitable designs are needed to test
strong hypotheses.
SCHYNS_9781785367274_t.indd 183
10/11/2017 15:20
184 Handbook of methods in leadership research
SIGNIFICANCE TESTING
When all variables are continuous and there are no interaction effects,
mediation is generally estimated as the product of the coefficients from the
direct effects that make up the indirect causal pathway. For example, this
estimator for both Figures 8.1(a) and 8.2(a) corresponds to the quantity
a × b, or ab. Given hypothetical values of the unstandardized coefficients
for Figure 8.1(a) listed next, a 5 1.50, b 5 0.50, and c 5 2.00, we can say
that the unstandardized indirect effect of X on Y equals ab 5 1.50(0.50)
5 0.75. The unstandardized direct effect of X on Y, or 2.00, is estimated
controlling for the effect of M on the Y (i.e., the other cause), so coefficient
c is analogous to a partial regression coefficient. The unstandardized total
causal effect of X is the sum of its direct effect and indirect effect. In other
words, for every 1-point increase in the original metric of X, we expect (1)
a 2-point increase in the original metric of Y through the direct effect, and
(2) another 0.75 increase through the indirect effect that involves M. The
sum of the direct and indirect effects is thus 2.75, which is the total effect
of X on Y. (The interpretations just stated require all the assumptions for
Figure 8.1(a) discussed earlier.)
Unstandardized indirect effects can be tested for statistical significance.
However, you should know that the outcomes of significance testing, or
p values, for indirect effects may be untrustworthy due to implausible
assumptions or inadequate sample sizes. Thus, statistical significance is
not a scientific gold standard in mediation analysis (or in any other kind
of statistical analysis). Sobel (1982) developed a hand-calculable significant test for unstandardized indirect effects that involve just three variables (i.e., there is a single mediator). In large random samples, the ratio
ab/SEab, where the denominator is an approximate standard error in the
Sobel method, is interpreted as a z (normal deviate) test of the unstandardized indirect effect. The Sobel test assumes normality of the sampling
distribution for coefficient ab, but this is a dubious assumption as product
estimators do not generally follow normal distributions. The Sobel test
also assumes random sampling, but most samples in mediation studies are
convenience (ad hoc) samples, so the term SEab probably does not measure
solely error variation in convenience samples. In samples that are neither
large nor representative, p values generated by the Sobel test can be quite
inaccurate.
More modern significance tests for the indirect effects from models
with either single or multiple mediators are based on a technique called
non-parametric bootstrapping (Preacher & Hayes, 2004, 2008). Briefly,
the computer randomly samples with replacement from the researcher’s
data file a large number (typically, 1000 or more) of generated data sets,
SCHYNS_9781785367274_t.indd 184
10/11/2017 15:20
Mediation analysis in leadership studies ­
185
each with the same number of cases as in the original data file. Next, the
computer derives the unstandardized estimator of the indirect effect in all
generated samples and then creates an empirical sampling distribution
given all of these results. Finally, the computer locates within the empirical
sampling distribution the values of the products estimator that correspond
to the 2.5 and 97.5 percentiles. Between these two points are 95 percent of
the results from the generated samples. If the value 0 does not fall with the
interval bounded by the 2.5 and 97.5 percentiles, then the null hypothesis
that the population product estimator equals 0 is rejected at the 0.05 level
of statistical significance; otherwise, the null hypothesis is retained.
Non-parametric bootstrapping makes no assumptions about the shape
of the population distribution of product estimators, other than that the
sample distribution matches the population distribution. Outcomes of
bootstrapped significance tests of unstandardized indirect effects may be
more accurate than those from the Sobel test in large and representative
samples, and sample size requirements for bootstrapped tests of indirect
effects are not as demanding compared with the Sobel test. But even
bootstrapped results can be very inaccurate in small or unrepresentative
samples. Such tests still assume random sampling, which almost never
happens in mediation studies. For more information about bootstrap
methods for testing the statistical significance of mediation effects, see
Mallinckrodt, Abraham, Wei, and Russell (2006).
What is more important than statistical significance in mediation studies
is the concept of substantive significance, which involves the evaluation of
results in terms of their practical, theoretical, or clinical significance in a
particular research context (Kline, 2013), not just whether results are statistically significant or not significant. Reporting and interpreting effect
sizes in mediation studies is one way to address substantive significance.
There are ways to measure the relative sizes of direct versus indirect effects
in standardized metrics that are directly comparable across different
studies (Lau & Cheung, 2012; Preacher & Kelley, 2011), when it is appropriate to analyse standardized effect sizes. That increasing numbers of
journals, including many in medicine, now require the reporting of effect
sizes is another motivation. The International Committee of Medical
Journal Editors (2015) puts it like this: ‘Avoid relying solely on statistical hypothesis testing, such as p values, which fail to convey important
information about effect size and precision of estimates’ (p. 14). There are
also methods in mediation analysis for interval estimation, which allow
the reporting of indirect effects with confidence intervals (Cheung, 2009).
Indirect effects with narrower confidence intervals are more precise than
indirect effects sizes with wider intervals.
Thoemmes (2015) described an indefensible use of significance testing
SCHYNS_9781785367274_t.indd 185
10/11/2017 15:20
186 Handbook of methods in leadership research
in mediation analysis that involves equivalent models. Suppose that the
product coefficient for the indirect effect of X on Y through M in Figure
8.1(a) is statistically significant. Next, the researcher reverses the arrow
between M and Y and thus specifies an equivalent model where the mediator is now variable Y and the outcome variable is M. For this respecified
model, the researcher finds that the product coefficient for the indirect
effect of X on M through Y is not significant. The researcher concludes that
these results support the original hypothesis that M mediates the effect of
X on Y but does not support the alternative hypothesis that Y mediates the
effect of X on M, but this conclusion is faulty. This is because the statistical significance of the indirect effect (or any other effect) can never be used
to infer whether one model should be preferred over another, if the two
models are equivalent. The only way to choose between two equivalent
models is through assumptions that are satisfied through research design
or predictions based on theory. Statistical analysis including significance
testing does not distinguish equivalent models.
ANALYSING MEDIATION AND MODERATION
TOGETHER
The classical product method for estimating indirect effects as products
of the coefficients for the direct effects among continuous variables makes
the strong assumption of no interactions. In mediation analysis, this
means that (1) the cause X and the mediator M do not interact; that is,
the effect of X on outcome Y does not depend on the level of M just as the
direct effect of M on Y does not vary across the levels of X. The assumption of no interaction also means that (2) there is no external variable with
which X interacts to render its direct effect on either M or Y conditional,
and (3) there is no external variable with which M interacts such that its
direct effect on Y is rendered conditional.
There are times when the assumption of no interactions is too restrictive. Fortunately, there are ways to represent both mediation and moderation (interaction) together in the same model. Doing so involves the
estimation of conditional indirect effects. Two modern analytical methods
for estimating conditional indirect effects, conditional process modeling
(CPM) (Hayes, 2013) – also called conditional process analysis – and
causal mediation analysis (CMA) (Pearl, 2014), are briefly described
next. Of the two methods, CPM has been applied mainly to models with
continuous mediators and outcomes, and when linear causal relations are
assumed. The CPM approach is becoming more familiar in disciplines
such as psychology and education. The CMA method is more general in
SCHYNS_9781785367274_t.indd 186
10/11/2017 15:20
Mediation analysis in leadership studies ­
187
that it can be applied to the analysis of mediators or outcomes that are categorical or continuous, and also when non-linear relations are assumed.
For example, the CMA method can be extended to linear-, logistic-, loglinear, or Poisson-type regression analyses (e.g., odds ratios or risk ratios
analysed for categorical outcomes). The CMA method also offers a constant definition of indirect effects across the kinds of variables and models
just mentioned. It is better known in disciplines such as epidemiology, but
that is changing (Vanderweele, 2015).
Two key concepts in CPM are those of (1) moderated mediation, also
known as a conditional indirect effect; and (2) mediated moderation,
where an interaction indirectly affects an outcome through a mediator.
Both types of conditional causal effects just mentioned are represented in
Figure 8.2(b) where the interaction between causal variables X and W is
represented by the symbol for a closed circle. This figure depicts the conceptual model described by Neves and Story (2015, p. 169). It specifies that
ethical leadership (X) interacts with leader reputation for performance
(W) to have a joint effect on subordinate affective commitment to the
organization, the mediator (M). This interaction hypothesis says that the
direct effect of X on M depends on W, and also that the direct effect of W
on M depends on X.
Both conditional direct effects just described for Figure 8.2(b) are the
first stages of the two indirect pathways listed next:
X → M → Y and W → M → Y
where the outcome Y is organizational deviance and variables involved in
conditional (i.e., moderated) direct effects are shown in bold type. Because
the first stage of each indirect path just listed is conditional, each indirect
effect is also conditional. This particular kind of moderated mediation
is also called first-stage moderation because the first path of an indirect
effect depends on an external variable (Edwards & Lambert, 2007). There
are other patterns, such as second-stage moderation where just the second
path of an indirect effect depends on an external variable, but all kinds of
moderated mediation refer to conditional indirect effects. It is also true in
Figure 8.2(b) that the joint effect of X and W on Y is specified as purely
indirect through M, which corresponds to mediated moderation. That is,
the interactive effect of X and W on Y is transmitted entirely through M.
Although the actual model analysed by Neves and Story (2015), had additional variables, such as age, gender, and education, both mediation and
moderation were analysed in their data set.
The alternative approach – that is, the method of CMA – assumes that
there is an interaction between the cause X and mediator M, and thus this
SCHYNS_9781785367274_t.indd 187
10/11/2017 15:20
188 Handbook of methods in leadership research
method routinely estimates conditional effects. Direct and indirect effects in
CMA are defined from a counterfactual perspective. A counterfactual does
not express what has happened but what could hypothetically happen under
different circumstances. The statement, “I would not have been late, if I had
set my alarm earlier,” is an example of a counterfactual. In real life, counterfactuals correspond to statement such as, “If I had only. . .” or “What if. . .?”
To use the CMA approach to estimate mediation, several effects must
be defined. When a cause X and a mediator M interact, the controlled
direct effect of X is defined as the average difference on Y, given a 1-point
change in X, if M were controlled as having the same level for all cases in
the population. There is a different value of the controlled direct effect for
each level of the mediator, so for a continuous M there are infinitely many
controlled direct effects.
The natural direct effect of X is the average difference in Y, if X were
to change by 1 point but M is kept to the level that it would have taken
without a 1-point change in X. Unlike the case for a controlled direct effect,
the level of M is not fixed to the same constant for all cases. Instead, the
mediator is allowed to vary, but only over values that would be naturally
observed if there was no change in X. If there is no interaction between
cause and mediator, then the controlled direct effect and natural direct
effect of X are equal for continuous variables in a linear model; otherwise,
estimates of natural direct effects and controlled direct effects may differ.
The natural indirect effect of X estimates the average change in Y as
the mediator M changes from values that would be observed assuming
no change in X, to the values that M would reach if X changes by 1 point.
That is, the effect if the outcome is influenced by the cause operating solely
through its influence on the mediator. The total causal effect of X is the
sum of its natural direct effect and the natural indirect effect on Y. In contrast, the controlled direct effect does not have a simple additive relation
with either the natural direct effect or natural indirect effect, so it is not
part of an effect decomposition in CMA.
When all variables are continuous in a linear model, calculating estimates of the controlled direct effect (CDE), natural direct effect (NDE),
natural indirect effect (NDE), and the total effect (TE) is just a matter
of rearranging terms in standard regression equations (Petersen, Sinisi,
& Van der Laan, 2006). To illustrate this point, the following numerical example is presented, based on some of the variables analysed by De
Vries, Roe, and Taillieu (2002) in a large sample of employees. The causal
variable X is leadership support, the mediator M is need for leadership,
and the outcome Y is job satisfaction. The hypotheses are that: (1) need
for leadership mediates some of the effect of leadership support on job
satisfaction, and (2) leadership support and need for leadership interact in
SCHYNS_9781785367274_t.indd 188
10/11/2017 15:20
Mediation analysis in leadership studies ­
189
their effect on job satisfaction. The hypotheses just stated are represented
in Figure 8.2(c).
The unstandardized regression equations for this example are listed next
in symbolic form:
M̂ 5 b0 + b1X(8.1)
Yˆ 5 q0 + q1 X + q2 M + q3 XM,
where b0 and q0 are the intercepts for, respectively, the regressions of M
on X and the regression of Y on X, M, and the product term XM, which
represents the interactive effect in the technique of moderated multiple
regression (Cohen, Cohen, West, & Aiken, 2003, Ch. 6). The coefficient
for predicting M from X is b1, and q1–q3 are the coefficients for, respectively, X, M, and XM when predicting Y.
Given Equation 8.1, the CDE, NDE, and NIE of X on Y can be
expressed as follows:
CDE 5 q1 + q3 m(8.2)
NDE 5 q1 + q3 b0
NIE 5 (q2 + q3) b1
In Equation 8.2, note that the CDE is defined for a particular constant
value of the mediator (M 5 m), and the NIE is defined at the predicted level
of the mediator, or b0, when X 5 MX, if X is a mean-deviated, or centered,
predictor (i.e., the scores on X are centered). Also note that if there is no
interaction between X and M, then q3 5 0 (see Equation 8.1). In this case:
CDE 5 NDE 5 q1,
which is the unconditional linear effect of X on Y (see Equation 8.2), and
the NIE equals the classical product estimator of the indirect effect, or
b1q2. Thus, a model with no interaction between the independent variable
and the mediator can be seen as a specific instance of a more general case
that includes the interaction.
Based on the summary statistics for variables X, M, X, and Y reported
by De Vries et al. (2002, p. 126), I conducted two linear regression analyses
in SPSS using the syntax listed in the Appendix, which also contains the
summary data for this example. Note that scores on variable X are centered.
The two unstandardized regression equations for this example are listed next:
M̂ 5 2.630 + 0.072 X
Ŷ 5 4.032 + 0.290 X − 0.062 M + 0.020 XM
SCHYNS_9781785367274_t.indd 189
(8.3)
10/11/2017 15:20
190 Handbook of methods in leadership research
In words, the predicted value of M, need for leadership, equals 2.630, if X,
leadership support, equals its mean (i.e., it does not vary). For each 1-point
increase in X, variable M is expected to increase by 0.072 points. The predicted score on Y, job satisfaction, is 4.032, if X equals its sample mean
and M 5 0. For each 1-point increase in X when M 5 0, job satisfaction is
predicted to increase by 0.290 points, controlling for M and the interaction
between X and M. Each 1-point increase in M predicts a 0.062 decrease in
Y when X equals the sample mean and controlling for all other effects. For
every 1-point increase in X, the slope of regression line for predicting Y
from M increases by 0.020 (and vice versa), controlling for all other effects.
Equation 8.3 implies that:
b0 5 2.630 and b1 5 0.072
q0 5 4.032, q1 5 0.290, q2 5 −0.062, and q3 5 0.020
Continuing with the same example, the direct effect of X, leadership
support, on Y, job satisfaction at a given level of need for leadership, M 5
m, is estimated as:
CDE 5 0.290 + 0.020 m
The researcher can select a particular value of m for the mediator and then
estimate the CDE by substituting this value in the formula just listed. The
direct effect of X on Y estimated at the level of M that would have been
observed if no change in X had occurred is estimated as:
NDE 5 0.290 + 0.020 (2.630) 5 0.343
The indirect effect of X on Y allowing M to vary as it would if no change
in X had occurred is estimated as
NIE 5 (−0.062 + 0.020) 0.072 5 −0.003
The total effect of X on Y is the sum of its natural direct and indirect
effects, or:
TE 5 0.343 − 0.003 5 0.340
Thus, the predicted score on job satisfaction (Y) is expected to increase by
0.34 points for every 1-point increase in leadership support (X) through
both its natural direct effect (0.343) and also its natural indirect effect
(−0.003) through need for leadership (M), assuming that X and M inter-
SCHYNS_9781785367274_t.indd 190
10/11/2017 15:20
Mediation analysis in leadership studies ­
191
act. The magnitude of the NIE for these data seems relatively small, but
it is a component of the total effect of leadership support on job satisfaction and thus should be tallied. Valeri and Vanderweele (2013) describe
macros for SPSS and SAS/STAT for the CMA method that allows
covariates, such as variables that control for treatment history. These
computer tools automatically calculate the CDE, NDE, NIE, and TE and
test each effect just listed for statistical significance, but do not forget the
cautions about significance testing in mediation analysis discussed earlier
in this chapter.
Thinking about direct or indirect effects from a counterfactual perspective is not immediately intuitive, but the CMA method offers a consistent
way of defining and estimating these effects in the presence of interaction for continuous or categorical mediators or outcomes. There are also
some freely available macros for CMA written for widely used computer
tools for general statistical analyses, including R, SPSS, SAS/STAT,
and Stata (Hicks & Tingley, 2011; Imai, Keele, & Yamamoto, 2010;
Valeri & Vanderweele, 2013), so the method is becoming more and more
accessible.
SUMMARY
Analysing mediation is not just a matter of drawing one-way arrows
between variables represented in a path diagram, calculating product
estimators, testing those estimators for significance, and then describing
the results as indicating mediation if those tests yield significant results.
Without a strong rationale about directionality, proper attention to the
many assumptions that underlie even the simplest of path diagrams with
indirect effects, careful measurement and selection of variables, and collection of data in strong designs that support causal inference due to
time precedence in measurement, there may be little hope that results
from a mediation study will actually represent the target phenomenon.
Researchers who plan to study mediation in leadership should be conversant with these issues; otherwise, the literature may wind up cluttered with
many studies where mediation is claimed to be analysed, but the results
have little basis for this interpretation.
NOTE
*
I would like to thank Dr. Rosalie Hall for her helpful comments on an earlier version of
this chapter.
SCHYNS_9781785367274_t.indd 191
10/11/2017 15:20
192 Handbook of methods in leadership research
REFERENCES
Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (2010). On making causal claims: A
review and recommendations. Leadership Quarterly, 21(6), 1086–1120.
Baker, D.F., Larson, L.M., & Surapaneni, S. (2015). Leadership intentions of young women:
The direct and indirect effects of social potency. Journal of Career Assessment, 24(4),
718–731.
Baron, R.M., & Kenny, D.A. (1986). The moderator–mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations. Journal of
Personality and Social Psychology, 51(6), 1173–1182.
Bullock, J.G., Green, D.P., & Ha, S.E. (2010). Yes, but what’s the mechanism? (Don’t expect
an easy answer). Journal of Personality and Social Psychology, 98(4), 550–558.
Cheung, M.W.L. (2009). Comparison of methods for constructing confidence intervals of
standardized indirect effects. Behavior Research Methods, 41(2), 425–438.
Cohen, J., Cohen, P., West, S.G., & Aiken, L.S. (2003). Applied multiple regression/­
correlation analysis for the behavioral sciences (3rd ed.). New York: Routledge.
Cole, D.A., & Maxwell, S.E. (2003). Testing mediational models with longitudinal data:
Questions and tips in the use of structural equation modeling. Journal of Abnormal
Psychology, 112(4), 558–577.
Cole, D.A., & Preacher, K.J. (2014). Manifest variable path analysis: Potentially serious and
misleading consequences due to uncorrected measurement error. Psychological Methods,
19(2), 300–315.
De Vries, R.E., Roe, R.A., & Taillieu, T.C.B. (2002). Need for leadership as a moderator of
the relationships between leadership and individual outcomes. The Leadership Quarterly,
13(2), 121–137.
Edwards, J.R., & Lambert, L.S. (2007). Methods for integrating moderation and mediation:
A general analytical framework using moderated path analysis. Psychological Methods,
12(1), 1–22.
Hallinger, P., & Heck, R.H. (1998). Exploring the principal’s contribution to school effectiveness: 1980–1995. School Effectiveness and School Improvement, 9(2), 157–191.
Hayduk, L.A., & Littvay, L. (2012). Should researchers use single indicators, best indicators, or multiple indicators in structural equation models? BMC Medical Research
Methodology, 12(159). Retrieved from http://www.biomedcentral.com/1471-2288/12/159
Hayes, A.F. (2013). Introduction to mediation, moderation, and process control analysis: A
regression-based approach. New York: Guilford.
Hicks, R., & Tingley, D. (2011). Causal mediation analysis. Stata Journal, 11(4), 605–619.
Hyman, H. (1955). Survey design and analysis: Principles, cases and procedures. Glencoe, IL:
Free Press.
Imai, K., Keele, L., & Yamamoto, T. (2010). Identification, inference, and sensitivity analysis for causal mediation effects. Statistical Science, 25(1), 51–71.
International Committee of Medical Journal Editors. (2015). Recommendations for the
conduct, reporting, editing, and publication of scholarly work in medical journals.
Retrieved from http://www.icmje.org/icmje-recommendations.pdf
Kenny, D.A. (2015). Mediation. Retrieved from http://davidakenny.net/cm/mediate.htm#CI
Kline, R.B. (2013). Beyond significance testing: Statistics reform in the behavioral sciences
(2nd ed.). Washington, DC: American Psychological Association.
Kline, R.B. (2015). The mediation myth. Basic and Applied Social Psychology, 37(4), 202–213.
Knight, C.R., & Winship, C. (2013). The causal implications of mechanistic thinking:
Identification using directed acyclic graphs (DAGs). In S.L. Morgan (Ed.), Handbook of
causal analysis for social research (pp. 275–299). New York: Springer.
Lau, R.S., & Cheung, G.W. (2012). Estimating and comparing specific mediation effects in
complex latent variable models. Organizational Research Methods, 15(1), 3–16.
Little, T.D. (2013). Longitudinal structural equation modeling. New York: Guilford.
MacKinnon, D.P., & Pirlott, A.G. (2015). Statistical approaches for enhancing causal inter-
SCHYNS_9781785367274_t.indd 192
10/11/2017 15:20
Mediation analysis in leadership studies ­
193
pretation of the M to Y relation in mediation analysis. Personality and Social Psychology
Review, 19(1), 30–43.
Mallinckrodt, B., Abraham, W.T., Wei, M., & Russell, D.W. (2006). Advances in testing
the statistical significance of mediation effects. Journal of Counseling Psychology, 53(3),
372–378.
Maxwell, S.E., & Cole, D.A. (2007). Bias in cross-sectional analyses of longitudinal mediation. Psychological Methods, 12(1), 23–44.
Neves, P. & Story, J. (2015). Ethical leadership and reputation: Combined indirect effects on
organizational deviance. Journal of Business Ethics, 127(1), 165–176.
Pearl, J. (2014). Interpretation and identification of causal mediation. Psychological Methods,
19(4), 459–481.
Petersen, M.L., Sinisi, S.E., & Van der Laan, M.J. (2006). Estimation of direct causal effects.
Epidemiology, 17(3), 276–284.
Preacher, K.J., & Hayes, A.F. (2004). SPSS and SAS procedures for estimating indirect
effects in simple mediation models. Behavior Research Methods, Instruments, & Computers,
36(4), 717–731.
Preacher, K.J., & Hayes, A.F. (2008). Asymptotic and resampling strategies for assessing
and comparing indirect effects in multiple mediator models. Behavior Research Methods,
40(3), 879–891.
Preacher, K.J., & Kelley, K. (2011). Effect size measures for mediation models: Quantitative
strategies for communicating indirect effects. Psychological Methods, 16(2), 93–115.
Riehl, C. (2012). Locating a focus on school–family–community partnerships within the
scholarship of educational leadership. In S. Auerbach (Ed.), School leadership for authentic
family and community partnerships: Research perspectives for transforming practice (pp.
10–28). New York: Routledge.
Selig, J.P., & Preacher, K.J. (2009). Mediation models for longitudinal data in developmental
research. Research in Human Development, 6(2–3), 144–164.
Sobel, M.E. (1982). Asymptotic intervals for indirect effects in structural equations models. In
S. Leinhart (Ed.), Sociological methodology (pp. 290–312). San Francisco, CA: Jossey-Bass.
Tafvelin, S., Armelius, K., & Westerberg, K. (2011). Toward understanding the direct and
indirect effects of transformational leadership on well-being: A longitudinal study. Journal
of Leadership & Organizational Studies, 18(4), 480–492.
Tate, C.U. (2015). On the overuse and misuse of mediation analysis: It may be a matter of
timing. Basic and Applied Social Psychology, 37(4), 235–246.
Thoemmes, F. (2015). Reversing arrows in mediation models does not distinguish plausible
models. Basic and Applied Social Psychology, 37(4), 226–234.
Trafimow, D. (Ed.) (2015). Disadvantages of mediation analyses in basic or applied social
psychology [special issue]. Basic and Applied Social Psychology, 37(4). Retrieved from
http://www.tandfonline.com/toc/hbas20/37/4?nav5tocList
Valeri, L., & Vanderweele, T.J. (2013). Mediation analysis allowing for exposure–mediator
interactions and causal interpretation: Theoretical assumptions and implementation with
SAS and SPSS macros. Psychological Methods, 18(2), 137–150.
Vanderweele. T.J. (2015). Explanation in causal inference: Methods for mediation and interaction. New York: Oxford University Press.
Wang, H., Law, K., Hackett, R., Wang, D., & Chen, Z. (2005). Leader–member exchange
as a mediator of the relationship between transformational leadership and followers’
performance and organizational citizenship behavior. Academy of Management Journal,
48(3), 420–432.
SCHYNS_9781785367274_t.indd 193
10/11/2017 15:20
194 Handbook of methods in leadership research
APPENDIX
SPSS Syntax for Linear Regression with Summary Statistics
comment regression analysis of summary data.
comment reported in De Vries, Roe, and Taillieu (2002, p. 126).
comment x, leadership support.
comment m, need for leadership.
comment xm, product term.
comment y, job satisfaction.
matrix data variables5x m xm y
/contents5mean sd n corr/format5lower nodiagonal.
begin data
0 2.63 0 3.87
1.00 .800 1.00 .58
717 717 717 717
.09
–.05 0
.49 –.04 .01
end data.
regression matrix5in(*)/variables5m x
/dependent5m/method5enter x.
regression matrix5in(*)/variables5x m xm y
/dependent5y/method5enter x m xm.
SCHYNS_9781785367274_t.indd 194
10/11/2017 15:20
9.
Person-oriented approaches to
leadership: a roadmap forward
Roseanne J. Foti and Maureen E. McCusker
INTRODUCTION
Organizational psychology is defined as the scientific study of people in
the workplace (Landy & Conte, 2013). To study a person means to study
how he or she “functions and develops as an active, intentional part of
an integrated person–environment system” (Magnusson & Stattin, 2006,
p. 401). Yet, with a few exceptions (e.g., Foti & Hauenstein, 2007; Meyer,
Stanley & Vandenberg, 2013), research in leadership has focused almost
exclusively on understanding relationships between separate dimensions
of people (e.g., values, traits, beliefs, perceptions, behaviors) and workrelated outcomes. From this “variable-oriented approach” (Block, 1971),
the units of analysis are the person’s individual components, or variables,
and the person is treated as a summation of the components. Although the
variable-oriented approach has been useful in examining dimension-level
cause-and-effect relationships across people, a less common alternative
approach, the “person-oriented approach,” can deepen an understanding
of these psychological concepts when studying people in organizations.
Person-oriented research treats the individual as an organized system of
dynamically interacting variables, which form a pattern within a person
over time (Magnusson, 1995). Thus, instead of addressing questions about
the individual components of people, the person-oriented approach seeks
to address questions related to a whole person, as a coherent, organized
totality (Bergman & Magnusson, 1997).
Research concerning leaders and leader development, as well as follower perceptions and dyadic relationships, is well suited for the personoriented approach because it inherently focuses on a whole person, not
simply his or her components. However, like most organizational research,
leadership inquiry has historically been variable-centered, resulting in a
potential misalignment of research questions, methodology and analysis
(Bergman & Vargha, 2013). Studying leaders and leadership as holistic and
parsimonious systems is appropriate and beneficial for two reasons: (1)
alignment of leader(ship) theory, methodology and analysis and (2) understanding of leadership processes beyond (a conglomerate of) dimensions,
195
SCHYNS_9781785367274_t.indd 195
10/11/2017 15:20
196 Handbook of methods in leadership research
toward a coherent totality consisting of patterns of interacting dimensions
over time. However, given the scarcity of person-oriented research in
­industrial-organizational psychology, organizational scientists attempting
to conduct person-oriented research often find themselves faced with an
array of challenging questions (Sterba & Bauer, 2010). Thus, in an attempt
to advance and refine leadership research, the goals of this chapter are to
generate insights into (1) the fundamental differences between variableoriented and person-oriented research, (2) the variety of methods that can
be used to conduct person-oriented research studies, (3) the factors that
researchers must consider when deciding whether and how to conduct a
person-oriented study, and (4) how the conclusions one can draw from
person-oriented research are different from (and can complement) the
information gleaned from more traditional variable-focused work.
We begin this chapter by first defining the person-oriented approach
both theoretically and methodologically, highlighting how its theory and
methods differ from the more traditional variable-oriented approach.
Next, we provide a comprehensive overview of the different types of
person-oriented analytical methods as well as when to use which technique. We then give one empirical example of the use of pattern-oriented
methods (i.e., latent profile analysis) for leadership research using previously collected data measuring people’s leadership perceptions using an
implicit leadership theory (ILT) scale (Epitropaki & Martin, 2004). We
conclude by discussing how person-oriented inquiry can further current
research in other areas within the organizational sciences as well as bridge
connections across different disciplines.
PERSON-ORIENTED APPROACH VS THE
VARIABLE-ORIENTED APPROACH
The person-oriented and variable-oriented approaches each contain two
facets: a theoretical facet and a methodological one (Bergman & Trost,
2006). Though interrelated, it is important to differentiate both facets
for two reasons. First, past research has confused the two facets or even
defined one facet by the other. For example, one may claim to use a
person-oriented approach solely because the methods used are person oriented (Sterba & Bauer, 2010). Second, the true person-oriented research
consists of both person-oriented theory and person-oriented methods, the
latter we refer to as pattern-oriented methods (Von Eye, 2010). It is not
uncommon for researchers studying organizational phenomena, including
leaders, to ground research in person-oriented theory but use variableoriented methods (Bergman & Trost, 2006). In addition to the fact that
SCHYNS_9781785367274_t.indd 196
10/11/2017 15:20
Person-oriented approaches to leadership ­
197
doing so results in a problematic theory–method mismatch (Bergman &
Andersson, 2010), it does not constitute the full person-oriented approach
(Magnusson & Stattin, 2006). In this section we will define both facets of
the person-oriented approach independently and explain how each deviates from the more familiar variable-oriented approach.
The Theory Facet
The emergence of variable-oriented theory in psychology dates back to the
early nineteenth century, when a rapid surge of novel research methods
and statistical analyses drew attention to quantitative measurement,
empirical analysis, and experimental replication. The result of this detailoriented perspective was a growth in the variable as the unit of measurement, analysis and, in turn, theory (Bergman & Wångby, 2014). Since
then, the variable-oriented approach has dominated most psychological
research, especially within organizational sciences, and is largely responsible for today’s advanced state of knowledge discovery.
According to Magnusson and Stattin (2006), the term “variable” as
a psychological concept represents a particular “aspect of individual
functioning” (p. 431). Thus, the theory underlying the variable-­oriented
approach is based in identifying and explaining causal relationships
between hypothetical constructs. In this way, all the variables are each
considered separate entities, which may vary both intrapersonally over
time and interpersonally across people (Bergman & Magnusson, 1997).
However, the relationships between the hypothetical constructs are
assumed to hold across people in a specified functional form (e.g., linear,
curvilinear). So, psychological variable-oriented theory is fundamentally nomothetic, as it seeks to explain behavior across all groups of
people using variable-level relationships (Bergman & Andersson, 2010;
MacDougall, Bauer, Novicevic & Buckley, 2014).
While variable-oriented theory views a person as a summation of many
variables, the theoretical facet of person-oriented theory is grounded in a
“holistic-interactionistic” view of people as organized gestalts who change
as “integrated totalities” with their environments. Although personoriented researchers claim that slightly different theoretical tenets underlie person-oriented research (Bergman & Magnusson, 1997; Bergman
& Wångby, 2014; Von Eye & Bogat, 2006), Sterba and Bauer (2010)
unify them into six principles: individual specificity, complex interaction,
interindividual differences in intraindividual change, pattern summary,
holism, and pattern parsimony. Next, we briefly describe each of the
principles and relate them to the study of leaders and leadership. These
principles and examples are consistent with a growing body of researchers
SCHYNS_9781785367274_t.indd 197
10/11/2017 15:20
198 Handbook of methods in leadership research
who ­conceptualize leaders as unified totalities of dynamically interacting
components (e.g., Foti & Hauenstein, 2007; Mumford et al., 2000) and
leadership as a process resulting from the dynamic interactions between
all the actors, their attributes, and the context within a particular social
system (e.g., DeRue & Ashford, 2010; Fairhurst & Uhl-Bien, 2012).
The first principle, individual specificity, states that the functioning,
process, and development of one’s behavior are at least partially unique to
that individual. For example, a leader’s development is influenced by many
different factors in a system, and how he or she operates given the many
factors is at least in part idiosyncratic (Day, Fleenor, Atwater, Sturm,
& McKee, 2014). The complex interaction principle states that psychological processes and behaviors involve complex, multilevel interactions,
such as Person × Person × Time × Situation × Group, which cannot be
neglected. This principle can be exemplified when considering perceptions
of a leader. The likelihood that one perceives another person as a leader
depends on (the pattern of) the target’s characteristics and on (the pattern
of) the perceiver’s characteristics, as well as the interaction of each of the
totalities within a given context (DeRue & Ashford, 2010). The third principle states that there exist interindividual differences in intraindividual
change. That is, given the complexity and uniqueness of interactions, some
people differ in how they change and/or remain constant, but, as stated by
the fourth principle, pattern summary, there is order and structure in these
differences. This lawfulness can be organized into patterns or profiles of
variables. For example, leaders possess different leadership styles (e.g.,
transformational, transactional, laissez-faire), as exhibited by different
sorts of behaviors within particular contexts. Together, these intraindividual differences in behaviors can be clustered to reflect heterogeneous
subgroup differences in leadership styles (Doucet, Fredette, Simard, &
Tremblay, 2015). Furthermore, the fifth principle of holism reflects the
person-oriented belief that since components of a process (i.e., variables)
are “inextricably interwoven and believed to interact,” they cannot be
interpreted outside of the system in which they operate (Bergman &
Trost, 2006, p. 604). So, in the previous example, reducing the pattern of
leadership behaviors to individual leader behaviors and interpreting them
in isolation is meaningless because doing so neglects the complex interactions that drive order and structure in the process. Finally, the principle of
pattern parsimony implies there are a finite number of patterns of profiles
that can emerge, despite the fact there are often many or infinitely more
potential patterns. That is, through a process of “self-organization” the
patterns form a limited number of clusters of similar profiles, reflecting
a certain order, or structure. O’Shea, Foti, Hauenstein, and Bycio (2009)
reflected this idea in a study that examined the pattern of behaviors exhib-
SCHYNS_9781785367274_t.indd 198
10/11/2017 15:20
Person-oriented approaches to leadership ­
199
ited by effective leaders. Although there were eight possible combinations
of the three dichotomized behaviors (transformational, contingent reward,
and passive), people typically clustered into three different profile types.
In sum, the primary difference between the variable-oriented and the
person-oriented approach, from a theoretical standpoint, is that the
former is a perspective that emphasizes understanding relationships
between constructs at the dimension level, whereas the latter is a perspective that emphasizes understanding a dynamic totality of interacting
dimensions.
The Method Facet
Methods are tools for analysing data to understand the processes operating within given psychological systems. For a correct application of statistical methods, however, it is crucial to recognize that they are tools for
analysis of data in the same way scalpels, forceps and clamps are tools for
surgery. No statistical tool has a value on its own in the research process;
it is only when a statistical tool matches the phenomena under investigation that it can contribute scientifically solid answers to relevant questions
(Magnusson & Stattin, 2006).
As with the theory facet, the measurement model of the person-oriented
approach differs from the measurement model of the variable-oriented
approach. The method facet of person-oriented theory is known as the
pattern-oriented method. The goal in variable-centered analysis is to
explain as much variance in an outcome variable by one or more predictor variables (Meyer et al., 2013). The units of analysis are variables, and
one data point, or datum, represents one person’s standing on a particular
dimension (Magnusson & Stattin, 2006). The variables are each assumed
to have specified, functional forms (e.g., linear, non-linear, or curvilinear) across all individuals and are regarded as distinct units. Similarly,
if interactions among variables are analysed, they are also specified and
modeled directly with the addition of product and polynomial terms
(Bauer & Shanahan, 2007). Analysis is typically conducted using general
or generalized linear models, such as analysis of variance, regression, multiple regression, or structural equation modeling (Bergman & Andersson,
2010) and results in nomothetic group generalizations about relationships
between variables.
Consider an example of the variable-oriented approach in studying
intelligence and leadership effectiveness: the measurement of person A
on the latent dimension of intelligence derives its psychological significance in relation to the positions for individuals, B, C, D and as shown
in Figure 9.1. Most psychological research, including leadership, uses
SCHYNS_9781785367274_t.indd 199
10/11/2017 15:20
200 Handbook of methods in leadership research
D
A B
C
Latent dimension: Intelligence
Variable-oriented approach
A
Latent dimension: Intelligence
Latent dimension: Agreeableness
Latent dimension: Conscientiousness
Latent dimension: Extraversion
Person-oriented approach
Source: Permission granted by John Wiley & Sons, Inc. for this modified version of
Figure 8.2(a) and (b) (p. 444) from Magnusson and Stattin (2006).
Figure 9.1 Measurement model for the variable-oriented and personoriented approach
variable-­oriented approaches. For example, when we are interested in
the relationship between perceptions of leader intelligence and leadership
effectiveness, we discuss our findings by saying that higher leader intelligence is related to greater effectiveness (Judge, Colbert, & Ilies, 2004). In
other words, we are saying that two variables, a score on intelligence and
a rating of leadership effectiveness, are statistically related to each other.
The strengths of variable-oriented analyses are many, including objectivity and clarity of the measurement and scales, control of confounds,
the ability to determine how much variance in the outcome is explained
by a particular variable, and the ability to make population-level causal
inferences based on statistical model testing (Bergman & Andersson, 2010;
Von Eye et al., 2006). However, these methods, and the subsequent statements that result, say nothing about individuals. The expectation is that
these aggregate group scores generalize to populations. In other words,
we make assumptions that interindividual differences are negligible or
random (Von Eye & Bogat, 2006).
While the goal of variable-oriented methods is to identify relationships between variables and account for variance, the purpose of
pattern-oriented methods is to identify subgroups in a population
SCHYNS_9781785367274_t.indd 200
10/11/2017 15:20
Person-oriented approaches to leadership ­
201
based on a particular configuration of variables (Meyer et al., 2013).
Bergman and Trost (2006) put it simply by stating that this type of
analysis entails first “identifying a subsystem relevant to the problem
under study, measuring its components, and studying them all together
as an undivided whole” (p. 604). Thus, the person as a whole is the unit
of analysis and one datum represents a person’s standing on a pattern
of dimensions (Magnusson & Stattin, 2006). These patterns, or profiles,
are assumed to capture a complex process of dynamic interactions of a
set of variables within a system or individual (Bergman & Andersson,
2010), as opposed to modeling each interaction individually (Bauer
& Shanahan, 2007). There are a variety of person-oriented analytic
methods (discussed below), but in general, these methods first identify
meaningful clusters based on a pattern of dimensions, then conduct subsequent analyses with the established profiles as predictors or outcomes
(e.g., Foti & Hauenstein, 2007). In this sort of analysis, the functional
form of variables and interactions are not specified, as it is assumed
that their reciprocal and interdependent interactions are unique to an
individual or the system (Magnusson & Stattin, 2006). So, by classifying individuals into types or typologies, pattern-oriented methods
allow for idiographic examinations of systems (people) in their totalities
(Bergman & Trost, 2006).
Recall the intelligence and leadership effectiveness example discussed
above. In the person-oriented approach to this example, which is depicted
in Figure 9.2, the measurement of person A on the latent construct of
intelligence derives its psychological significance from its position in a
configuration of data for the same individual using his or her position in
the latent constructs of agreeableness, conscientiousness and extraversion.
These latent constructs are assumed to represent simultaneously working
dimensions in the system of leader traits. The implication of this measurement model is that statistics yield information about the individual, and
generalizations refer to individuals.
Moreover, as shown in Figure 9.2, each latent construct takes meaning
from all the other variables in the pattern. The same position for different individuals (A and B) on the latent construct of leader intelligence
may differ entirely in its significance for the rating of leadership effectiveness for the two individuals. That is, the same level of intelligence takes
meaning from the different levels of agreeableness, conscientiousness and
extraversion.
Thus, to describe individuals adequately, the researcher has to recognize
that any sample or population is rarely a homogeneous group. Rather, the
researcher examines the heterogeneity of possible patterns and groupings
of individuals that make sense from a theoretical perspective.
SCHYNS_9781785367274_t.indd 201
10/11/2017 15:20
202 Handbook of methods in leadership research
AB
Latent dimension: Intelligence
Latent dimension: Agreeableness
Latent dimension: Conscientiousness
Latent dimension: Extraversion
Source: Permission granted by John Wiley & Sons, Inc. for this modified version of
Figure 8.3 (p. 445) from Magnusson and Stattin (2006).
Figure 9.2 Profiles for two individuals on three latent dimensions of leader
perceptions
PATTERN-ORIENTED METHODS
In the latter part of the previous section we provided a general review of
pattern-oriented methods to highlight the differences from the traditional
methods. Given that the primary purpose of this chapter is to provide
a roadmap for conducting person-oriented leadership research, the following section provides a more extensive description of several different pattern-oriented analytical methods. In addition, we indicate which
sorts of research questions are answered by each technique, which we
hope will guide researchers in aligning their research methodology with
research questions and theory. A summary of this guide is included in
Table 9.1.
The growing trend of the person-oriented approach in the behavioral
sciences (Bergman & Wångby, 2014), along with recent calls for researchers to develop more advanced pattern-oriented analytical tools (e.g.,
Vandenberg & Stanley, 2009), has resulted in a recent growth in the
number of pattern-oriented methods available. Accordingly, we have categorized the methods into three classes of analyses, adapted from Bergman
and Wångby’s classification (2014): (1) cluster analysis and its extensions,
(2) configural frequency analysis and its extensions, and (3) model-based
classification analysis and its extensions. It should be noted that all three
classes of methods include multiple different procedures, and describing
all of them is beyond the scope of this chapter. So, we have limited our
description to the methods most commonly employed and discussed in the
literature.
SCHYNS_9781785367274_t.indd 202
10/11/2017 15:20
Person-oriented approaches to leadership ­
203
Table 9.1
Summary of pattern-oriented methods
Method
Type
Indicator Data
When to Use
Cluster
analysis
LICUR
Exploratory
Cross-sectional
Exploratory
Longitudinal
ISOA
Exploratory
Longitudinal
CFA
Exploratory
Cross-sectional
P-CFA
Exploratory
Longitudinal
A-CFA
Exploratory
Longitudinal
Bayesian
CFA
Model based
Cross-sectional
(currently)
LCA
Model based
Cross-sectional
LPA
Model based
Cross-sectional
RMLCA
Model based
Longitudinal
LCGM/
LPGM
Model based
Longitudinal
LTA
Model based
Longitudinal
When expecting lack of
homogeneity in the sample
When examining large
developmental shifts
When examining patterns of
stability or change of both
cluster structure and individual
When assuming stability in cluster
classification
When seeking to explore all
possible combinations of
patterns
When assuming independence of
time points within classes over
time
When assuming auto-associations
between time points within
classes over time
When working with prior
information
When examining composites of
types and antitypes
When working with categorical
indicators
When working with continuous
indicators
When working with categorical
indicators
When working with few indicators
and three or more time points
When it is assumed growth follows
a particular functional form
When working with large sample
sizes and three or more time
points
When examining whether profiles
or classes are stable across time
When examining whether people
tend to remain in or transition
among classes
SCHYNS_9781785367274_t.indd 203
10/11/2017 15:20
204 Handbook of methods in leadership research
Cluster Analysis and its Extensions
The simplest and possibly most straightforward pattern-oriented method
is cluster analysis because it requires minimal external knowledge and
information to form groups (Von Eye & Bergman, 2003). The goal of
this primarily exploratory method is to disaggregate data from one entire
sample into a number of smaller, non-overlapping groups, or clusters
(Bergman & Wångby, 2014). This is done by identifying a collection of
cases (e.g., people) in the sample that are more similar to each other than to
the other cases and grouping them into their own cluster. Although there
are multiple analytical techniques to do so, the most common is Ward’s
(1963) procedure, which uses the squared Euclidian distance between data
points to minimize variance within groups and maximize variance between
them. The result is a set of groups that each contain “relatively homogeneous” cases (intragroup similarity), but that differ in a meaningful way from
the other groups (intergroup difference).
This idea of meaningful interpretation is crucial in cluster analysis. Von
Eye and Bergman (2003) claim the most important problem associated
with cluster analysis is the tendency for researchers to trust the resulting
clusters as meaningfully interpretable when they may in fact be non-­
interpretable artifacts of the analyses. Thus, safeguarding the trustworthiness of the cluster solution, as well as using robust measurements and
appropriate sample size is of paramount importance in cluster analysis.
Cluster analysis can be conducted on both cross-sectional and longitudinal data. Although cross-sectional cluster analysis can be considered in
part person oriented as it focuses on a pattern of the individual, it is not
entirely person oriented because it neglects the process component of the
person as a dynamic system (Bergman & Andersson, 2010). Rooted in
the holistic-interactionistic research paradigm, the individual is seen as an
organized whole with elements operating together to achieve a functioning system. This dynamic system results from interactions among its components (Bergman & Wångby, 2014). Examples of components include
behaviors, biological factors, and environment factors. Dynamic systems
are characterized by moments of equilibrium and moments of disequilibrium, thus these person processes can be captured only with longitudinal
data. As a result, researchers have developed longitudinal approaches to
cluster analysis, and there are two main ways to do so.
The first method is to conduct a cluster analysis at each time point individually, and then compare the resulting clusters across time points. One
example of a longitudinal extension of cluster analysis is called LICUR
(linking clusters after the removal of a residue; Bergman, Magnusson, &
El-Khouri, 2000), which first removes residue from each time point, runs
SCHYNS_9781785367274_t.indd 204
10/11/2017 15:20
Person-oriented approaches to leadership ­
205
a cross-sectional-like cluster analysis at each time point, then links all time
points in the sequence to test for different types of cluster membership
configurations (Bergman & Magnusson, 1997). This method is particularly useful for questions relating to stability and change of both the individual and the structure of the clusters (Bergman & Wångby, 2014), as well
as for exploring periods characterized by “dramatic developmental shifts”
(Bergman & Magnusson, 1997, p. 300).
The second type of longitudinal cluster analysis is based on the assumptions that the system of cluster classifications remains across all time
points (i.e., is time invariant) and that individuals belong to one cluster
of this system at each point in time. An example of this method is ISOA
(I-states as objects analysis; Bergman, Nurmi, & Von Eye, 2012), which
first uses cluster analysis to identify a time-invariant clustering system,
then each individual is assigned to a cluster at every time point. Finally
each person’s sequence of cluster membership over time is examined
(Bergman & Magnusson, 1997). ISOA is best suited for small samples sizes
over shorter time periods.
In sum, cluster analysis methods can be very useful exploratory methods
to identify and describe non-homogeneous subgroups in data based on a
pattern of variables. However, clustering methods provide limited practical utility in research, given their fundamentally descriptive nature (Sterba
& Bauer, 2010). Thus, the following two methods describe probability and
model-based pattern-oriented methods that allow researchers to make
more advanced inferences about the clusters and causal relationships.
Configural Frequency Analysis and its Extensions
The goal of configural frequency analysis (CFA)1 is to examine all possible
combinations of a pattern of variables and determine which ones occur
more or less frequently than expected by a base model (Von Eye, 2002).
This grants the researcher the ability to study “all possible value patterns
directly” to better understand the various emergent structures in the data
(Bergman & Wångby, 2014, p. 34) and to make predictions regarding the
frequency of occurrence of particular patterns. The steps to conduct a
standard CFA are as follows, as described by Von Eye (2002): first, to
allow for a reasonable number of possible configurations, all variables are
dichotomized (e.g., low, high) or trichotomized (e.g., low, medium, high),
unless the variable is naturally grouped, as with nominal or ordinal data.
Second, all possible configurations of variables and levels are listed and
assigned to cells. Then, a base model is specified, which serves as a referent for comparison of the frequency of each of the possible configurations
of variables. The frequency of each observed pattern is then compared
SCHYNS_9781785367274_t.indd 205
10/11/2017 15:20
206 Handbook of methods in leadership research
to the base model.2 If the pattern of variables occurs significantly more
frequently than is expected by the base model, it is considered a “type,”
whereas if a pattern occurs significantly less frequently than is expected, it
is considered an “antitype.” It should be noted that in classical CFA, the
base model is one of variable independence, or of main effects, so that the
emergence of types and antitypes only results when there are significant
interactions between the variables (Von Eye et al., 2006). The final step
includes interpretation of the resulting types and antitypes, based on the
research question and purpose of the study (recall Von Eye and Bergman’s
2003 emphasis on meaningful interpretation of clusters/patterns/types).
Like cluster analysis, the resulting types and antitypes from CFA are
often used as predictors or outcomes in subsequent analyses. A study by
O’Shea et al. (2009) exemplifies how CFA can be used to answer novel
leadership questions. The purpose of their study was to test Bass’s (1985)
claim that optimal leaders exhibit both transformational and transactional leadership behaviors. To do so, the researchers measured three
variables of leadership behaviors: transformational behaviors (TRF),
contingent reward behaviors (CR; i.e., transactional) and passive management-by-exception behaviors (P-MBE); each was dichotomized. They also
measured leadership effectiveness as measured by a series of subordinate
outcomes. They then conducted a CFA using the eight possible high–low
combinations to examine which configurations were likely to occur more
(types) or less (antitypes) frequently than chance. Three configurations
emerged as types (high on TRF, CR, P-MBE; high TRF, high CR, low
P-MBE; and low TRF, low CR, high P-MBE), and the remaining three
configurations emerged as antitypes. Upon identification of the types, the
researchers carried out planned comparisons of the effectiveness of each
leadership type. In support of their hypotheses and Bass’s (1985) claim,
the most effective leaders were those types who exhibited both transformational and transactional leadership behaviors.
The previous example reflects both the exploratory power of CFA to
answer questions related to the probability of occurrence of particular patterns as well as its predictive power to test their antecedents and outcomes.
In addition, CFA is unique in that it treats both frequently occurring configurations and rare, or outlying, patterns as meaningful. However, like
cross-sectional cluster analysis, classical CFA is only truly person-oriented
to the extent that it incorporates longitudinal development of a person.
Work by Von Eye and colleagues have developed extensions of CFA for
longitudinal data, allowing for examination of configural changes over time
(Von Eye, 2002; Von Eye, Mun, & Bogat, 2009). There are two longitudinal extensions, each with differing base models. The first method, P-CFA,
resembles classical CFA in that the base model treats time points on one or
SCHYNS_9781785367274_t.indd 206
10/11/2017 15:20
Person-oriented approaches to leadership ­
207
multiple variables as independent, and types (antitypes) are those patterns
across variables and time that occur more (less) frequently than expected
given the base model. The second method, A-CFA, uses an auto-associative
base model that instead of treating each time point on each variable as independent (as in P-CFA), multiple time points on each variable are treated
as repeated observations of the same variable. In this way, “types and antitypes reflect longitudinal relationships among categorical variables after
auto-associations are taken into account” (Von Eye et al., 2008).
One final extension of CFA we will mention is one rarely used but holds
potential for overcoming some of the disadvantages of standard CFA
methods: Bayesian CFA. Bayesian logic is grounded in the notion that the
probability of occurrence of any given value can be estimated from two
sources: the observed data and prior information (Gutiérrez-Peña & Von
Eye, 2000). For the purposes of this chapter, we will not discuss in depth
how to conduct a Bayesian CFA (see Gutiérrez-Peña & Von Eye, 2000
for more detailed explanation), but we do note its benefits with regard to
CFA. First, since Bayesian methods rely on joint probabilities of possible
pattern types and antitypes, there is no need to adjust the experiment-wise
alpha level. Second, this method calculates the probability of occurrence
for all cross-classification patterns, allowing for the possibility of composites of types and antitypes. In view of the flexibility afforded by the
Bayesian approach to analysis, it is no surprise its influence is growing
rapidly in the organizational sciences in very recent years, and we anticipate this growth will be mirrored in the person-oriented domain.
To summarize, the class of CFA methods and its extensions grant
researchers the ability to explore and test all potential combinations of
patterns in their data, as well as measure each one’s probability of occurrence. Accordingly, this class of methods is very useful for researchers
interested in questions necessitating a comparison of all possible combinations of variables, those examining which combinations are most common,
or those interested in studying unconventional patterns or trajectories
(i.e., the antitypes). In addition, the vast number of planned comparisons
results in restrictions on the experiment-wise error rate. Given that CFAs
are exploratory in nature, researchers are increasingly turning to modelbased pattern-oriented methods, which use “more rigorous criteria for
determining the number of clusters or classes” (Meyer et al., 2013, p. 197).
We describe several different ones below.
Model-based Analyses
Model-based methods differ on a theoretical level from the previous
methods reviewed. This class of methods is grounded in the idea that there
SCHYNS_9781785367274_t.indd 207
10/11/2017 15:20
208 Handbook of methods in leadership research
exists unobservable heterogeneity (or latent categorical classes) in a population, which manifests as configurations of observable variables (Wang &
Hanges, 2011). The goal of model-based methods is to uncover how many
and what types of latent classes or profiles exist, similar to factor analysis
(Vandenberg & Stanley, 2009). There are multiple different model-based
methods; however, in this chapter we focus on six of the most prevalent
ones: latent class analysis (LCA), latent profile analysis (LPA), repeated
measures latent class analysis (RMLCA), latent class growth modeling/
latent profile growth modeling (LCGM/LPGM), growth mixture modeling (GMM) and latent transition analysis (LTA). LCA and LPA are very
similar, except that the former is used for categorical predictors, while the
latter is used for continuous predictors. LTA, LCGM and LPGM are all
longitudinal extensions of the LCA and LPA.
Simply put, in LCA and LPA, one first specifies a latent model with the
fewest number of possible latent classes or profiles, and then latent classes
or profiles are added to the model until it fits the data well (Von Eye &
Bergman, 2003). Once the model has been specified, individuals are classified into the latent classes or profiles based on their relative probabilities.
Finally, predictors and outcomes of classes or profiles are studied per
traditional statistical analyses (MacDougal et al., 2014). LCA and LPA
methods work in the same way, except that the former is used when the
indicator variable is categorical and the latter is used when the indicator
variable is continuous.
There are several advantages of this method. First, the estimation of the
model’s parameters produces confidence intervals as well as estimations of
model fit (Bergman & Wångby, 2014). Second, individuals are classified
into the groups only after the latent groups have been established. Third, it
uses a “probabilistic classifying approach,” allowing for there to be uncertainty in class membership (Wang & Hanges, 2011). Finally, LPA can
model curvilinear relationships between observed variables (MacDougall
et al., 2014).
Next we will discuss four longitudinal extensions of the cross-sectional
model-based analysis discussed above. The first one is repeated measures
latent class analysis (RMLCA), also known as longitudinal latent class
analysis (LLCA). The goal of RMLCA is to identify subgroups of individuals based on their differing patterns of change over time. Thus, the
criterion for classifying individuals into particular classes is the pattern
of change of a small number of variable(s) over time. So, each latent
class corresponds to a particular pattern of change in an outcome over
time (Collins & Lanza, 2010). In addition, “grouping variables,” such as
demographic attributes, can be included to determine if the longitudinal
patterns differ based on additional characteristics (Lanza & Collins, 2006,
SCHYNS_9781785367274_t.indd 208
10/11/2017 15:20
Person-oriented approaches to leadership ­
209
p. 553). Similar to cross-sectional LCA, the indicator variables in RMLCA
must be categorical, but the outcome variables and covariates may be
either categorical or continuous. Although RMLCA can be conducted
using many different indicators of the latent class, it is best for data with
just a few indicators and three or more time points (Collins & Lanza,
2010).
A similar longitudinal extension is latent class growth analysis/latent
profile growth analysis (LCGA/LPGA). The difference between LCGA/
LPGA and RMLCA is that these LCGA/LPGA extensions assume that
growth or development over time follows a particular functional form
(Vandenberg & Stanley, 2009). That is, individuals are grouped together
based on similar patterns of change in levels and slope of the indicator
variable(s) over time (Sterba & Bauer, 2010). The goal in creating classes
using LCGA/LPGA is to minimize the variability of individuals’ trajectories within classes, but maximize the variability in trajectories between
classes. LCGA/LPGA assumes that all individuals within a particular
class are homogeneous in terms of change; they all follow the same trajectory over time and any within-class deviations in trajectories are due to
random noise (ibid.). As with the rest of the model-based analyses, the
resulting classes characterized by particular growth trajectories can then
be compared to other covariates or used as predictors or outcomes in
subsequent analyses. LCGA/LPGA methods are best when working with
large data sets consisting of at least three or more time points.
The third longitudinal extension is growth mixture modeling (GMM;
Muthén, 2004). GMM is a pattern-oriented extension of the more traditional, variable-oriented latent growth curve modeling (Wang & Bodner,
2007). (See Hall in Chapter 13 of this volume for a more detailed discussion of the application of latent growth curve modeling for leadership
research.) Similar to LCGA/LPGA, GMM allows researchers to uncover
classes of longitudinal change trajectories. However, GMM differs from
the former in that it relaxes the assumption of intraclass homogeneity
by allowing the growth parameters to vary systematically across people
within classes. Thus, although there may exist qualitatively different
classes of trajectories, there also may exist quantitative variation in the
intercept and slope within each class; hence the word “mixture” (Sterba &
Bauer, 2010).
Longitudinal transition analysis (LTA) is the fifth type of longitudinal
pattern-oriented analysis. Similar to GMM, LTA does not assume uniform
homogeneity in class over time. However, what makes LTA unique is its
ability to model changes (or “transitions”) among classes or profiles classto-class or profile-to-profile across time points (Collins & Lanza, 2010).
It does so by calculating the probability of being in a ­particular class or
SCHYNS_9781785367274_t.indd 209
10/11/2017 15:20
210 Handbook of methods in leadership research
profile group at a given time point, conditional on the class membership
in the immediately preceding time point (Meyer et al., 2013). Thus, results
of LTA estimate the probability that one will remain in his or her class or
profile group over time. LTA is useful for answering questions about (1)
whether particular profiles or classes are stable or variable across time and
(2) whether people tend to remain in particular classes or transition among
them (Wright & Hallquist, 2014).
LEADER PERCEPTIONS: MATCHING
MEASUREMENT TO OUR RESEARCH QUESTION
In the next section, we walk readers through an empirical example related
to the measurement of implicit leadership theories (ILTs) and why use of
a person-oriented approach and methods matches our research question.
ILTs are schemas that guide perceptions of leaders by providing a set
of assumptions and expectations about how leaders behave and how to
respond to them (Foti, Bray, Thompson, & Allgood, 2012; Lord, Foti, &
De Vader, 1984). Perceivers hold ILTs about leaders based on direct and
indirect experiences, highlighting the importance of development and environment as essential determinants of ILTs. Over the last two decades, the
conceptualization of ILTs has become increasingly complex. The classical
perspective of leadership perceptions was based on a graded categorical
structure defined by central or prototypical characteristics, but possessed
fuzzy boundaries (Lord et al., 1984; Lord, Foti, & Phillips, 1982). More
recent research suggests that ILTs are created dynamically on the basis of
contextual input and that connectionist network models provide a better
representation of leadership categories and how they emerge in a particular situation (Shondrick & Lord, 2010). In connectionist models of ILTs,
information processing about leaders is accomplished by multiple trait
units acting in parallel with the resulting pattern of processing being meaningful (Lord, Brown, Harvey, & Hall, 2001). Moreover, characteristics of
leaders can be incorporated into the dynamic models, so that the prototype itself might vary depending on the gender (Foti, Knee, & Backert,
2008; Foti & Wills, 2015) or ethnicity of the leader (Sy et al., 2010) or the
affective state of perceivers (Boyd & Foti, 2016). Thus, one might expect
both strong individual differences in category structures among perceivers who have different experiences, and also strong contextual effects that
reflect constraints from different situations.
A connectionist model of ILTs matches a person-oriented approach in
several key aspects. First, connectionist models of ILTs emphasize contextual constraints; similarly, the person-oriented approach is concerned with
SCHYNS_9781785367274_t.indd 210
10/11/2017 15:20
Person-oriented approaches to leadership ­
211
individuals and how individuals operate in or are influenced by the context
in which they exist (Magnusson, 1988). In other words, the person in the
environment is just as important as the person by environment interaction
and the environment itself. Second, in connectionist models, constraints
activate different meaningful prototype patterns (Lord et al., 2001). A
hallmark characteristic of the person-oriented approach is that variables
in and of themselves have limited meaning (Bergman & Magnusson,
1997). It is the profile of these variables as part of an indivisible pattern
that takes on meaning and begins to describe individuals. Finally, each
variable takes its meaning from the other variables in the pattern to form
the coherent whole. Thus, when we assume that the relationships among
variables are not uniform across all the values that a variable might take,
we can develop profiles, patterns, or configurations that describe individuals, not scores on the variables (Bogat, 2009).
Are There Distinct Profiles of Leader Perceptions?
Our research question investigates whether there are distinct profiles of
leader perceptions. We collected data from 709 university students (60
percent female). Using the 21-item scale developed by Epitropaki and
Martin (2004), students rated how characteristic the items were of a leader,
with no definition of the term provided. Responses were provided on a
nine-point Likert-type scale, where 1 indicated “not at all characteristic”
and 9 indicated “extremely characteristic.” The 21 characteristics appear
in Table 9.2. If we were going to use a variable-oriented method, we
would typically analyse the data using exploratory or confirmatory factor
analysis to identify the structure of ILTs. More specifically, the objective
for a variable-oriented method is to identify the number and nature of the
factors or latent variables that produce the observed covariation and variation in the 21 manifest variables. Typical factor analytic dimensions for
the ILT scale appear in Table 9.2. The assumption of factor analysis is that
Table 9.2
21-item implicit leadership scale and associated factors
Sensitivity
Intelligence
Dedication
Dynamism
Tyranny
Masculinity
Understanding
Sincere
Helpful
Clever
Knowledgeable
Educated
Intelligent
Motivated
Dedicated
Hard-working
Energetic
Strong
Dynamic
Domineering
Pushy
Manipulative
Conceited
Selfish
Loud
Masculine
Male
Source:
Epitropaki & Martin (2004).
SCHYNS_9781785367274_t.indd 211
10/11/2017 15:20
212 Handbook of methods in leadership research
correlations between manifest variables arise because of their dependency
on one or more of the same factors. Thus, Understanding and Sincere are
highly correlated because both are influenced by a common underlying
factor of Sensitivity. Moreover, the relationships between these variables
and their way of functioning in the totality of an individual are the same
for all individuals. In our factor analysis example, each variable has the
same factor loading for all individuals and reflects what is characteristic
for the average individual.
However, does this analysis accurately reflect theorizing about ILTs?
We think not. Given ILTs develop based on individuals’ direct and indirect experiences AND can vary based on the characteristics of the perceiver and the context, it follows we should investigate within-individual
patterns across the 21 characteristics. We can do this with latent profile
analysis (LPA).
Basic ideas of LPA
As previously mentioned, in LPA, individuals can be divided into subgroups based on an unobservable construct. The construct of interest is
the latent variable and subgroups are called latent profiles. Latent profile
analysis utilizes a categorical latent variable and continuous indicators
(Collins & Lanza, 2010; Foti, Thompson, & Allgood, 2011), which differs
from factor analysis where both the latent variable and the indicators, or
observed variables, are continuous (Collins & Lanza, 2010). More specifically, latent profile analysis is often used to identify and describe a set
of mutually exclusive and exhaustive latent profiles whose members are
characterized by similar response sets of manifest indicators. The primary
goal is to maximize the homogeneity within groups (i.e., individuals within
a profile should look similar) and maximize the heterogeneity between
groups (i.e., individuals between profile groups should look different).
In LPA, the use of multiple indicators provides a basis for estimating
measurement error. When measurement error is present, many individuals’ responses do not point unambiguously to membership in a particular
group, thus true profile membership is unknown (Lanza, Flaherty, &
Collins, 2003). Key assumptions of LPA are that the latent profile indicators are (1) continuous and normally distributed within profiles (Bauer &
Curran, 2003) and (2) independent within profiles (conditional independence). The latter assumption is also present in many other latent variable
modeling techniques such as item response theory.
Model estimation
Using a person-oriented approach, a graphical representation of our
research question is seen in Figure 9.3. That is, we want to ascertain
SCHYNS_9781785367274_t.indd 212
10/11/2017 15:20
Person-oriented approaches to leadership ­
213
Leader
Profiles
Understanding
Figure 9.3
Sincere
…
Male
Graphical representation of person-oriented leader perceptions
whether there are distinct profiles of leader perceptions based on the 21
manifest indicators. In latent profile models, the data are used to estimate
the number of profiles in the population, the relative size of each profile and
the probability of a particular response to each manifest indicator, given
profile membership. The first step is model identification. This requires
the specification and testing of multiple profile solutions, typically from
one profile to ten profiles, using maximum likelihood estimation. Many
estimation procedures require initial values for the parameters to start the
estimation procedures. However, if different starting values produce very
different estimates and different log-likelihoods, then the model is not well
identified. Thus, the recommendation is to use many different sets of starting values, approximately 100 or more, and inspect the distribution of loglikelihood values. In a well-identified model, starting values only affect the
number of iterations required for the model to converge (Lanza, Bray, &
Collins, 2013). Typically, LPA models are fit without the use of any parameter restrictions, which allows the profiles to be different from each other.
If there are model estimation issues, however, restricting the variances to
be equal across profiles will greatly reduce the unknowns in the model and
can help with model identification. Finally, most LPA software can handle
missing data. If missing data on a variable depends on the variable itself,
this is referred to as missing not at random (MNAR). If missing data on a
variable does not depend on the variable itself, this is referred to as missing
at random (MAR). Most LPA procedures use a maximum likelihood
routine that adjusts for data that is MAR but not for NMAR.
SCHYNS_9781785367274_t.indd 213
10/11/2017 15:20
214 Handbook of methods in leadership research
The second step is model selection. Various model selection information
criteria have been proposed for comparing models with different numbers
of profiles. From these models, the designation of the “best-fitting” model
is determined using a variety of statistical indicators; including the Akaike
information criterion (AIC; Akaike, 1974), the Bayesian information
criterion (BIC; Schwarz, 1978) and the sample size-adjusted BIC (Sclove,
1987). All three indicators are penalized log-likelihood test statistics, which
penalize models for estimating too many parameters; moreover, both versions of the BIC further penalize models by sample size. A fourth statistical indicator is the bootstrapped likelihood ratio test (BLRT; McLachlan
& Peel, 2000). The BLRT in effect estimates a “difference” distribution by
which different models can be compared through the use of repeated sampling methods. None of these statistical indicators can determine model
fit in isolation. A determination of the best-fitting profile solution then
is based on which model has lower values for these fit indicators (lower
values indicates better relative fit). Finally, to help in the determination of
the optimal number of profiles, the interpretability of each profile must be
considered; specifically whether or not a specific profile solution is more
consistent with past theory and empirical research.
Returning to our example, models with one to ten profiles were fit
to assess leader perceptions. Models were fit using Mplus Version 6
(Muthén & Muthén, 1998–2010). Models with more than ten profiles
were not considered due to model instability (i.e., difficulty replicating the maximum likelihood solution). In all models, profile-specific
means and variances were free to vary across all manifest indicators.
Table 9.3 summarizes the fit criteria for the leader profile models. All
of the statistical criteria considered here indicated an improvement in
model fit as the number of profiles increased. Given that decrements in
the BIC appeared to attenuate between the three-profile and six-profile
solutions, these four solutions were substantively assessed (Lawrence
& Zyphur, 2011). The four-profile solution appeared superior to the
three-profile solution; each of the profiles in the four-profile solution
were adequately sized, all three profiles from the three-profile solution
appeared in the four-profile solution, and the additional profile in the
four-profile solution was interpretable based on leadership theory. In
addition, all profiles from the four-profile solution appeared in the
five-profile and six-profile solutions, and the larger models appeared to
differ from the smaller ones only by separating out smaller subgroups of
participants from the larger ones. These smaller subgroups were increasingly difficult to interpret theoretically, as they showed patterns of
means on the indicators inconsistent with the factor structure or existing
leadership theory. Given the tradeoffs among model parsimony, statisti-
SCHYNS_9781785367274_t.indd 214
10/11/2017 15:20
Person-oriented approaches to leadership ­
215
Table 9.3 Model fit information for latent profile analyses of leader
perceptions
Leader Profiles
No. of
profiles
1
2
3
4
5
6
7
8*
Number of free Log-likelihood
parameters
42
64
86
108
130
152
174
196
–34 273.471
–32 787.506
–31 837.777
–31 448.804
–31 145.515
–30 959.300
–30 830.608
–30 677.573
AIC
BIC
a-BIC
Entropy
R2
68 630.943
65 703.011
63 847.555
63 113.607
62 551.031
62 222.600
62 009.216
61 747.147
68 830.193
66 006.631
64 255.544
63 625.966
63 167.759
62 943.697
62 834.683
62 676.982
68 696.814
65 803.386
63 982.434
63 282.990
62 754.917
62 460.990
62 282.111
62 054.545
1.00
0.88
0.91
0.90
0.91
0.89
0.89
0.90
Note: * Model was not identified. Italic font indicates selected model. AIC = Akaike
information criterion; BIC = Bayesian information criterion; a-BIC = sample size adjusted
BIC. Heavy emphasis was placed on the utility and theoretical interpretation of a solution.
This approach to LPA model selection has been used elsewhere in the organizational
literature (Lawrence & Zyphur, 2011). In all models, within-profile indicator means and
variances were free to vary.
cal fit, and theory, the four-profile solution was selected as optimal to
describe leader perceptions.
The third step in our analysis is model interpretation. In interpreting
our models, we are interested in both latent profile homogeneity and latent
profile separation. Although homogeneity is not as straightforward to
interpret in LPA as in LCA or factor analysis, profile separation is still a
very helpful concept to consider. It is the degree to which the latent profiles can clearly be distinguished from each other. Homogeneity is analogous to the concept of saturation in factor analysis, whereas separation
is analogous to the concept of simple structure in factor analysis (Lanza,
Bray, & Collins, 2013). Looking at Table 9.4, we see the two estimated
parameters for the four-profile solution: latent profile prevalences and
profile-specific means. Prevalences are the probability of membership in
a particular latent profile. Profile-specific means are analogous to factor
loadings and express the relationship between manifest and latent variables. The profile-specific means form the basis for interpreting the latent
structure. Profile-specific variances are the third parameter estimated in
most LPA software and refer to the variance of a specific manifest indicator (e.g., understanding) given membership in a particular latent profile.
For our data, the profiles were labeled Prototypical (29 percent prevalence, n 5 206), Laissez-faire (24 percent prevalence, n 5 170), Autocratic
SCHYNS_9781785367274_t.indd 215
10/11/2017 15:20
216
SCHYNS_9781785367274_t.indd 216
10/11/2017 15:20
Dedication
Intelligence
7.37
Helpful
7.65
7.44
7.53
Knowledgeable
Educated
Intelligent
7.93
8.03
Dedicated
Hardworking
Within-profile mean across all items
8.05
Motivated
Within-profile mean across all items
7.13
Clever
Within-profile mean across all items
6.81
Sincere
8.747
8.811
8.684
8.747
8.234
8.407
8.289
8.454
7.785
8.362
8.522
8.149
7.879
7.998
7.796
7.842
6.806
6.874
6.730
7.237
6.381
7.147
7.346
6.919
8.186
8.176
8.104
8.278
7.782
7.839
7.816
7.908
7.565
6.751
7.224
6.306
6.723
7.176
0.31
Autocratic
Within-profile item means
7.09
Understanding
0.24
Laissez-faire
8.415
Prototypical
3
Sensitivity
Overall item means
2
0.29
Item
1
Profile
Parameter estimates for leader profiles from latent profile analysis of ILT scale items
Latent profile membership
proportions
Factor
Table 9.4
6.462
6.330
6.426
6.630
6.248
6.276
6.224
6.262
6.228
5.313
5.579
5.116
5.244
0.16
Anti-prototypical
4
217
SCHYNS_9781785367274_t.indd 217
10/11/2017 15:20
7.35
7.19
Strong
Dynamic
4.96
4.65
3.90
3.63
6.02
Pushy
Manipulative
Conceited
Selfish
Loud
4.25
Male
Within-profile mean across all items
4.45
Masculine
Within-profile mean across all items
5.50
Domineering
Within-profile mean across all items
7.40
Energetic
3.476
3.364
3.588
3.504
5.665
1.979
2.116
2.982
3.572
4.707
7.988
7.973
7.956
8.046
3.085
3.036
3.134
3.619
4.870
2.465
2.700
3.547
3.828
4.201
6.681
6.628
6.621
6.795
5.803
5.618
5.988
6.494
7.235
5.396
5.882
6.571
6.773
7.106
7.768
7.541
7.901
7.861
5.089
5.091
5.088
5.576
5.925
5.039
5.167
5.702
5.762
5.861
5.940
5.932
6.295
6.205
Note: Model stability and identification for all models were addressed by using multiple sets of random starting values (500 sets for initial stage
optimization, 50 sets for final stage optimization), and maximum likelihood estimation with standard errors robust to non-normality was used to
estimate all models (Muthén & Muthén, 1998–2010).
Masculinity
Tyranny
Dynamism
218 Handbook of methods in leadership research
(31 percent prevalence, n 5 220), and Anti-prototypical (16 percent prevalence, n 5 113). Prototypical and Anti-prototypical profiles were expected
based on the work of Epitropaki and Martin (2004). The Prototypical
profile had the highest profile means for Sensitivity, Intelligence,
Dedication, Dynamism; the lowest profile means for Tyranny; and lower
than average means for Masculinity. The Anti-prototypical had the lowest
profile means on Sensitivity, Intelligence, Dedication, and Dynamism and
higher than average means on Tyranny and Masculinity. The Laissez-faire
profile was expected based on work in the area of transformational and
transactional leadership (Hinkin & Schriesheim, 2008) and was characterized by a pattern of average responses to all items. The Autocratic profile
displayed a pattern of lower than average means for Sensitivity; higher
than average means for Intelligence, Dedication, and Dynamism; and the
highest profile means on Tyranny and Masculinity. The Autocratic profile
was also expected in the literature as autocratic leaders score particularly
low on the factor of consideration, as identified by the Ohio State studies
(Judge, Piccolo, & Ilies, 2004).
Thus for our data, Prototypical, Laissez-faire, Autocratic, and Antiprototypical profiles of leader perceptions emerged. Interestingly, there
was no dominant profile of leader perceptions for university students,
supporting the idea that there is variability in people’s perceptions of
typical leaders. Given that transformational leadership theory dominates the leader development literature (Avolio, 2005), it is noteworthy
that a majority of participants did not perceive a typical leader to be
Prototypical. In contrast, the prevalence of the Anti-prototypical and
Autocratic profiles reinforces the need for more research to focus on the
“darker” side of leadership (Schyns & Schilling, 2013). The Laissez-faire
profile is consistent with theory characterizing this type of leader as generally failing to take responsibility for managing. Recent work on the role
of followers in the leadership process emphasizes both an active and a
passive dimension of followership (Carsten, Uhl-Bien, West, Patera, &
McGregor, 2010). Perhaps endorsing a Laissez-faire leader profile allows
followers to take a more active role in the coproduction of leadership
(Carsten & Uhl-Bien, 2012).
Although beyond the scope of our example, structural features can
be added to LPAs. For example, covariates can be added to explore
whether these variables predict latent profile membership. For example,
does leadership self-efficacy impact latent profile membership of leader
perceptions? Alternatively, LPA with multiple groups can be performed
to examine differences in the probability of profile membership as a function of gender. More interestingly, a distal outcome can be predicted from
latent profile membership. For example, do profiles of leader perceptions
SCHYNS_9781785367274_t.indd 218
10/11/2017 15:20
Person-oriented approaches to leadership ­
219
predict leader–member exchange relationships? Recently, Lanza, Tan, and
Bray (2013) have developed a flexible model-based approach to empirically derive and summarize the profile-dependent density functions of
distal outcomes with categorical, continuous, or count distributions.
STRENGTHS AND WEAKNESSES OF THE PERSONORIENTED APPROACH
We began with the variable-oriented approach, which according to
Bergman and Andersson (2010), has grown to be “almost synonymous
with the scientific approach” due to its dominance and flexibility (p. 155).
One of the greatest strengths of the variable-oriented approach is the objectivity it affords researchers due to what are typically clear and precise measurements and reliable, valid scales. In addition, its often rigorous control
of confounding variables allows for the ability to make strong causal inferences, to quantify the amount of variance in an outcome explained by a
particular variable (ibid.), and to assemble a parsimonious model (Foti &
Hauenstein, 2007) of relationships that can be generalized to a population.
However, the variable-oriented approach carries with it several disadvantages, many of which can be addressed using the person-oriented
approach. First, the variable-oriented approach assumes homogeneity of
the population. In other words, it is believed that variables under study
behave the same way across all individuals and systems to which results
are generalized. Furthermore, it is assumed that the interactions within
a system are linear. However, given the importance of context and the
myriad different interactions of variables occurring within a dynamic
system, this is most often not the case (Reitzle, 2013). One advantage
of the person-oriented approach is that it overcomes this issue because
it acknowledges that there are population subgroups characterized by
patterns of variables and their non-linear interactions, resulting in something entirely different than the simple sum of the variables (Bergman &
Andersson, 2010). This way of thinking is more closely aligned with the
conceptualizations and research questions of processes and individuals as
holistic systems. It captures the reciprocal nature of interactions between
a person and his or her environment, such that the concept under study
becomes a “person–environment system” (Magnusson & Stattin, 2006,
p. 425). In addition, the variables forming the patterns within the system
do not have to be at the same level. The pattern-oriented approach can
capture characteristics and dynamics of individuals, dyads, groups, and
organizations (Wiegand, Jeltsch, Hanski, & Grimm, 2013). A final advantage is that examining many different components of the process together
SCHYNS_9781785367274_t.indd 219
10/11/2017 15:20
220 Handbook of methods in leadership research
can result in a more complete understanding of the processes as well as
potential for greater predictive accuracy (MacDougall et al., 2014).
Like the variable-oriented approach, the person-oriented approach
is not without its shortcomings. First, given that the person-oriented
approaches are underutilized in the behavioral sciences, pattern-oriented
methods are not particularly well developed. However, this trend appears
to be changing, as recent growth in person-oriented approaches has driven
a parallel growth in advanced pattern-oriented methods. Furthermore,
of the existing methods, many are descriptive in nature, which makes
causal relationships harder to define. Also disadvantageous is the requirement for a large sample size (especially for the longitudinal methods) and
highly reliable measures, since the methods do not account for measurement error. Additionally, interpretation of the meaning of the resulting
patterns is based on the indicator/variable means within each profile (see
­Table 9.4). It should be noted, however, that this process is analogous
to the interpretation of factors in the variable-oriented factor analysis.
Our final note on disadvantages relates to replication of the pattern. The
difficulty or inability of replicating patterns across different samples is
often mentioned as a disadvantage in person-oriented research. However,
research by Foti and colleagues suggests otherwise (e.g., Bray, Foti,
Thompson, & Wills, 2014; Foti, Bray, Thompson, & Allgood, 2012; Foti
& Coyle, 2015; Foti & Thompson, 2015). Findings of their research show
that although the prevalence of particular patterns may vary depending on
the sample, the number and type of patterns remain stable and replicable.
It is important to note that although person- and variable-oriented
approaches differ in both theory and method, neither is necessarily
superior to the other. Rather, each should be seen as complementary
and meaningful in their own regard. As Von Eye et al. (2006) explain,
variable-oriented approaches are best suited for research questions aimed
at understanding general trends or relationships between variables in a
“well-specified population” (p. 1002). Person-oriented approaches are
appropriate for answering holistic questions about a person and his or
her development over time, as well as for providing explanation for individuals or groups of individuals who deviate from the aggregate average.
However, we advocate for the integration of both approaches in research
so as to offset their weakness and maximize their strengths. This enables
gathering of information about distinct subgroups while also allowing for
generalizations to be made across total samples.3
In discussing the strengths and weaknesses of both the variable- and
person-oriented approaches directly above, we hope to help readers
understand the value in including both approaches in their research. In
particular, we hope to convince leadership researchers that doing so could
SCHYNS_9781785367274_t.indd 220
10/11/2017 15:20
Person-oriented approaches to leadership ­
221
deepen an understanding of leadership processes by addressing questions
that variable-oriented methods cannot. The field of leadership has mastered the variable-oriented approach; almost all leadership research has
historically taken this perspective.
THE ROADMAP FORWARD
Although leadership researchers have uncovered a great deal of knowledge using the variable-oriented approach, we believe the field of leadership, in general, has not taken full advantage of the person-oriented
approach to answer important and difficult questions. Specifically, there
are three key areas within the leadership domain particularly suited to the
person-oriented research: leadership and time, leadership and context, and
leadership and non-linearity.
The role of time in leadership has largely been neglected in research,
as most has been primarily cross-sectional (Shamir, 2011). According to
David Day (2014), there are four areas within the leader and leadership
domain in need of closer temporal examination: (1) the effects of leader
behavior, (2) perceptions of leaders and emergent leadership, (3) the
development of dyadic leader–follower relationships, and (4) leader development. Given that the person-oriented approach is rooted in research
on individual development, an inherently longitudinal process (Bergman
& Magnusson, 1997), the person-oriented approach conceptually (and
methodologically) incorporates time. Thus, it is fit for addressing questions related to these four temporal issues. In this section, however, we
focus mainly on the fourth: leader development. Traditionally, leader
development research has focused on the growth of individual leadership
skills or competencies (Ruderman, Clerkin, & Connolly, 2014), and, with
only a few exceptions (e.g., Day & Sin, 2011; Mumford et al., 2000), has
been almost exclusively cross-sectional (Day, 2014). However, personoriented theory states that individual variables behave differently based on
the environment and other interacting variables in the system (Bergman
& Andersson, 2010). This is also the case with the skills and competencies involved in leader development. In addition, leader development is
inherently yoked with time because development implies change, which
can only happen over time (Day, 2014). Thus, conducting research using
a person-oriented approach, which captures complex interactions over
time, aligns leader development methodology with leader development
theory. In addition, it can be used to test longitudinal models of leader
development. For example, Murphy and Johnson (2011) proposed a
dynamic “lifetime development” model of leader development, which
SCHYNS_9781785367274_t.indd 221
10/11/2017 15:20
222 Handbook of methods in leadership research
claims that leadership development is driven by a complex and dynamic
interaction of early developmental factors, leader identity, self-regulation,
leader effectiveness and individual context, all occurring over time. The
person-oriented approach can bring researchers one step closer to testing
the patterns of variables and trajectories in these sorts of leader development models. Thus, one pathway on the roadmap forward for personoriented research and leadership is to address complex temporal aspects of
leaders and leadership that cannot be addressed from a variable-oriented
perspective.
Another avenue for advancement of leadership research is a better integration of context, and the person-oriented approach provides a means
to do so. The context in leadership research can refer to the environment,
the situation, the task, other individuals, or other dyads in a system. The
influence of context in leadership has consistently been found to be an
important one (Liden, Sparrowe, & Wayne, 1997), yet for the most part,
the context has largely been regarded as a separate entity, divorced from
the individual. When context is included in leadership studies, it is mostly
modeled using a person × interaction term, ignoring the idea that a person’s structure and dynamics are in part defined by his or her context
(Bergman & Magnusson, 1997). That is, the context and situations are
regarded as modifiers, not partners in the leadership process. Since context
is regarded as part of a holistic unit using the pattern-oriented approach,
there is no need to specify the context, as in the variable-oriented research.
Furthermore, even when the context is identified, it is unlikely that it is
soundly measured and properly defined (Bogat, 2009). In addition, the
importance of both time and context can be highlighted with the notion
of trajectories of classes over time. That is, groups may interact differently
within different environments, resulting in context-dependent group trajectories. The pattern-oriented approach can also address this issue. Thus,
a second pathway on the roadmap forward for person-oriented research
and leadership is by focusing on the many different forms of context: situation, task, actors, and dyads.
Finally, the concept of non-linearity of interactions is a very important
one when studying systems. As discussed above, non-linearity refers to
the idea that two variables in a system take on different forms depending on the other components in the system (Magnusson & Stattin, 2006).
Bauer and Shanahan (2007) conducted the only test of non-linearity using
person-oriented and variable-oriented approaches. Results from their
study showed that creating a three-way interaction using variable-oriented
methods was not the same as studying the non-linearity of system interactions. Although their example was related to child academic competence
and dropout, not leadership, these findings underscore the importance
SCHYNS_9781785367274_t.indd 222
10/11/2017 15:20
Person-oriented approaches to leadership ­
223
of the person-oriented approach to study the complexity of interactions
within social process, including leadership. Thus, our third pathway on the
roadmap forward for person-oriented research and leadership is to realize
the usefulness of non-linear interactions in leadership processes.
Though we have provided some recommended foci for advancing
leadership research with the pattern-oriented approach, we conclude
this chapter with two novel examples of future research ideas using the
person-oriented approach. The first example relates to leadership perceptions. Although researchers have previously used the person-oriented
approach to study perceptions of leadership (e.g., Foti et al., 2012), most
has been conducted exclusively at the individual level. That is, individuals’
perceptions of a leader target are typically clustered to identify subgroups
of individuals. Given the growth in studying leadership processes at the
dyad level (e.g., Balkundi & Kilduff, 2006) as well as recent attention
to dyadic data analysis methods (see Yammarino & Gooty, Chapter 11
this volume), one interesting and novel idea is to use the person-oriented
approach to study perceptions of dyad partners. For example, one might
measure and subsequently identify subgroups of dyadic perceptions of
both leaders and followers, then measure their relations to other interpersonal variables, such as communication, conflict, or LMX. This would
facilitate a deeper understanding of the dynamic interactions underlying
the process of claiming and granting, proposed by DeRue and Ashford
(2010).
The second example also takes a different perspective on the personoriented approach. Most of what we have discussed up to this point
has focused on using the pattern-oriented methods for discovery, or to
uncover/explore relationships between variables or constructs. However,
pattern-oriented methods can also be used from a measurement perspective. In other words, they can be used to develop new scales of measurement. The idea behind this is to use the pattern-oriented methods to
identify which manifest variables are best for measuring a particular
construct. First, many different possible variables are measured, then they
are classified into profile, and finally, those variables that increase the
variance within a profile (homogeneity) and/or decrease variance between
profiles are dropped. The result is a new set of variables useful for measuring the construct of interest. One example of where this technique could be
useful is in measuring leadership perceptions. Perhaps the person-oriented
approach could provide new ways to measure perceptions of leaders by,
for example, incorporating goal attainment, leader emotional expressions, follower emotional expression, and events. These could, in turn, be
­compared to those scales used to date.
SCHYNS_9781785367274_t.indd 223
10/11/2017 15:20
224 Handbook of methods in leadership research
CONCLUSION
We believe the person-oriented approach holds great promise for understanding leadership processes. By moving away from a primary focus on
prediction of outcomes and percentage variance explained to a focus on
examining how variables co-occur both relationally and over time, greater
understanding in leadership can be achieved.
NOTES
1. Note: In person-oriented research, the acronym “CFA” is used to abbreviate configural
frequency analysis and is not to be confused with the “CFA” of confirmatory factor
analysis, a variable-oriented technique used to test the underlying factor structure of a
set of observed variables.
2. Note: Given the large number of significance tests involved in CFA, it is recommended
researchers make adjustments to the alpha level to control for Type I errors (e.g.,
Bonferroni adjustment; Von Eye, 2002).
3. Note: We advise researchers conducting variable-oriented and person-oriented analysis simultaneously to select variables with strong underlying factor structure (Foti &
Thompson, 2016).
REFERENCES
Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on
Automatic Control, 19(6), 716–723.
Avolio, B.J. (2005). Leadership development in balance: Made/born. Mahwah, NJ: Lawrence
Erlbaum Associates.
Balkundi, P., & Kilduff, M. (2006). The ties that lead: A social network approach to leadership. The Leadership Quarterly, 17(4), 419–439.
Bass, B.M. (1985). Leadership and performance beyond expectations. New York: Free Press.
Bauer, D.J., & Curran, P.J. (2003). Distributional assumptions of growth mixture models:
Implications for overextraction of latent trajectory classes. Psychological Methods, 8(3),
338–363.
Bauer, D.J., & Shanahan, M.J. (2007). Modeling complex interactions: Person-centered and
variable-centered approaches. In T.D. Little, J.A. Bovaird, & N.A. Card (Eds.), Modeling
contextual effects in longitudinal studies. Mahwah, NJ: Lawrence Erlbaum.
Bergman, L.R., & Andersson, H. (2010). The person and the variable in developmental psychology. Journal of Psychology, 218(3), 155–165.
Bergman, L.R., & Magnusson, D. (1997). A person-oriented approach in research on developmental psychopathology. Development and Psychopathology, 9(2), 291–319.
Bergman, L.R., & Trost, K. (2006). The person-oriented versus the variable-oriented
approach: Are they complementary, opposites, or exploring different worlds? MerrillPalmer Quarterly. Special Issue: Person-Centered and Variable-Centered Approaches to
Longitudinal Data, 52(3), 601–632.
Bergman, L.R., & Vargha, A. (2013). Matching method to problem: A developmental
science perspective. European Journal of Developmental Psychology, 10(1), 9–28.
Bergman, L.R., & Wångby, M. (2014). The person-oriented approach: A short theoretical
and practical guide. Eesti Haridusteaduste Ajakiri, 2(1), 29–49.
SCHYNS_9781785367274_t.indd 224
10/11/2017 15:20
Person-oriented approaches to leadership ­
225
Bergman, L.R., Magnusson, D., & El-Khouri, B.M. (2000). Studying individual development
in an interindividual context: A person-oriented approach. Mahwah, NJ: Erlbaum.
Bergman, L.R., Nurmi, J.-E., & Von Eye, AA. (2012). I-states-as-objects-analysis (ISOA):
Extensions of an approach to studying short-term developmental processes by analyzing
typical patterns. International Journal of Behavioral Development, 36(3), 237–246.
Block, J. (1971). Lives through time. Berkeley, CA: Bancroft Books.
Bogat, G.A. (2009). Is it necessary to discuss person-oriented research in community psychology? American Journal of Community Psychology, 43(1–2), 22–34.
Boyd, K., & Foti, R.J. (April, 2016). Measuring the effects of contextual constraints on perceptions of leadership. Paper presented at the annual meeting of the Society for Industrial
and Organizational Psychology, Anaheim, CA.
Bray, B.C., Foti, R.J., Thompson, N.J., & Wills, S.F. (2014). Disentangling the effects of
self leader perceptions and ideal leader prototypes on leader judgments using loglinear
modeling with latent variables. Human Performance, 27(5), 393–415.
Carsten, M.K., & Uhl-Bien, M. (2012). Follower beliefs in the co-production of leadership.
Zeitschrift für Psychologie, 220(4), 210–220.
Carsten, M.K., Uhl-Bien, M., West, B.J., Patera, J.L., & McGregor, R. (2010). Exploring
social constructions of followership: A qualitative study. The Leadership Quarterly, 21(3),
543–562.
Collins, L.M., & Lanza, S.T. (2010). Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. New York: Wiley.
Day, D.V. (2014). Introduction: Leadership and organizations. In D.V. Day (Ed.), The
Oxford handbook of leadership and organizations (pp. 3–12). New York: Oxford University
Press.
Day, D.V., & Sin, H.P. (2011). Longitudinal tests of an integrative model of leader development: Charting and understanding developmental trajectories. The Leadership Quarterly,
22(3), 545–560.
Day, D.V., Fleenor, J.W., Atwater, L.E., Sturm, R.E., & McKee, R.A. (2014). Advances
in leader and leadership development: A review of 25 years of research and theory. The
Leadership Quarterly, 25(1), 63–82.
DeRue, D.S., & Ashford, S.J. (2010). Who will lead and who will follow? A social process of
leadership identity construction in organizations. Academy of Management Review, 35(4),
627–647.
Doucet, O., Fredette, M., Simard, G., & Tremblay, M. (2015). Leader profiles and their
effectiveness on employees’ outcomes. Human Performance, 28(3), 244–264.
Epitropaki, O., & Martin, R. (2004). Implicit leadership theories in applied settings: Factor
structure, generalizability, and stability over time. Journal of Applied Psychology, 89(2),
293–310.
Fairhurst, G.T., & Uhl-Bien, M. (2012). Organizational discourse analysis (ODA):
Examining leadership as a relational process. The Leadership Quarterly, 23(6), 1043–1062.
Foti, R.J., & Coyle, P.T. (2015). Patterns of leader and follower perceptions: How are they
related? Paper at the Implicit Followership Theories and People’s Dispositions as Drivers
of Leadership Effectiveness Symposium presented at the annual meeting of the Academy
of Management Conference, Vancouver, BC.
Foti, R.J., & Hauenstein, N.M.A. (2007). Pattern and variable approaches in leadership
emergence and effectiveness. Journal of Applied Psychology, 92(2), 347–355.
Foti, R.J., & Thompson, N.J. (2016). Judgments of leadership: Profiles of implicit theories
and personality. Paper at the Current Perspectives on Person-centered Leadership Research
Symposium at the annual meeting of the Society for Industrial and Organizational
Psychology, Anaheim, CA.
Foti, R.J., & Wills, S.F. (2015). Perceptions of female emergent leaders: Similarities and
differences. Paper at the Understanding Perceptions of Female and Male Leaders
Symposium at the annual meeting of the Academy of Management Conference,
Vancouver, BC.
Foti, R.J., Bray, B.C., Thompson, N.J., & Allgood, S.F. (2012). Know thy self, know thy
SCHYNS_9781785367274_t.indd 225
10/11/2017 15:20
226 Handbook of methods in leadership research
leader: Contributions of a pattern-oriented approach to examining leader perceptions. The
Leadership Quarterly, 23(4), 702–717.
Foti, R.J., Knee, R., & Backert, R.G. (2008). Multi-level implications of framing leadership
perceptions as a dynamic process. The Leadership Quarterly, 19(2), 178–194.
Foti, R.J., Thompson, N.J., & Allgood, S.F. (2011). The pattern oriented approach: A framework for the experience of work. Industrial and Organizational Psychology: Perspectives on
Science and Practice, 4(1), 122–125.
Gutiérrez-Peña, E. and Von Eye, A. (2000). A Bayesian approach to configural frequency
analysis. Journal of Mathematical Sociology, 24(2), 151–174.
Hinkin, T.R., & Schriesheim, C.A. (2008). An examination of “nonleadership”: From
laissez-faire leadership to leader reward omission and punishment omission. Journal of
Applied Psychology, 93(6), 1234–1248.
Judge, T.A., Colbert, A.E., & Ilies, R. (2004). Intelligence and leadership: A quantitative review and test of theoretical propositions. Journal of Applied Psychology, 89(3),
542–552.
Judge, T.A., Piccolo, R.F., & Ilies, R. (2004). The forgotten ones? The validity of consideration and initiating structure in leadership research. Journal of Applied Psychology, 89(1),
36–51.
Landy, F.J. & Conte, J.M. (2013). Work in the 21st century: An introduction to industrial and
organizational psychology. New York: McGraw-Hill.
Lanza, S.T., & Collins, L.M. (2006). A mixture model of discontinuous development in
heavy drinking from ages 18 to 30: The role of college enrollment. Journal of Studies on
Alcohol, 67(4), 552–561.
Lanza, S.T., Bray, B.C., & Collins, L.M. (2013). An introduction to latent class and latent
transition analysis. In J.A. Schinka, W.F. Velicer, & I.B. Weiner (Eds.), Handbook of psychology (2nd ed., Vol. 2, pp. 691–716). Hoboken, NJ: Wiley.
Lanza, S.T., Flaherty, B.P., & Collins, L.M. (2003). Latent class and latent transition
analysis. In J.A. Schinka & W.F. Velicer (Eds.), Handbook of psychology, Vol. 2: Research
methods in psychology (pp. 663–685). Hoboken, NJ: John Wiley & Sons.
Lanza, S.T., Tan, X., & Bray, B.C. (2013). Latent class analysis with distal outcomes: A flexible model-based approach. Structural Equation Modeling, 20(1), 1–26.
Lawrence, B.S., & Zyphur, M.J. (2011). Identifying organizational faultlines with latent class
cluster analysis. Organizational Research Methods, 14(1), 32–57.
Liden, R.C., Sparrowe, R.T., & Wayne, S.J. (1997). Leader–member exchange theory:
The past and potential for the future. Research in Personnel and Human Resources
Management, 15, 47–120.
Lord, R.G., Brown, D.J., Harvey, J.L., & Hall, R.J. (2001b). Contextual constraints on
prototype generation and their multilevel consequences for leadership perceptions. The
Leadership Quarterly, 12(3), 311–338.
Lord, R.G., Foti, R.J., & De Vader, C.L. (1984). A test of leadership categorization theory:
Internal structure, information processing, and leadership perceptions. Organizational
Behavior and Human Performance, 34(3), 343–378.
Lord, R.G., Foti, R.J., & Phillips, J.S. (1982). A theory of leadership categorization. In J.G.
Hunt, V. Sekaran, and C. Schriesheim (Eds.), Leadership: Beyond establishment views (pp.
104–121). Carbondale, IL: Southern Illinois University Press.
MacDougall, A.E., Bauer, J.E., Novicevic, M.M., & Buckley, M.R. (2014). Toward the
pattern-oriented approach to research in human resources management: A review of configurational and category theorizing, methods and applications. Research in Personnel and
Human Resources Management, 32, 177–240.
Magnusson, D. (1988). Individual development from an interactional perspective: A longitudinal study. Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Magnusson, D. (1995). Individual development: An integrated model. In P. Moen, G.H.
Elder Jr., & K. Luscher (Eds.), Examining lives in context. Perspectives on the ecology of
human development (pp. 19–60). Washington, DC: APA.
Magnusson, D., & Stattin, H. (2006). The person in context: A holistic-interactionistic
SCHYNS_9781785367274_t.indd 226
10/11/2017 15:20
Person-oriented approaches to leadership ­
227
approach. In R.M. Lerner & W. Damon (Eds.), Handbook of child psychology: Vol. 1.
Theoretical models of human development (6th ed., pp. 404–464). Hoboken, NJ: Wiley.
McLachlan, G.J., & Peel, D. (2000). Finite mixture models. New York: Wiley.
Meyer, J.P., Stanley, L.J., & Vandenberg, R.J. (2013). A person-centered approach to the
study of commitment. Human Resource Management Review, 23(2), 190–202.
Mumford, M.D., Zaccaro, S.J., Johnson, J.F., Diana, M., Gilbert, J.A., & Threlfall, K.V.
(2000). Patterns of leader characteristics: Implications for performance and development.
The Leadership Quarterly, 11(1), 115–133.
Murphy, S.E., & Johnson, S.K. (2011). The benefits of a long-lens approach to leader development: Understanding the seeds of leadership. The Leadership Quarterly, 22(3), 459–470.
Muthén, B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In D. Kaplan (Ed.), Handbook of quantitative methodology
for the social sciences (pp. 345–368). Thousand Oaks, CA: Sage.
Muthén, L.K., & Muthén, B.O. (1998–2010). Mplus user’s guide (6th ed.). Los Angeles, CA:
Muthén & Muthén.
O’Shea, P.G., Foti, R.J., Hauenstein, N.M.A., & Bycio, P. (2009). Are the best leaders
both transformational and transactional? A pattern-oriented analysis. Leadership, 5(2),
237–259.
Reitzle, M. (2013). Introduction: Doubts and insights concerning variable- and person-­
oriented approaches to human development. European Journal of Developmental
Psychology, 10(1), 1–8.
Ruderman, M.N., Clerkin, C., & Connolly, C. (2014). Leadership development beyond
competencies: Moving to a holistic approach (White Paper). The Center for Creative
Leadership.
Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics, 6(2), 461–464.
Schyns, B., & Schilling, J. (2013). How bad are the effects of bad leaders? A meta-analysis of
destructive leadership and its outcomes. The Leadership Quarterly, 24(1), 138–158.
Sclove, S.L. (1987). Application of model-selection criteria to some problems in multivariate
analysis. Psychometrika, 52(3), 333–343.
Shamir, B. (2011). Leadership takes time: Some implications of (not) taking time seriously in
leadership research. The Leadership Quarterly, 22(2), 307–315.
Shondrick, S.J., & Lord, R.G. (2010). Implicit leadership and followership theories:
Dynamic structures for leadership perceptions, memory, and leader–follower processes.
In G. Hodgkinson, & J. Ford (Eds.), International review of industrial and organizational
psychology (Vol. 25, pp. 1–33). Chichester, UK: John Wiley & Sons.
Sterba, S.K., & Bauer, D.J. (2010). Matching method with theory in person-oriented developmental psychopathology research. Development and Psychopathology, 22(2), 239–254.
Sy, T., Shore, L.M., Strauss, J., Shore, T.H., Tram, S., Whiteley, P., & Ikeda-Muromachi,
K. (2010). Leadership perceptions as a function of race–occupation fit: The case of Asian
Americans. Journal of Applied Psychology, 95(5), 902–919.
Vandenberg, R.J., & Stanley, L.J. (2009). Statistical and methodological challenges for commitment researchers: Issues of invariance, change across time, and profile differences. In
H.J. Klein, T.E. Becker, & J.P. Meyer (Eds.), Commitment in organizations: Accumulated
wisdom and new directions (pp. 383–416). Florence, KY: Routledge/Taylor and Francis
Group.
Von Eye, A. (2002). Configural frequency analysis – Methods, models, and applications.
Mahwah, NJ: Erlbaum.
Von Eye, A. (2010). Developing the person-oriented approach: Theory and methods of
analysis. Development and Psychopathology, 22(2), 227–285.
Von Eye, A., & Bergman, L.R. (2003). Research strategies in developmental psychopathology: Dimensional identity and the person-oriented approach. Development and
Psychopathology, 15(3), 553–580.
Von Eye, A., & Bogat, G.A. (2006). Person orientation – concepts, results, and development.
Merrill Palmer Quarterly, 52(3), 390–420.
Von Eye, A., Bogat, G.A., & Rhodes, J.E. (2006). Variable-oriented and person-oriented
SCHYNS_9781785367274_t.indd 227
10/11/2017 15:20
228 Handbook of methods in leadership research
perspectives of analysis: The example of alcohol consumption in adolescence. Journal of
Adolescent Research, 29(6), 981–1004.
Von Eye, A., Mun, E.Y., & Bogat, G.A. (2008). Temporal patterns of variable relationships in person-oriented research: Longitudinal models of configural frequency analysis.
Developmental Psychology, 44(2), 437–445.
Wang, M., & Bodner, T.E. (2007). Growth mixture modeling: Identifying and predicting
unobserved subpopulations with longitudinal data. Organizational Research Methods,
10(4), 635–656.
Wang, M., & Hanges, P.J. (2011). Latent class procedures: Applications to organizational
research. Organizational Research Methods, 14(1), 24–31.
Ward, J.H. (1963). Hierarchical grouping to optimize an objective function. Journal of the
American Statistical Association, 58(301), 236–244.
Wiegand, T., Jeltsch, F., Hanski, I., & Grimm, V. (2003). Using pattern-oriented modeling
for revealing hidden information: A key for reconciling ecological theory and application.
Oikos, 100(2), 209–222.
Wright, A.G.C., & Hallquist, M.N. (2014). Mixture modeling methods for the assessment
of normal and abnormal personality, part II: Longitudinal models. Journal of Personality
Assessment, 96(3), 269–282.
SCHYNS_9781785367274_t.indd 228
10/11/2017 15:20
10. Multi-level issues and dyads in leadership
research
Francis J. Yammarino and Janaki Gooty
Multiple levels of analysis in theory building and theory testing are critical
and have a long history in leadership research, as evidenced in works by
Dansereau, Alutto, and Yammarino (1984); Dansereau and Yammarino
(1998a, 1998b); DeChurch, Hiller, Murase, Doty, and Salas (2010);
Dionne and Dionne (2008); Dionne et al. (2014); Gooty, Serban, Thomas,
Gavin, and Yammarino (2012); Gooty and Yammarino (2011, 2016);
Markham (2010, 2012); Schriesheim, Castro, Zhou, and Yammarino
(2001), Yammarino and Dansereau (2009, 2011), and Yammarino,
Dionne, Chun, and Dansereau (2005). These authors, among others, have
noted the importance of clearly specifying the levels of analysis at which
phenomena are expected theoretically, and ensuring the measurement of
constructs and data analytic techniques correspond to the asserted levels
of analysis, so that inference drawing is not misleading.
In this chapter, we explicate a set of key multi-level issues, both theory
and method related, for leadership research. We then focus on the most
neglected and poorly understood level of analysis in leadership – dyads –
and develop a set of key issues related to three methodological dyadic
approaches in the leadership field. Finally, we provide some recommendations involving multi-level issues in theory and methods for leadership
researchers.
MULTI-LEVEL ISSUES IN LEADERSHIP
Levels of analysis are inherent in theoretical formulations in leadership
research. They are implicit or assumed, can be explicitly incorporated,
and are used to develop the boundary conditions under which a theory is
expected to hold. Understanding how and if levels are specified permits
an examination of the potential or degree of prevalence of theoretical misspecification. Moreover, identification of relevant levels-of-analysis issues
may help account for mixed, inconsistent, and contradictory findings in
prior leadership research. Without explicit incorporation of levels-ofanalysis issues, incomplete understanding of a construct or phenomenon
229
SCHYNS_9781785367274_t.indd 229
10/11/2017 15:20
230 Handbook of methods in leadership research
may lead to faulty measures, inappropriate data analytic techniques, and
erroneous conclusions. As such, levels of analysis must also be accounted
for methodologically in leadership research. In brief, “[t]heory without
levels of analysis is incomplete; data without levels of analysis is incomprehensible” (Yammarino et al., 2005, p. 904).
Four Levels in Leadership
The phrase “levels of analysis” refers to the entities, units, or objects of
observation (e.g., Miller, 1978). Typically, hierarchically ordered, lowerlevel entities such as persons are nested or embedded in higher-level entities such as dyads or groups. In leadership research, we are interested
in people in work organizations in terms of four key levels of analysis
(e.g., Dansereau et al., 1984; Yammarino & Dansereau, 2009, 2011;
Yammarino et al., 2005). First, a focus on the level of independent individuals or persons (e.g., leaders or followers) allows for the exploration
of individual differences. Second, dyads, consisting of two-person groups
with interpersonal relationships such as leader–follower roles, involve
one-to-one interdependence between individuals. Third, groups, including
workgroups and teams, are a collection of individuals who are interdependent and interact on a face-to-face or “virtual” (i.e., non-colocated)
basis with one another. Fourth, collectives are clusterings of individuals
that are larger than groups and whose members are interdependent based
on a hierarchical structuring or a set of common or shared expectations.
Collectives can include groups of groups, departments, functional areas,
strategic business units, organizations, firms, and industries.
These four levels of analysis constitute different lenses for the examination, both conceptually and empirically, of people in organizations. Due
to the hierarchical and nested structure of levels, viewing people from
increasingly higher levels of analysis necessarily means the number of
entities decreases (e.g., there are fewer groups than persons in an organization) and the size of the entity increases (e.g., there are a larger number of
people in collectives than in groups).
Three Alternatives per Level
While some prefer to assert only two views for each level (e.g., groups
and group effects are viable or they are not), there are others who prefer
to consider three alternatives for each level of analysis (see, for example,
Klein, Dansereau, & Hall, 1994). Dansereau et al. (1984) and Yammarino
and Dansereau (2009, 2011), among others, distinguish conceptually
between two different (relevant) views of any level of analysis – wholes and
SCHYNS_9781785367274_t.indd 230
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
231
parts. Klein et al. (1994) call these homogeneous and heterogeneous views,
respectively. The third view, called independent, indicates that the focal
entities (e.g., groups) are not relevant and other entities, often lower-level
ones (e.g., individuals), are plausible (see Dansereau et al., 1984; Klein et
al., 1994).
A wholes view is defined as a focus between entities but not within them.
Differences between entities are viewed as valid, and differences within
entities are viewed as random (error). This perspective can be viewed as
a between-units case in which: (1) members of a unit are homogeneous
(Klein et al., 1994); (2) the whole unit is of importance, entities display
similarity among members; and (3) relationships among members of units
are positive with respect to constructs of a theory and a function of differences between units.
A parts view is defined as a focus within entities but not between them.
Differences within entities are treated as valid, and differences between
entities are considered to be random (error). This perspective can be
viewed as a within-units case, also known as a “frog pond” effect, in
which: (1) members of a unit are heterogeneous (Klein et al., 1994); (2)
a member’s position relative to other members is of importance, entities
display complementarity among members; and (3) relationships among
members of units are negative with respect to constructs of a theory and
a function of differences within units. The mechanisms that hold together
parts are interdependence and behavioral integration (typically at lower
levels) or functional integration and social structure (often at higher levels)
(Dansereau et al., 1984; Miller, 1978; Yammarino & Dansereau, 2009,
2011).
Various authors indicate that effects also may not be evidenced at a
specific focal level (Dansereau et al., 1984; Klein et al., 1994; Miller, 1978;
Yammarino & Dansereau, 2009, 2011). In this case of independence, two
possible conclusions for a focal level are either a focus both between and
within entities (sometimes called equivocal), or error between and within
entities (sometimes called inexplicable or null). In both of these cases, the
focal level of analysis does not clarify understanding of the constructs, variables, or phenomena of interest, and other levels must be considered. The
members of a unit are (1) independent, (2) free of the unit’s influence, and
(3) relationships among members of units are independent with respect
to constructs of a theory and a function of differences between members
(e.g., persons) independent of higher-level units (e.g., groups).
In leadership research, when “person” is the focal level of analysis, the
person (wholes) and the interdependent genes, properties, or behaviors
over time (parts) within the person are the potential units of analysis. If the
entities are independent at this level, potential higher levels are the dyad or
SCHYNS_9781785367274_t.indd 231
10/11/2017 15:20
232 Handbook of methods in leadership research
group. For example, cognitive ability is considered relatively stable and not
expected to change at multiple points in time in persons, ensuring independence between persons (a wholes view), whereas state affect, for example, for
persons is transient and multiple measurements of this could change or shift
over time for persons (a parts view). In leadership, wholes, a homogeneous perspective, can also imply a focus on “stability,” such as person (e.g.,
leader) traits that remain constant over time. Parts, a heterogeneous perspective, can also imply a focus on “change,” such as person (e.g., leader)
characteristics that shift over time (see Dansereau et al., 1984; Dansereau,
Yammarino, & Kohles, 1999; Yammarino & Dansereau, 2009, 2011).
When “group” is the focal level of analysis, the group (wholes) and the
interdependent persons or dyads (parts) within the group are the potential
units of analysis. For example, the presence of a common leader across
multiple dyads within the same group creates a potential lower-level
(dyad-level) dependency (a parts view). If the entities are independent at
this level (for example, when considering team-level diversity, we do not
expect potential dyad-level dependencies), there are differences among
teams (a wholes view), and potential higher levels are collectives such as
the organization, strategic group, or industry.
Specific, Emergent, and Cross Levels
In leadership research, assuming only one level of analysis or choosing
only one level without consideration of other levels can either mask effects
or indicate effects when none exist (e.g., Dansereau et al., 1984; Hackman,
2003; Yammarino et al., 2005; Yammarino & Dansereau, 2009, 2011).
Moreover, levels in organizational settings rarely exist independently of
one another, and are typically nested, cross-classified, or somehow linked
(e.g., Gooty & Yammarino, 2011, 2016). These linkages necessitate that
multiple levels and their effects be considered simultaneously.
In leadership, micro- versus macro-theories, processes, and concepts can
be either contrasted or viewed as analogous, depending on various levels-ofanalysis formulations. Adjacent levels can interact in alignment, misalignment, or opposition to one another, and can be intertwined via complex
processes (e.g., Klein & Kozlowski, 2000; Miller, 1978). The focus here for
“linking” (multiple) levels of analysis is on three general types of multiplelevel formulations – level-specific, emergent, and cross-level f­ormulations
– from both wholes and parts perspectives for leadership work.
Level-specific relationships
First, relationships among constructs may be hypothesized to hold at
a lower level (e.g., person) but not at a higher level (e.g., group). This
SCHYNS_9781785367274_t.indd 232
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
233
possibility is discussed as a discontinuity thesis (Miller, 1978), as levelspecific formulations (Dansereau et al., 1984; Miller, 1978), or empirically
as disaggregated, individual, or level-specific effects (Pedhazur, 1982;
Robinson, 1950). In these cases, the higher level of analysis is not relevant
for understanding the theoretical constructs.
Wholes at a lower level may not always emerge at a higher level (independent). This level-specific wholes formulation means that members are
homogeneous with respect to the constructs of interest in all lower-level
entities (e.g., groups), and higher-level entities (e.g., collectives) are not relevant. Examples of level-specific wholes formulations include various personality and leadership approaches at the individual level (see Dionne et
al., 2014; Yammarino & Dansereau, 2009; Yammarino et al., 2005); many
team-based and shared leadership approaches at the group level (Dionne
et al., 2014; Klein & Kozlowski 2000; Yammarino & Dansereau, 2009;
Yammarino et al., 2005); and organizational missions and visions and
strategic leadership at the organization level (Dansereau & Yammarino,
1998a, 1998b; Yammarino et al., 2005; Yammarino & Dansereau, 2009).
Parts at a lower level may not always emerge at a higher level (independent). This level-specific parts formulation means that members are
heterogeneous with respect to the constructs of interest in all lower-level
entities (e.g., groups), and higher-level entities (e.g., collectives) are not
relevant. Examples of level-specific parts formulations include newer
approaches that consider personality and leadership changes across time
(see Yammarino & Dansereau, 2009) and vertical dyad linkage (VDL) and
leader–member exchange (LMX in- and out-group) leadership approaches
at the group level (Dansereau & Yammarino, 1998a, 1998b; Yammarino
et al., 2005).
Emergent relationships
Second, relationships among constructs may not be asserted at a lower
level but are hypothesized to manifest themselves at a higher level of
analysis. This possibility is also discussed as a type of discontinuity thesis
(Miller, 1978), as emergent formulations that hold at a higher level (e.g.,
group) after not being asserted or found to hold at a lower level (e.g.,
person) (Dansereau et al., 1984; Miller, 1978), empirically as higher-level
effects that do not disaggregate, or as emergent effects (Miller, 1978;
Robinson, 1950). In these cases, the lower level of analysis is not relevant
for understanding the theoretical constructs.
For an emergent wholes formulation, constructs are expected to hold
at a higher (e.g., group) level where members of a higher-level entity
are homogeneous with respect to the constructs after not having been
expected or observed at a lower level (independent). Examples of ­emergent
SCHYNS_9781785367274_t.indd 233
10/11/2017 15:20
234 Handbook of methods in leadership research
wholes formulations include concepts in relation to leadership such
as group cohesion, shared mental models, as well as shared and team
leadership at the team level (DeChurch et al., 2010; Klein & Kozlowski,
2000; Yammarino et al., 2005) and several of the Global Leadership
and Organizational Behavior Effectiveness research program (GLOBE)
cultural values dimensions at the society level (House, Hanges, Javidan,
Dorfman, & Gupta 2004).
For an emergent parts formulation, constructs are expected to hold at
a higher (e.g., group) level where members are heterogeneous with respect
to the constructs after not having been expected or observed at a lower
level (independent). Examples of emergent parts formulations include
concepts such as vertical dyad linkage, perhaps leader–member exchange
(see Dansereau & Yammarino, 1998a, 1998b), intra-group conflict, and
compatible mental models at the team level (Klein & Kozlowski, 2000),
and aspects of organizational subcultures at the organization level (Katz
& Kahn, 1978).
Cross-level relationships
Third, relationships among constructs also may be hypothesized to hold
at higher (e.g., collective) and lower (e.g., group) levels of analysis. This
possibility is discussed as a homology thesis (Miller, 1978), empirically
as aggregated or ecological effects (Pedhazur, 1982; Robinson, 1950),
and, in the traditional sense, as cross-level explanations (Behling, 1978;
Dansereau et al., 1984; Miller, 1978; Yammarino & Dansereau, 2009,
2011). Note the use here is contrary to how the term “cross-level” is used in
much contemporary work (e.g., Bryk & Raudenbush, 1992; Raudenbush
& Bryk, 2002; Rousseau, 1985) as the effect of a higher-level variable on
a lower-level association (cross-level moderation) or variable (cross-level
direct effect). Our use of “cross-level” is in line with older cited traditional
work as well as in the “hard” and physical sciences (e.g., Behling, 1978;
Dansereau et al., 1984; Miller, 1978; Robinson, 1950), and are statements
about relationships among variables that are likely to hold equally well
at a number of levels of analysis (e.g., X and Y are positively related for
persons and for groups). Such cross-level formulations and effects specify
patterns of relationships replicated across levels of analysis. Models of this
type are uniquely powerful and parsimonious because the same effect is
manifested at more than one level of analysis (e.g., E 5 mc2, which holds
at multiple levels of analysis, is a cross-level formulation for us and in traditional and “hard” sciences work; in contrast, most contemporary work
in leadership would not term this equation as a cross-level effect).
Wholes at a lower level also can aggregate or manifest themselves as
wholes at a higher level. This cross-level wholes formulation means that
SCHYNS_9781785367274_t.indd 234
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
235
members are homogeneous with respect to the constructs of interest in
all entities (e.g., groups and collectives) at both levels of analysis, but the
entities (e.g., groups and collectives) differ from one another. Examples of
cross-level wholes formulations include various GLOBE cultural dimensions that are consistent from the individual to organization to society
levels (House et al., 2004), and professional and functional titles and
associated expectations about them that hold from the individual to dyad
to group to organization levels (see Dansereau et al., 1984; Yammarino &
Dansereau, 2009).
Wholes at a lower level can aggregate or manifest themselves as parts
at a higher level. This cross-level parts formulation means that members
are homogeneous with respect to the constructs of interest in all the
lower-level entities (e.g., groups), and these differ from one another; in
all higher-level entities (e.g., collectives), however, there is heterogeneity
because members within the entities differ from one another. Examples of
cross-level parts formulations include functional area task differences and
subsystem subcultures from the individual to organization and group to
organization levels (Katz & Kahn, 1978).
Time and Levels
Another important multi-level issue for leadership research is the theoretical specification and empirical test of potentially changing variables and
phenomena (e.g., traits, properties) and shifting levels of analysis (entities)
over time (see Dansereau et al., 1999; Yammarino & Dansereau, 2011).
Leadership must be viewed from a longitudinal and dynamic perspective to fully understand relevant phenomena that are stable or shifting,
changing, and developing over time. While much has been written about
changing variables or constructs over time and strategies for analysing
these changes (e.g., Chan, 1998; Yammarino et al., 2005), relatively little
has been written about changing entities (levels of analysis) over time (for
exceptions, see Dansereau et al., 1984, 1999; Yammarino & Dansereau,
2009, 2011).
In their approach, for example, Dansereau et al. (1999) consider the
plausible units of analysis (wholes, parts, independence, and null) at two
points in time to account for entity changes over time. A 16-cell (4 × 4)
matrix results for understanding changes or stability in levels of analysis
over time. For various levels of analysis, Dansereau et al. (1999) define,
describe, and illustrate the following components: four types of stable conditions (in terms of wholes, parts, lower-level independent units, and null
over time); three types of changes that move the focus up from a lower to a
higher level (transformation up from parts to wholes, level change up from
SCHYNS_9781785367274_t.indd 235
10/11/2017 15:20
236 Handbook of methods in leadership research
independent units to wholes, and level change up from independent units
to parts); three types of changes that move the focus down from a higher
to a lower level (transformation down from wholes to parts, level change
down from wholes to independent units, and level change down from parts
to independent units); and six types of changes that indicate the beginning
or end of a level (three “emergents” from null or nothing to wholes, parts,
or independent units, and three “ends” from wholes, parts, or independent
units to null or nothing).
As an example, a portion of Klein and Kozlowski’s (2000) work can
be cast in terms of the Dansereau et al. (1999) framework. Klein and
Kozlowski (2000) offer two forms for the emergence or change of entities over time – composition and compilation. Composition, based on
isomorphism, suggests that lower-level entities (e.g., persons), based on
shared mental models and similar information and expertise, combine in
a linear or pooled fashion with stable or uniform interactions over time,
resulting in higher-level entities (e.g., teams) viewed from a homogeneous
perspective. In this case, lower-level wholes shift to higher-level wholes
over time. Compilation, based on discontinuity, suggests that lower-level
entities (e.g., persons), based on compatible mental models and diverse
information and expertise, combine in a non-linear or adaptive fashion
with irregular or non-uniform interactions over time, resulting in higherlevel entities (e.g., teams) viewed from a heterogeneous perspective. In this
case, lower-level wholes shift to higher-level parts over time.
Fallacies, False Dichotomies, and Levels
In leadership research, it is critical to avoid committing a “fallacy of
the wrong level” (Dansereau, Cho, & Yammarino, 2006; Dansereau &
Yammarino, 2006). The most common form of these fallacies involves
inferring associations at a level other than where data are collected or
analyses are conducted. An ecological fallacy occurs when lower levels of
analysis are presumed to be mere disaggregations of higher-level entities.
In contrast, an individual fallacy occurs when higher levels of analysis
are presumed to be mere aggregations of lower-level entities. Work by
Robinson (1950) and others (e.g., Dansereau et al., 1984, 2006; Miller,
1978; Rousseau, 1985) highlight the problems and issues associated with
such fallacies.
While Dansereau et al. (1984) and Yammarino and Dansereau (2009,
2011) note that similarities may be apparent between micro-elements at
lower levels and macro-elements at higher levels, they also make clear that
these elements are not identical. Persons cannot be viewed as just disaggregated groups or organizations, and groups or organizations cannot
SCHYNS_9781785367274_t.indd 236
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
237
be seen as just aggregated persons. Processes, constructs, relationships,
theories, and other aspects of these entities can vary by level of analysis,
such that some can emerge or hold at one level and not at other levels (e.g.,
Gooty & Yammarino, 2011, 2016; Klein & Kozlowski, 2000; Morgeson &
Hoffman, 1999).
Dansereau and Yammarino and colleagues (Dansereau et al., 1984,
2006; Dansereau & Yammarino, 2006; Markham, 2010; Yammarino &
Dansereau, 2009; Yammarino et al., 2005) elaborate on various aspects
of fallacies of the wrong level for leadership research. They also offer a
variety of potential solutions to, and ways to avoid, such ecological and
individual fallacies in theory and hypothesis formulation, measurement,
data analysis, and inference-drawing procedures.
Relatedly, various Aristotelian “either-or” controversies, which have
existed for several millennia, can be expressed as, and often degrade to,
levels-related false dichotomies and inappropriate dualisms. In terms
of multi-level issues in leadership research, these dichotomies typically
relate to some version of “the individual versus the environment” such as
the “person–situation” and “genes–learning/development” debates. The
goal is to resolve Aristotelian “either-or” controversies with a “Galilean
approach” characterized by integration of the components of each duality.
In leadership, there is a long history related to the “person–situation”
debate (e.g., Bass, 2008; Dansereau et al., 1999; Yammarino & Dansereau,
2009, 2011). This discussion can also be framed in terms of an “individual–
environment” debate (e.g., are leaders born or made?), or the micro–macro
distinction (e.g., is leadership in or of the organization?), where an integration can occur to resolve the debate by viewing higher levels of analysis as
the context (situation or environment) for or boundaries on lower levels of
analysis (Dansereau et al., 1984, 1999; Yammarino & Dansereau, 2009).
For example, rather than assuming person (lower-level) or situation
(higher-level) explanations of effects, it is plausible to offer a person-bysituation or interactional explanation for leader behavior. Likewise, it is
the person within situation (context) that matters for explanation of leadership phenomena; again, there is an appeal to integrate these notions in
terms of lower and higher levels of analysis.
Moreover, Bass (2008) develops the idea and importance of “context”
as a higher level of analysis within which leadership behavior occurs and
without which behaviors often become uninterpretable. For example,
individual (e.g., leader) behavior occurs within a group context, individual
and group behaviors (e.g., of followers) occur with an organizational (still
higher-level) context, and so on. In this way, lower and higher levels of
analysis together, in integration, explain the phenomena of interest and
avoid the “either-or” controversy.
SCHYNS_9781785367274_t.indd 237
10/11/2017 15:20
238 Handbook of methods in leadership research
Analytics and Levels
Given the diverse multi-level issues addressed here for leadership research,
a variety of multi-level methods and data analytic tools are required for
assessing and testing these notions empirically. A complete review of
these approaches is beyond the scope here, but some approaches will be
developed more fully in the following section. At this point, suffice it to
say that a core issue is the potential dependency created among observations that may not be independent due to nesting, embeddedness, and
cross-­classification of entities at multiple levels of analysis (see Dansereau
et al., 1984; Gooty & Yammarino, 2011). As such, many common analytic ­techniques (e.g., OLS regression) are inappropriate for multi-level
analyses.
Approaches that are appropriate, however, for analysing multi-level
notions in leadership work include random coefficient models (RCM)
via hierarchical linear modeling (HLM) and various multi-level routines
in Mplus and Stata, multi-level structural equation models (MLSEM),
within and between analysis (WABA), and multi-level routines in the R
software package (for details and leadership examples, see Dansereau
et al., 1984, 2006; Dansereau & Yammarino, 2000, 2006; Gooty &
Yammarino, 2011, 2016; Klein & Kozlowski, 2000; Yammarino, 1998;
Yammarino & Markham, 1992). All these analytic tools for assessing multiple levels of analysis have procedures for testing multi-level meditation,
moderation, and longitudinal (processual, developmental) notions while
accounting for dependencies in the data.
Another though very different analytic tool for multi-level leadership
work is dynamic computational modeling and simulation (e.g., Dionne
& Dionne, 2008; Dionne, Sayama, Hao, & Bush, 2010; Sayama, 2015).
Dionne and Dionne (2008), for example, developed a computational
model for a levels-based comparison of four types of leadership that
represent three different levels – individual, dyad, and group – across a
dynamic group decision-making optimization scenario. Computational
modeling, including agent-based modeling and simulation, of complex
non-linear relations among various traits, behaviors, properties, levels
of analysis, and environmental characteristics for leadership phenomena is a rapidly growing area of work. These computational tools for
modeling complex and dynamic multi-level notions can also account
for potential dependencies created by multiple embedded or interrelated entities and levels of analysis in leadership research (see Sayama,
2015).
SCHYNS_9781785367274_t.indd 238
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
239
DYADS IN LEADERSHIP
Having considered a variety of critical levels of analysis issues for both
theory and method that apply to all levels for leadership research, we focus
now on a particular level of analysis – dyads – which is the most recent
in the literature. In organizational behavior, for example, there is a long
history (from at least the 1960s) of considering individuals, groups and
(since the 1980s or so) teams, and organizations. The mention of dyads,
however, outside the realm of leadership is still rare today. But in the leadership realm, the use of the term “dyad” has been around since at least the
1970s (e.g., Dansereau, Graen, & Haga, 1975). Despite this history, dyads
were often considered only as part of a group (e.g., dyads within groups,
or in-group and out-group dyads). While the focus on dyads as a separate
unique level of analysis has been around in leadership work since the 1980s
(e.g., Dansereau et al., 1984), only very recently has this levels perspective
really come to the forefront in leadership (see Gooty & Yammarino, 2011,
2016). As such, we elaborate some important multi-level theoretical and
methodological issues for dyads, the most neglected and least understood
level of analysis in leadership research.
Foundations of Dyads
As previously explained, dyadic relationships involve one-to-one (twoparty) linkages between people, and they are omnipresent in work settings
and critical in leadership research. Indeed, the prevalence of dyads in work
settings and yet the neglect of their examination has produced numerous
calls for more theory, methods, and empirical research on this “forgotten”
level of analysis (see Gooty & Yammarino, 2011; Kenny, Kashy, & Cook,
2006). The study of dyadic relationships falls in the realm of multi-level
research and theory testing, but dyads are the least studied level of analysis
relative to individuals, groups or organizations (see Gooty & Yammarino,
2011, 2016; Krasikova & LeBreton, 2012; Schriesheim et al., 2001). In the
handful of studies where dyadic relationships are an explicit focus, considerable misalignment of theory, measurement and inferences exists. For
example, LMX theory is based on dyadic relationships between a leader
and each follower, whereas most empirical studies measure and analyse
relationships at the individual and group levels (for reviews, see Gooty et
al., 2012; Schriesheim et al., 2001). Even in studies where the dyadic level is
implemented, problems still abound with the statistical methods in use, and
there is a serious dearth of comprehensive conceptual and methodological treatments of the dyad as a level of analysis (see, for reviews, Gooty &
Yammarino, 2011, 2016; Kenny et al., 2006; Krasikova & LeBreton, 2012).
SCHYNS_9781785367274_t.indd 239
10/11/2017 15:20
240 Handbook of methods in leadership research
Dependencies for Dyads
Within- and between-dyad dependencies
Dyadic research explains and tests the relationship aspects between two
members of that dyad. The dyad members might vary in their perception
of relationship factors (e.g., communication, quality of exchanges) as well
as demonstrate some degree of similarity. From a measurement standpoint, any observations provided by both members of a dyad describing
relationship factors are “dependent”; that is, they are not independent of
one another. This type of linkage creates the dyad and is also known as
the within-dyad dependence (Kenny et al., 2006). In addition, dyads could
also be related due to nesting within a higher level (e.g., groups, organizations) and/or due to shared partners (i.e., one dyad member, a leader
for example, occurs in more than one dyad), so there is between-dyad
dependence.
Both types of dependencies, within and between dyads, create theoretical and methodological complexities. From a theoretical perspective, the
dependence per se within each dyad and between dyads in a group might
be of interest to researchers. For example, a typical LMX study seeks
to understand both types of dependencies. Even in studies where these
dependencies are not of theoretical interest, Gooty and Yammarino (2011,
2013), Kenny and Judd (1986), and Bliese and Hanges (2004) note that
these dependencies need to be accounted for in the research design and
data analysis. In particular, Gooty and Yammarino (2011) summarize
the statistical errors resulting from ignoring the dyad, and its associated
dependencies, as a valid level of analysis: biased standard errors, incorrect
inferences due to errors in significance testing, and more Type 1 errors (for
a comprehensive review of these issues, see Gooty & Yammarino, 2011;
Kenny & Judd, 1986; Schriesheim et al., 2001). The dependencies in dyads
could occur due to many causes such as space, time, or common fate (i.e.,
common external stimuli created by the work group influences). While all
of these are plausible sources of dependencies in leadership research, the
last one is particularly prevalent; that is, dyads are nested or embedded
within groups.
Dyad dependencies and groups
Dyads could be cast as two-person groups, although some dyads (e.g.,
where each dyadic member has a specific distinguishable role such as
husband and wife) might not be. Dependencies within and between dyads
can arise with i observations nested in j dyads that are housed in k groups.
Where these i observations are unique to the dyad they occupy, nested
hierarchical data structures accommodate the interdependence in lower-
SCHYNS_9781785367274_t.indd 240
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
241
level units. For example, two-person project teams (i.e., a dyad) can be
part of a larger team consisting of many such dyads. In these nested data
structures, each of the i observations and j dyads at lower levels are not
presumed independent unlike single-level data structures. These nested
data structures, however, do presume that each lower-level unit is uniquely
nested within a higher-level unit (e.g., a dyadic member belongs only to
one dyad and one group). One could imagine organizational settings
wherein this data structure applies to peer–peer relationships. In leadership research, however, where the leader is a key focus at the individual,
dyad and group levels, this assumption of unique membership is routinely
violated. The consequences of violating this assumption are similar to
what we articulated above with not accounting for dependencies in data.
To recap, ignoring lower-level dependencies could cause Type 1 and Type
2 errors and reduced power to detect Level 1 (i.e., lower-level) effects (see
also Bliese & Hanges, 2004; Dansereau & Yammarino, 2006; Gooty &
Yammarino, 2011, 2016; Kenny et al., 2006; Yammarino et al., 2005).
Very simply, most studies in leadership research include leaders and
followers at Level 1 (i.e., individual level) where the leader could occur in
multiple dyads and in a work group. In this instance, the lower-level unit
(i.e., leader) is not only nested within a higher-level unit (e.g., dyad), but
also resides in multiple higher-level units (i.e., several dyads) simultaneously. This non-unique higher-level membership violates a key assumption in nested data structures.
Cross-classification
Following Gooty and Yammarino (2011), one way to treat this issue of
non-unique dyadic membership is via cross-classification of the leader by
both the dyad and the group. This design differs from traditional nested
data structures in important ways. In particular, the leader and follower
are both modeled at Level 1 (individual level). The dependency within and
between dyads is modeled via the dyad-level membership, cross-classified
by the group. This cross-classification scheme affords the advantage of
modeling dyad-level variables (e.g., trust, LMX) at the appropriate dyadic
level and precludes the assumption that lower-level units are uniquely
nested in higher-level units. Rather, the dyads could be within the same
work group (i.e., work unit or organization-level dependencies) that is
modeled via the group factor.
Dependence at the dyad level could exist in two forms: within each dyad
and between dyads as depicted in Figures 10.1 and 10.2 (adapted from
Gooty & Yammarino, 2011). There could be within-dyad dependency
(indicated by double-headed arrows in Figures 10.1(a) and 10.2(a)) attributable to partner/member effects, mutual influence and/or common fate.
SCHYNS_9781785367274_t.indd 241
10/11/2017 15:20
242
SCHYNS_9781785367274_t.indd 242
10/11/2017 15:20
F2
F1
L2
L1
(b) Level 1
Level 2
Level 3
L1, F1
Dyad 1
L2, F2
Dyad 2
L3, F3
Dyad 3
Independent/unique dyads
Adapted from Gooty & Yammarino (2011).
Figure 10.1
Source:
Note: Each leader and follower belongs to one dyad (a). Each leader–follower dyad is independent at the dyad level as the leader and follower are
unique to that dyad. All three dyads might, however, belong to the same organization creating a higher-level dependency captured via the group
factor. Nested data structure to represent independent dyads (b).
(a)
F3
L3
Group
243
SCHYNS_9781785367274_t.indd 243
10/11/2017 15:20
L
(b) Level 1
Level 2
L
Group
F1
Dyad 1
F2
Dyad 2
F3
Dyad 3
Dependent/non-unique dyads
Adapted from Gooty & Yammarino (2011).
Figure 10.2
Source:
Note: Leader belongs to three dyads; followers belong to one dyad (a). The leader is common to all three dyads, creating between-dyad
dependencies. There might be certain dependencies between all the members in the group (leader and followers) driven by organization-level (or
work-unit) properties captured via the group factor. Cross-classified data structure to represent dependent dyads; solid arrows indicate dyad
membership, dashed arrows indicate group membership (b).
(a)
F3
F2
F1
244 Handbook of methods in leadership research
Partner effects exist when the actions of one dyadic member affect the outcomes of the other (e.g., a leader’s ability to provide resources could affect
follower task performance). Mutual influence refers to the reciprocal
effects of dyadic members’ actions and outcomes (e.g., a leader engages in
assigning tasks, a follower completes the assigned task and requests more
challenging tasks or autonomy). When both members of the dyad experience common environmental stimuli, they are subject to a final source of
within-dyad dependency known as common fate (e.g., a leader and follower are both bound by the same organizational policies and procedures)
(see also Gooty & Yammarino, 2011; Kenny et al., 2006).
Independent and dependent dyads
Dependence between dyads can be conceptualized via higher-level linkages and common members/partners. In the former, dyads entail unique
membership and are nested within groups and/or organizations, also
known as independent/pure dyads (Dansereau et al., 1984; Gooty &
Yammarino, 2011; Kenny et al., 2006), as shown in Figure 10.1(a). In such
instances, each of the dyad members (i.e., L1–F1, L2–F2, and L3–F3)
appears in only one dyad and thus contributes to only one interpersonal
relationship. When the dyads have this structure, theory, measurement
and analyses could emerge at the individual, dyadic and group levels as
described further in the following paragraphs.
Theory for independent dyads could suggest, for example, that LMX is
positively associated with performance at the dyadic level (e.g., Gooty &
Yammarino, 2011, 2016). Measurement then for this theoretical assertion
is depicted in Figure 10.1(b) where leaders and followers both appear at
Level 1 providing matched ratings of their LMX relationship and the performance of the follower all at the individual level. The dyadic-level measurement of LMX could be a direct consensus construct (e.g., Chan, 1998)
appearing at Level 2, presuming leaders and followers agree regarding
their relationship. Such dyadic LMX is presumed to vary between dyads
(e.g., LMX). These dyads could be embedded in work groups subject
to the same procedural justice formulations, for example. Level 3 then
captures such variables that vary between work groups (e.g., procedural
justice). This example of independent/pure dyads represents nested data
structures in which dyadic and group membership is unique; that is, each
leader and follower appears in one dyad only and each dyad appears in
one work group only. In general, theory in such cases could be at the individual, dyadic and/or group levels. Measurement, analyses and inference
drawing should then be aligned with such presumed theoretical suppositions. We also note that the case of independent work dyads is more easily
accessible to multi-level researchers cognitively as it follows the nested
SCHYNS_9781785367274_t.indd 244
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
245
data structure configuration that is widely prevalent in leadership, teams
and organizational behavior research.
The next configuration, however, is much less known in the organizational sciences (for exceptions, see Gooty & Yammarino, 2011, 2016).
Specifically, while multi-level data structures in the organizational sciences
have largely presumed nested data, in practice this assumption is violated
when the dyad level is of interest as dyadic membership is rarely unique.
Such designs entail an additional level of dependency as shown in Figure
10.2. Kenny et al. (2006) have called such dyads the “one with many”
design, while Dansereau et al. (1984) and Yammarino and Dansereau
(2009, 2011) label such configurations “dyads within groups.” More
recently, Gooty and Yammarino (2011, 2013) indicated that such dyads
represent cross-classified dyads. For example, in leadership research
(e.g., LMX and its predecessor, vertical dyad linkage theory; Dansereau
& Yammarino, 1998a, 1998b), a typical leader is part of multiple dyads
(e.g., the appearance of the same leader, L, in multiple L–F1, L–F2, L–F3
dyads in Figure 10.2). In addition to dependencies attributable to common
group membership as in the case of independent dyads above, here dyads
are also dependent on the common leader across multiple dyads. Just like
the case of independent dyads, theory, measurement and analyses could
be at the individual, dyadic and/or group levels here. Analytical strategies,
however, must account for the cross-classification of the leader by dyad
and group. Such analytical strategies via three different methodological
approaches and associated statistical techniques are detailed in the next
section.
Three Dyadic Approaches
The above explication about dyads and the potential dependencies within
and between dyads permits a consideration of three primary methodological approaches to dyads in leadership research. These approaches, as
summarized in Table 10.1, are called (1) actor–partner interaction model
(APIM) which includes one with many (OWM) designs and the social
relations model (SRM) (e.g., Campbell & Kashy, 2002; Kenny et al., 2006;
Kenny & Livi, 2009; Krasikova & LeBreton, 2012); (2) random coefficient model (RCM) typically implemented via hierarchical linear model
(HLM) and hierarchical cross-classified model (HCM) analytic software
(e.g., Bryk & Raudenbush, 1992; Gooty & Yammarino, 2011, 2016;
Raudenbush & Bryk, 2002; Raudenbush, Bryk, Cheong, Congden and Du
Toit et al., 2004) (but also by Mplus and R software); and (3) within and
between analysis (WABA) typically implemented via Data Enquiry that
Tests Entity and Correlational-Causal Theories (DETECT) software (e.g.,
SCHYNS_9781785367274_t.indd 245
10/11/2017 15:20
246 Handbook of methods in leadership research
Table 10.1
Comparison of dyadic approaches
Issue
APIM/OWM/SRM
RCM/HLM/HCM
WABA/DETECT
References
Campbell & Kashy
(2002); Kenny et al.
(2006); Kenny & Livi
(2009); Krasikova &
LeBreton (2012)
Background
& tradition
Psychological sciences
– interpersonal
& romantic
relationships
Do dyad- and
individual-level
effects affect
individual-level
scores? Rely on
theory and partitions
based on individual
or higher levels
Dansereau et al.
(1984); Gooty &
Yammarino (2011,
2016); Schriesheim
(1995); Yammarino
(1998); Yammarino &
Dansereau (2009, 2011)
Organization sciences –
behavior of individuals,
groups, & organizations
Views of
dyads
Two choices:
homogeneity/wholes
or independent/none
Bryk & Raudenbush
(1992); Gooty &
Yammarino (2011,
2016); Raudenbush
& Bryk (2002);
Raudenbush et al.
(2004)
Education –
individual student
achievement &
performance
Do dyad- and
individual-level
variables affect
individual-level
outcomes? Rely on
theory and justify
with aggregation
procedures whether
individual or higher
levels
Two choices:
homogeneity/wholes
or independent/none
Nature of
dyads
Independent/unique
or dependent/
non-unique
Individual level
Independent/unique
or dependent/
non-unique
Individual level
Individual or higher
level
Dyad or higher level
Individual or higher
level
Dyad or higher level
Dyad or higher level
Variance partitioning
and statistical
significance tests
using regression and
SEM approaches
Inferential statistical
significance tests using
maximum likelihood,
etc. algorithms
Tests of statistical &
practical significance
using ANOVA and
correlation/regression
approach
Key question
Dependent
variables
Independent
variables
Mediators &
moderators
Basis of
analyses
Do variables and
relationships operate at
the dyad level? Tests,
after assertions based
on theory, whether
individual, within-dyad,
or between-dyad level
Three choices:
homogeneity/wholes,
heterogeneity/parts, or
independent/none
Dependent/non-unique
or independent/unique
Dyad level
Dyad or higher level
Note: APIM = actor–partner interdependence model; OWM = one with many; SRM =
social relations model; RCM = random coefficient modeling; HLM = hierarchical linear
modeling; HCM = hierarchical cross-classified models; WABA = within and between
analysis; DETECT = Data Enquiry that Tests Entity and Correlational/Causal Theories.
SCHYNS_9781785367274_t.indd 246
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
247
Dansereau et al., 1984; Gooty & Yammarino, 2011, 2016; Schriesheim,
1995; Yammarino, 1998; Yammarino & Dansereau, 2009, 2011).
For all three approaches, dyadic or relationship leadership research
requires measurement from both parties on the constructs of interest. These
measurements could be individuals reporting on their own states (e.g.,
follower and leader each report about their own LMX) or could focus on
assessments of the relationship per se, as reported by each party (e.g., leader
and follower both report about the mutual level of LMX in their relationship). This form of measurement from both parties on the same construct
simultaneously is called reciprocal data/measures (by Kenny et al., 2006) or
matched data/measures (by Dansereau et al., 1984; Gooty & Yammarino,
2011; Yammarino & Dansereau, 2009). Note that simply collecting a measurement from one party (e.g., a follower) on one variable (e.g., LMX), and
from the other party (e.g., a leader) on another variable (e.g., performance),
does not constitute dyadic research – it is individual differences research as
dyadic measurement does not occur for any shared constructs.
The other similarities and differences among the characteristics of these
dyadic approaches are shown in the table. In particular, the three dyadic
approaches arise from different disciplinary and statistical traditions
that are designed to test different types of multi-level questions. APIM
arises out of the psychological sciences where the focus is interpersonal
and romantic relationships, and a key question is whether dyad- and
individual-level effects affect individual-level scores. It relies on theory
and variance partitions based on individual or higher levels. RCM arises
out of education research where the focus is individual student achievement and performance, and a key question is whether higher-level variables (e.g., student–teacher relations, schools, and school districts) affect
individual-level outcomes (e.g., individual student achievement). It relies
on theory and justification with aggregation procedures for individual or
higher levels. WABA arise out of the organization sciences where the focus
is individual, group, and organizational behaviors, and a key question is
whether variables and relationships operate at the dyad and higher levels.
It relies on tests, after assertions based on theory, whether individual,
within-dyad, or between-dyad level effects are relevant.
In both the APIM and RCM approaches, when considering dyads
as a level of analysis or the entities of focus, there are two plausible
views: homogeneity/wholes or independent/none (as described above).
Essentially, there are or are not dyads present. In contrast, in WABA
there are three plausible views: homogeneity/wholes, heterogeneity/parts,
or independent/none (as described above). In this case, after identifying that dyads are present, then one must decide/ask whether the dyadic
partners are similar vs positioned relative to one another. All three dyadic
SCHYNS_9781785367274_t.indd 247
10/11/2017 15:20
248 Handbook of methods in leadership research
approaches are equipped to handle independent (unique) and dependent
(non-unique) dyads; the former being the original focus of the APIM and
RCM via HLM, and the latter being the original focus of WABA and
more recently RCM via HCM.
In terms of different types of variables, the APIM and RCM dyadic
approaches are similar. For these two approaches, dependent variables
operate at the individual level, independent variables operate at the individual or higher levels, and mediator and moderator variables operate
at the dyad or higher levels. In WABA, however, dependent variables
operate at the dyad level, independent variables operate at the dyad or
higher levels, and mediator and moderator variables operate at the dyad
or higher levels.
For data analysis, APIM relies on variance partitioning and statistical significance tests using regression and structural equation modeling
approaches. In APIM, the basic SRM equation (see Kenny et al., 2006;
Kenny & Livi, 2009) that applies for actor i with partner j in group k is:
Xijk 5 mk + aik +bjk +gijk
(10.1)
where Xijk is the score for person i rating person j, mk is the group mean,
aik is person i’s actor effect, bjk is person j’s partner effect, and gijk is the
relationship effect. Variances for the random variables are: s2m, s2a, s2b,
s2g. Also, at the individual level, generalized reciprocity is a person’s actor
effect correlated with that person’s partner effect, sab; and, at the dyad
level, dyadic reciprocity is two members’ correlated relationship effects,
sgg. Last, there is an overall intercept or grand mean, uk. In total, seven
parameters are estimated: one mean, four variances, and two covariances.
For data analysis, RCM relies on inferential statistical significance tests
using maximum likelihood and other algorithms. In RCM implemented
via the HLM software, there can be three-level models with individuals
nested within dyads nested within groups. The slope and intercept from
these Level 1 regressions are modeled as Level 2 outcomes. A typical
three-level unconditional model, say with individuals (i) nested in dyads
(j) nested in groups (k), in RCM (HLM) looks as follows (for details and
definitions, see Gooty & Yammarino, 2011):
Level 1: Yijk 5 π0jk + eijk
Level 2: π0jk 5 b00k + r0jk
Level 3: b00k 5 g000 + u00k
(10.2)
(10.3)
(10.4)
RCM via the use of HCM (hierarchical cross-classified random modeling,
or simply cross-classification) is used when lower-level units belong to
SCHYNS_9781785367274_t.indd 248
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
249
multiple higher-level units (i.e., dependent dyads). A typical Level 1 and
Level 2 unconditional model with individuals in dyads cross-classified by
groups is as follows (for details and definitions, see Gooty & Yammarino,
2011):
Level 1: Yijk 5 π0jk + eijk
Level 2: π0jk 5 q0 + b00j+ c00k
(10.5)
(10.6)
Note that Equations 10.2 (independent dyads) and 10.5 (dependent dyads)
are identical. The key difference is in the higher-level models. Equations
10.3 and 10.4 build in the dyad and group levels one step at a time as
they align with the assumption that lower-level units are uniquely nested
in higher-level units. From a statistical standpoint, the nested model
(modeled via Equations 10.2–10.4) rests on the assumption that observations within each equation are independent. Equation 10.6 (HCM),
however, builds in the column factor term {c00k} modeled simultaneously
with the dyad factor term {b00k}, thus accounting for the dependency
between dyads and the group simultaneously.
For data analysis, WABA relies on tests of both statistical and practical
significance (the latter tests are not dependent on degrees of freedom and
are based in coordinate-free geometry) using ANOVA and ­correlation/
regression approaches. In WABA, the total correlation between two
variables x and y is partitioned into within and between components,
also known as the basic WABA equation (see Dansereau et al., 1984;
Yammarino, 1998; Yammarino & Markham, 1992):
Total Correlation 5 Between Component + Within Component
rxy 5hbx hby rbxy + hwx hwy rwxy(10.7)
where hbx and hby are the between etas for variables x and y, hwx and hwy
are the corresponding within etas, and rbxy and rwxy are the corresponding between- and within-cell correlations. rxy is the traditional raw score
(total) correlation between variables x and y. When individual and dyad
levels (i 5 individuals [both sources], d 5 dyads [dependent or independent]) are considered, the equation is:
rxyi 5 hbxd hbyd rbxyd + hwxd hwyd rwxyd(10.8)
And when dyad and group levels (d 5 dyads [dependent], j 5 groups) are
considered, the equation is:
rxyd 5 hbxj hbyj rbxyj + hwxj hwyj rwxyj(10.9)
SCHYNS_9781785367274_t.indd 249
10/11/2017 15:20
250 Handbook of methods in leadership research
RECOMMENDATIONS FOR LEADERSHIP
RESEARCH
As noted above, multi-level issues and multiple levels of analysis are
critical in leadership research because theory and theory-building considerations without levels of analysis are incomplete; and likewise, data and
theory-testing considerations without levels of analysis are incomprehensible. As most, actually all, leadership phenomena are multi-level in nature
we offer some thoughts and recommendations to summarize the importance of these issues (also see Dionne et al., 2014; Gooty & Yammarino,
2011; Yammarino et al., 2005).
In particular, we hope to impress upon readers that levels of analysis are
not simply an analytical issue. Levels of analysis considerations begin with
theory. The following are our recommendations for theory in multi-level
leadership research:
1.Define the level of analysis of the unit(s) of interest; that is, the entity
(entities) to which theoretical generalizations apply. For example,
some constructs in leadership research exist at the individual level
(such as follower or leader perceptions of the LMX relationship).
Others, such as when LMX is defined as the quality of the exchange
relationship, should reside at the dyadic level. Gooty and Yammarino
(2013) discuss emergence processes (e.g., shared realities) that allow
for dyadic LMX (based on consensus among dyadic partners) and
dyadic dispersion LMX (the degree to which dyadic partners differ)
to exist at the dyadic level. Similarly, the degree to which followers
differ in their perceptions of LMX with the leader in a workgroup is
LMX differentiation that draws upon key compilational processes.
The key point here is that constructs and the associations among them
will not always emerge at higher levels. And when they do, they might
change functional form. For example, dyadic LMX behaves much like
individual LMX with regard to follower performance, whereas dyadic
dispersion LMX attenuates the individual-level LMX – performance
association (e.g., Gooty & Yammarino, 2011, 2016).
2.Define the level of analysis of the associated concepts, constructs,
variables, and relationships.
3.Keeping in mind that not all constructs and relationships among them
emerge at higher levels, emergence processes must be articulated that
provide a theoretical justification for how and why constructs and
their associations emerge at higher levels.
4.Specify the boundary conditions, including and based upon levels of
analysis, for everything articulated in 1, 2, and 3 here.
SCHYNS_9781785367274_t.indd 250
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
251
The theoretical steps delineated above lead to the following recommendations regarding measurement in multi-level leadership research:
1.Construct measures at the same level of analysis depicted in the theory,
models, and hypotheses. In some instances, this might be made possible because the construct only exists at the higher level (e.g., group
size) and, in other cases, it might be derived from lower-level ratings
as noted below.
2.When the measurement of constructs at their appropriate level is not
directly possible or feasible, then lower-level ratings are combined in
compositional or compilational processes to construct higher-level
measures (see Chan, 1998). Compositional models assume consensus
in lower-level ratings and require appropriate justification (consensus,
agreement, aggregation) indices such as the rwg, ICC1, ICC2, and so
on (for a review, see LeBreton & Senter, 2008). Compilational models
rely on variability and do not require such consensus justification but
require reliable between-unit differences to be displayed.
3.Further, validate a measure, even an established measure, if it has been
modified or adapted to account for various or different levels of analysis than originally intended (e.g., for a change in referent or entities).
Next, our recommendations for data analysis in multi-level leadership
research are the following:
1.Permit theory (variables, relationships, and levels of analysis) to determine the multi-level technique to be used.
2.Employ appropriate multi-level techniques if the entities of interest
are at a level of analysis above the individual level.
Finally, below are our recommendations for drawing inferences in
multi-level leadership research:
1.Include levels of analysis in both theory (i.e., as the entities) and data
(i.e., as the samples and subjects).
2.State which relationships hold across different levels of analysis in
terms of multi-level, cross-level, emergent, and level-specific models.
In terms of dyads and the inherent dependencies both within and between
dyads, a particular focus here, our recommendations are the following:
1.Conceptually, explicate why and how dyads are expected to form and
operate in leadership research. This is important because leadership
SCHYNS_9781785367274_t.indd 251
10/11/2017 15:20
252 Handbook of methods in leadership research
relationships demonstrate some form of dependency, and elucidating
the nature of this dependency and how it could impact consequences
is critical.
2.Methodologically, at both the research design (measurement) and
the data analysis stages of leadership research, modeling dyads as
a level of analysis requires a consideration of dependencies (or lack
thereof). If researchers are interested in only the individual level,
a rare but possible situation in leadership research, then single- or
one-sided reports (measurements) from only one dyadic partner’s
perspective might be appropriate. If the dyadic relationship per
se is of interest, however, the more typical situation in leadership
research, then matched (or reciprocal) reports (measurements)
from both dyadic partners are appropriate. In these cases, the three
dyadic methodological and analytic approaches reviewed here –
APIM, RCM, and WABA – can be quite useful for leadership
researchers.
In conclusion, we have presented a variety of multi-level issues that are
important in leadership research. From both theoretical and methodological perspectives, we stressed the multi-level nature of leadership research
and the need to address such issues in the research process. We also considered, in some detail, dyads and dyadic dependencies relevant for leadership research. Dependencies within and between dyads give rise to cases
of independent and dependent dyads, and these are treated somewhat
differently in the three dyadic methodological and analytic approaches
presented here. Our hope is simply that by raising awareness of these issues
leadership researchers will understand better the nature of multi-level
research and implement it appropriately in their studies to further faster
the state of leadership science.
REFERENCES
Bass, B.M. (2008). The Bass handbook of leadership. New York: Free Press.
Behling, O. (1978). Some problems in the philosophy of science of organizations. Academy of
Management Review, 3, 193–201. doi: 10.5465/AMR.1978.4294841
Bliese, P.D., & Hanges, P.J. (2004). Being both too liberal and too conservative: The perils of
treating grouped data as though they were independent. Organizational Research Methods,
7(4), 400–417. doi: 10.1177/1094428104268542
Bryk, A.S., & Raudenbush, S.W. (1992). Hierarchical linear models. Newbury Park, CA:
Sage.
Campbell, L.J., & Kashy, D.A. (2002). Estimating actor, partner, and interaction effects
for dyadic data using PROC MIXED and HLM5: A user-friendly guide. Personal
Relationships, 9(3), 327–342. doi: 10.1111/1475-6811.00023
SCHYNS_9781785367274_t.indd 252
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
253
Chan, D. (1998). Functional relations among constructs in the same content domain at different levels of analysis: A typology of composition models. Journal of Applied Psychology,
83(2), 234–246. doi: 10.1037/0021-9010.83.2.234
Dansereau, F., & Yammarino, F.J. (Eds.) (1998a). Leadership: The multiple-level approaches
(Part A: Classical and new wave) (Vol. 24 of Monographs in organizational behavior and
industrial relations). Stamford, CT: JAI Press.
Dansereau, F., & Yammarino, F.J. (Eds.) (1998b). Leadership: The multiple-level approaches
(Part B: Contemporary and alternative) (Vol. 24 of Monographs in organizational behavior and industrial relations). Stamford, CT: JAI Press.
Dansereau, F., & Yammarino, F.J. (2000). Within and between analysis: The variant paradigm as an underlying approach to theory building and testing. In K.J. Klein & S.W.J.
Kozlowski (Eds.), Multilevel theory, research, and methods in organizations: Foundations,
extensions, and new directions (pp. 425–466) (SIOP Frontiers Series). San Francisco, CA:
Jossey-Bass.
Dansereau, F., & Yammarino, F.J. (2006). Is more discussion about levels of analysis really
necessary? When is such discussion sufficient? The Leadership Quarterly, 17(5), 537–552.
doi: 10.1016/j.leaqua.2006.07.002
Dansereau, F., Alutto, J.A., & Yammarino, F.J. (1984). Theory testing in organizational
behavior: The variant approach. Englewood Cliffs, NJ: Prentice-Hall.
Dansereau, F., Cho, J., & Yammarino, F.J. (2006). Avoiding the “fallacy of the wrong level”:
A within and between analysis (WABA) approach. Group and Organization Management,
31(5), 536–577. doi: 10.1177/1059601106291131
Dansereau, F., Graen, G., & Haga, W.J. (1975). A vertical dyad linkage approach
to leadership within formal organizations: A longitudinal investigation of the rolemaking process. Organizational Behavior and Human Performance, 13(1), 46–78. doi:
10.1016/0030-5073(75)90005-7
Dansereau, F., Yammarino, F.J., & Kohles, J. (1999). Multiple levels of analysis from a
longitudinal perspective: Some implications for theory building. Academy of Management
Review, 24(2), 346–357. doi: 10.5465/AMR.1999.1893940
DeChurch, L.A., Hiller, N.J., Murase, T., Doty, D., & Salas, E. (2010). Leadership across
levels: Levels of leaders and their levels of impact. The Leadership Quarterly, 21(6), 1069–
1085. doi: 10.1016/j.leaqua.2010.10.009
Dionne, S.D., & Dionne, P.J. (2008). Levels-based leadership and hierarchical group decision optimization: A simulation. The Leadership Quarterly, 19(2), 212–234. doi: 10.1016/j.
leaqua.2008.01.004
Dionne, S.D., Gupta, A., Sotak, K.L., Shirreffs, K., Serban, A., Hao, C.,. . .Yammarino,
F.J. (2014). A 25-year perspective on levels of analysis in leadership research. The
Leadership Quarterly, 25(1), 6–35. doi: 10.1016/j.leaqua.2013.11.002
Dionne, S.D., Sayama, H., Hao, C., & Bush, B.J. (2010). The role of leadership in
shared mental model convergence and team performance improvement: An agentbased computational model. The Leadership Quarterly, 21(6), 1035–1049. doi: 10.1016/j.
leaqua.2010.10.007
Gooty, J., & Yammarino, F.J. (2011). Dyads in organizational research: Conceptual
issues and multi-level analyses. Organizational Research Methods, 14(3), 456–483. doi:
10.1177/1094428109358271
Gooty, J., & Yammarino, F.J. (2016). The leader–member exchange relationship: A
multi-source, cross-level investigation. Journal of Management, 42(4), 915–935. doi:
10.1177/0149206313503009
Gooty, J., Serban, A., Thomas, J.S., Gavin, M.B., & Yammarino, F.J. (2012). Use
and misuse of levels of analysis in leadership research: An illustrative review of
leader–member exchange. The Leadership Quarterly, 23(6), 1080–1103. doi: 10.1016/j.
leaqua.2012.10.002
Hackman, J.R. (2003). Learning more by crossing levels: Evidence from airplanes, hospitals,
and other orchestras. Journal of Organizational Behavior, 24(8), 905–922. doi: 10.1002/
job.226
SCHYNS_9781785367274_t.indd 253
10/11/2017 15:20
254 Handbook of methods in leadership research
House, R.J., Hanges, P.J., Javidan, M., Dorfman, P.W., & Gupta, V. (Eds.) (2004). Culture,
leadership, and organizations: The GLOBE study of 62 societies. Thousand Oaks, CA: Sage
Publications.
Katz, D., & Kahn, R.L. (1978). The social psychology of organizations. New York: Wiley.
Kenny, D.A., & Judd, C.M. (1986). Consequences of violating the independence assumption in analysis of variance. Psychological Bulletin, 99(3), 422–431. doi:
10.1037/0033-2909.99.3.422
Kenny, D.A., & Livi, S. (2009). A componential analysis of leadership: Using the social
relations model. Research in Multi-Level Issues, 8 (Multi-Level Issues in Organizational
Behavior and Leadership), 147–191.
Kenny, D.A., Kashy, D.A., & Cook, W.L. (2006). Dyadic data analysis. New York: Guilford.
Klein, K.J., & Kozlowski, S.W.J. (2000). From micro to meso: Critical steps for conceptualizing and conducting multi-level research. Organizational Research Methods, 3(3),
211–236. doi: 10.1177/109442810033001
Klein, K.J., Dansereau, F., & Hall, R.J. (1994). Levels issues in theory development, data
collection, and analysis. Academy of Management Review, 19(2), 195–229. doi: 10.5465/
AMR.1994.9410210745
Krasikova, D.V., & LeBreton, J. (2012). Just the two of us: Misalignment of theory and
methods in examining dyadic phenomena. Journal of Applied Psychology, 97(4), 739–757.
doi: 10.1037/a0027962
LeBreton, J.M., & Senter, J.L. (2008). Answers to 20 questions about inter-rater reliability and inter-rater agreement. Organizational Research Methods, 11(4), 815–852. doi:
10.1177/1094428106296642
Markham, S.E. (2010). Leadership, levels of analysis, and déjà vu: Modest proposals for
taxonomy and cladistics coupled with replications and visualization. The Leadership
Quarterly, 21(6), 1121–1143. doi: 10.1016/j.leaqua.2010.10.011
Markham, S.E. (2012). The evolution of organizations and leadership from the ancient
world to modernity: A multilevel approach to organizational science and leadership. The
Leadership Quarterly, 23(6), 1134–1151. doi: 10.1016/j.leaqua.2012.10.011
Miller, J.G. (1978). Living systems. New York: McGraw Hill.
Morgeson, F.P., & Hofmann, D.A. (1999). The structure and function of collective
constructs: Implications for multilevel research and theory development. Academy of
Management Review, 24(2), 249–265. doi: 10.5465/AMR.1999.1893935
Pedhazur, E.J. (1982). Multiple regression in behavioral research. New York: Holt, Rinehart,
& Winston.
Raudenbush, S.W., & Bryk, A.S. (2002). Hierarchical linear models: Applications and data
analysis methods (2nd ed.). Thousand Oaks, CA: Sage.
Raudenbush, S.W., Bryk, A.S., Cheong, Y.F., Congdon, R., & Du Toit, M. (2004). HLM
6: Hierarchical linear and nonlinear modeling. Lincolnwood, IL: Scientific Software
International, Inc.
Robinson, W.S. (1950). Ecological correlations and the behavior of individuals. American
Sociological Review, 15(3), 351–357. Retrieved from http://www.jstor.org/stable/2087176
Rousseau, D.M. (1985). Issues of level in organizational research: Multi-level and cross-level
perspectives. Research in Organizational Behavior, 7, 1–37. Retrieved from digitalcollections.library.cmu.edu/awweb/awarchive?type5file&item549182
Sayama, H. (2015). Introduction to modeling and analysis of complex systems. Albany, NY:
Open SUNY.
Schriesheim, C.A. (1995). Multivariate and moderated within- and between-entity analysis
(WABA) using hierarchical multiple linear regression. The Leadership Quarterly, 6(1),
1–18. doi: 10.1016/1048-9843(95)90002-0
Schriesheim, C.A., Castro, S.L., Zhou, X.T., & Yammarino, F.J. (2001). The folly of theorizing “A” but testing “B”: A selective level-of-analysis review of the field and a detailed leader–
member exchange illustration. The Leadership Quarterly, 12(4), 515–551. doi: 10.1016/
S1048-9843(01)00095-9
Yammarino, F.J. (1998). Multivariate aspects of the varient/WABA approach: A discus-
SCHYNS_9781785367274_t.indd 254
10/11/2017 15:20
Multi-level issues and dyads in leadership research ­
255
sion and leadership illustration. The Leadership Quarterly, 9(2), 203–227. doi: 10.1016/
S1048-9843(98)90005-4
Yammarino, F.J., & Dansereau, F. (Eds.) (2009). A new kind of OB (organizational behavior). Research in Multi-Level Issues, 8, 3–60.
Yammarino, F.J., & Dansereau, F. (2011). Multi-level issues in evolutionary theory, organization science, and leadership. The Leadership Quarterly, 22(6), 1042–1057. doi: 10.1016/j.
leaqua.2011.09.002
Yammarino, F.J., & Markham, S.E. (1992). On the application of within and between analysis: Are absence and affect really group-based phenomena? Journal of Applied Psychology,
77(2), 168–176 (correction, p. 426). doi: 10.1037/0021-9010.77.2.168
Yammarino, F.J., Dionne, S.D., Chun, J.U., & Dansereau, F. (2005). Leadership and levels
of analysis: A state-of-the-science review. The Leadership Quarterly, 16(6), 879–919. doi:
10.1016/j.leaqua.2005.09.002
SCHYNS_9781785367274_t.indd 255
10/11/2017 15:20
11. A social network approach to examining
leadership
Markku Jokisaari
For many leadership scholars, leadership is essentially a social phenomenon. For example, ‘most definitions of leadership stress social or interpersonal influence processes as key elements’ (Zaccaro & Klimoski, 2001,
p. 10), and central theories of leadership focus on social exchange between
leaders and their followers (Dinh et al., 2014). However, although leadership as per its definition is a relational process that includes relations with
other people, leadership research has primarily focused on leaders’ attributes and behaviour – particularly others’ perceptions of their attributes
and behaviour – rather than on the network of relations in which leadership is embedded. Further, research focusing on relationships has typically
examined dyadic relationship characteristics, such as the quality of the
working relationship between a leader and her or his followers, whereas
the wider social environment around leader–follower dyads has gained
much less research attention.
The social network approach provides both the theory and methodology for a detailed examination of the characteristics of the social environment of leadership (for reviews, see Balkundi & Kilduff, 2005; Carter,
DeChurch, Braun, & Contractor, 2015; Sparrowe, 2014; Sparrowe &
Liden, 1997). Theories related to the social network approach provide
concepts to help define and perceive this social context, and they further
argue that the social context provides both opportunities and constraints
for individual and group behaviour and related success (e.g., Kilduff et al.,
2006; Tichy, Tushman, & Fombrun, 1979; Wellman, 1988). Specifically,
the social network approach argues that social interaction and exchange
within and between organizations includes not only formal channels but
also informal social relations and interactions that affect how actors access
resources and accomplish their organizational goals (e.g., Granovetter,
1985). In other words:
[. . .]both formal and informal organizational elements generate a web of interactions connecting actors. These interactions, whether formally designed or
informally emergent, are conduits through which organizational actors coordinate efforts, share goals, exchange information, and access resources that affect
behaviours and performance outcomes. (McEvily et al., 2014, pp. 302–303)
256
SCHYNS_9781785367274_t.indd 256
10/11/2017 15:20
A social network approach to examining leadership ­257
For example, opinion leaders do not necessarily have a formal leader position; rather, they are often people who are well connected between social
groups through their social networks (e.g., Burt, 2005). Scholars adopting
the social network approach further argue that by focusing on informal
social contexts – that is, social networks – researchers can examine ‘how
work really gets done in organizations’ (Cross & Parker, 2004).
Similarly, the social network approach provides a tool to examine
central research questions in leadership, since leadership is embedded
in organizational environments that are characterized by both formal
and informal relations and interactions (e.g., Balkundi & Kilduff, 2005;
Fernandez, 1991). First, many leadership scholars argue that leadership
is centrally related to social influence among people and the coordination of people’s efforts toward a common goal (e.g., Kaiser, Hogan, &
Kraiger, 2008). In this regard, the social network approach argues that to
understand social influence and the coordination of common efforts, one
has to examine how people are connected to each other and to focus on
the wider social environment rather than formal dyadic relations between
a leader and her followers (e.g., Fernandez, 1991). Second, a central part
of leadership concerns access to resources in order to enhance the performance of work groups (e.g., Kaiser et al., 2008). In this respect, the
social network approach suggests that leaders’ ability to acquire resources
depends on not only their formal position but also their informal relations within and outside the organization (e.g., Fernandez, 1991). For
example, earlier research on social networks and leadership has shown
that leaders’ position in internal and external social networks of their
work groups is related to how well their work groups perform (Mehra,
Dixon, Brass, & Robertson, 2006). Third, a central research question
among leadership scholars is related to leadership development – that is
to say, how leadership evolves over time between a leader and his or her
followers (e.g., Day, Fleenore, Atwater, Sturm, & McKee, 2014). Again,
the social network approach suggests that in order to understand leadership development, a researcher should focus on not only the connections
between leaders and followers but also their relations outside their focal
work group (Balkundi & Kilduff, 2005). For example, earlier research has
shown that leaders’ position in social networks of the workplace is related
to others’ perceptions of them as leaders (e.g., Venkataramani, Green, &
Schleicher, 2010). In this regard, the social network approach thus provides theories and concepts to examine both formal and informal aspects of
leadership.
The methodology of the social network approach provides tools to
measure and analyse social environments in detail, and in the social
network approach specific measures have been introduced to examine the
SCHYNS_9781785367274_t.indd 257
10/11/2017 15:20
258 Handbook of methods in leadership research
attributes of social networks and individuals’ positions in these networks.
These network measures can relate to the individual, dyadic, triadic,
group, and whole network levels (Wasserman & Faust, 1994), and these
characteristics of social networks have important consequences for leadership. For example, earlier research has shown that leaders’ position in
social networks (i.e., the individual level) is related to others’ perception of
leaders’ helpfulness (Galunic, Ertuk, & Gargiulo, 2012), and the network
structure of a work group (i.e., the group level) is related work group
performance (Sparrowe, Liden, Wayne, & Kraimer, 2001). Furthermore,
a central methodological question for a researcher engaged in social
network research is whether to use a whole or personal network design. A
whole network design refers to a bounded social environment, such as a
department in an organization, in which all or most of the people report
their relations or connections to each other (e.g., Wasserman & Faust,
1994), whereas a personal network refers to a focal individual’s direct ties
and her or his perceptions of the relations among these ties in the network.
These networks are often called ‘ego networks’, and the focal actor who
reports his or her ties is called the ‘ego’ in the network literature. Thus,
social network methods offer possibilities to examine how leadership
manifests in different levels of a social environment depending on the
research questions.
In this chapter, I will first briefly present the central characteristics
of the social network approach and discuss how they might relate to
leadership research. Thereafter, I consider the methodology of the social
network approach, including the study design, sampling and data collection methods, and methods to measure social networks. I further discuss
central measures of networks for use in data analysis and statistical
inference for network data. I also provide examples from the leadership
research that has capitalized on social network methodology and theory.
Finally, I present a research example in detail that includes the collection of network data and application network analysis to understand the
­characteristics of social networks.
SOCIAL NETWORK APPROACH AND LEADERSHIP
RESEARCH
The social network approach focuses on the relations between actors,
such as leaders’ relations in the organization and how these relations,
or lack of relations, have important consequences for their actions (e.g.,
Wellman, 1988). Actors may be individual persons, social groups, organizations, or other collectives. A relation can include different types of
SCHYNS_9781785367274_t.indd 258
10/11/2017 15:20
A social network approach to examining leadership ­259
interaction or exchange between actors, such as ‘gives support to’, ‘asks
advice from’ or ‘discusses with’. A relationship can be directional when an
actor gives or sends something to another actor, or non-directional when
actors mutually contribute to the interaction or exchange. Furthermore,
the general pattern of relations between actors constitutes the structure
of the network, and each actor has his or her personal relations and
position within and between networks. Overall, actors and their actions
are viewed as interdependent rather than independent (e.g., Wellman,
1988).
The social network approach further argues that the network structure
of the individuals and groups has important consequences for their goaldirected action. First, the network structure affects how actors coordinate
their action toward common goals (e.g., Coleman, 1988; Oh, Labianca, &
Chung, 2006). Moreover, the characteristics of network ties define how a
group’s members are able to form ‘bonding ties’ (e.g., Adler & Kwon, 2002)
and related trust with each other and to coordinate their action toward
their goals. Second, individuals’ positions and relations affect how they
are able to acquire resources through their network ties to accomplish their
goals (e.g., Burt, 2005). That is, social networks are important channels
for resources such as information and credentials (e.g., Lin, 2001). Third,
an actor’s position in the network is related to his or her reputation in the
group and how others perceive him or her (e.g., Kilduff & Krackhardt,
1994; Mehra, Dixon et al., 2006). For example, a central position in the
network is often related to prestige within the group (Mehra, Dixon et al.,
2006).
These central consequences of network structure and position – that is,
coordination and social influence, access to resources, and reputation –
also play important roles in the leadership literature, as noted above. First,
many leadership scholars argue that leadership involves coordinating
and influencing people to attain common goals (e.g., Kaiser et al., 2008).
Influencing and motivating others is in many ways a social process, and
earlier research on social networks has shown that a group’s network
structure can foster the leader’s efforts to motivate and coordinate group
members’ behaviour to achieve group goals. For example, research shows
that how group or team members are connected to each other affects
team performance (Balkundi & Harrison, 2006): when more people in
the team are connected to each other – that is, a high-density network –
the team achieves higher performance. One argument why this group
network structure benefits group performance is that when people in the
network know each other, it enhances the maintenance of group norms
and trust and eases the monitoring of whether group member behaviour is
in accordance with group goals (e.g., Coleman, 1988).
SCHYNS_9781785367274_t.indd 259
10/11/2017 15:20
260 Handbook of methods in leadership research
Second, leadership involves acquiring the resources needed by the group
to achieve their goals – often in a competitive environment (e.g., Kaiser
et al., 2008). Social capital theory in particular argues that personal and
group success depends in many ways on network-based resources – social
capital (e.g., Burt, 1992, 2005; Lin, 2001) such as information, advice
and credentials. In addition, as mentioned above, people’s networks can
enhance group norms and trust between people, which in turn enhance
goal attainment (Coleman, 1988). Most of the earlier research on social
networks and leadership appear to have used social capital theory as a
theoretical framework (for a review, see Carter et al., 2015).
Among network scholars, two primary explanations have been used to
explain why the characteristics of social networks are important channels
for resources – that is, network structure and social resources (e.g., Burt,
1992). First, network scholars have argued that resources such as information are more homogeneous within the group rather than between groups
(e.g., Burt, 1992; Granovetter, 1973). Thus, to obtain additional resources,
one must often reach outside one’s typical social circle. How one does this
depends in turn on one’s network structure and related ties. Specifically,
Granovetter (1973) famously argued that weak ties, such as acquaintances,
often provide a bridge between different social groups and, consequently,
enable access to information and opinions that are not available in one’s
own social circles. Along similar lines, Burt (1992) argued that when a person’s network includes people who do not know each other, this supports
access to resources. Specifically, when a network includes people who do
not know each other, there are ‘structural holes’ (i.e., missing connections
between persons) in the network. Furthermore, when there are structural
holes in a person’s network, that person is in a brokerage position between
the people who often represent different groups. Figure 11.1 presents a
network diagram to illustrate weak ties and structural holes in a network.
In the network diagram, actors are typically represented by dots, and lines
indicate relations between actors. For example, Heather is in the middle
of the diagram, and she has structural holes in her network, because she
has connections, for example, to Jan and Gretel, who themselves are not
connected. Furthermore, in the network literature the argument has been
that these types of relations that connect different groups are often weak
ties, called ‘bridging weak ties’ (Granovetter, 1973, p. 1371).
Second, social resources theory (Lin, 1982, 2001) emphasizes that the
resources available through social networks depend on the social contact’s position or rank. That is, a social contact’s position in the societal
or organizational hierarchy is important, because it enables access to
resources, such as credentials and social influence (ibid.). In addition,
social resources theory postulates that actors’ own status and weak rather
SCHYNS_9781785367274_t.indd 260
10/11/2017 15:20
A social network approach to examining leadership ­261
Ann
Les
Bernard
Heather
Gretel
Jan
Paul
Ed
Chrissie
Teresa
Gil
Peter
Figure 11.1
An example of network diagram
than strong ties in the network are related to high-status contacts (Lin,
2001). Earlier research on leadership and social networks also supports
the argument that the characteristics of social networks are related to
leadership and group outcomes, presumably because they enable access
to resources. For example, earlier research on leaders’ positions in social
networks found that the position of work group leaders in external and
internal networks is related to the effectiveness of their work group
(Mehra, Dixon et al., 2006). Another study showed that a focal group’s
ties to other group leaders are related to group effectiveness (Oh, Chung,
& Labianca, 2004). Earlier research has also established that a relationship
exists between network position and power (Brass & Burkhardt, 1993).
Finally, a central question in leadership research concerns how leadership evolves over time; that is, how an individual develops as a leader
and how leadership evolves between leaders and followers within a
social environment (e.g., Day et al., 2014). This leadership development
process is inherently a social phenomenon, because the focus is on how
the social exchange between leaders and followers evolves over time and
how others perceive the leader’s characteristics and behaviour. The social
network approach can be used to examine, for example, how leadership
evolves as leaders’ connections with others change in social networks
over time. Earlier research on leadership and networks has shown that
SCHYNS_9781785367274_t.indd 261
10/11/2017 15:20
262 Handbook of methods in leadership research
leaders’ ­position in a social network is related to their reputation as a
leader (Mehra, Dixon et al., 2006), others’ perception of their charisma
(Balkundi, Kilduff, & Harrison et al., 2011), their status in the organization (Venkataramani et al., 2010), and their promotion to a leader role
(Parker & Welch, 2013). For example, Balkundi and colleagues (2011)
found that a leader’s central position in a social network is related to
others’ ­perception of his or her charisma as a leader. Earlier research
also indicates that perceptions of leader charisma tend to be contagious
through network ties (Pastor et al., 2002); that is, employees’ perceptions
of leader charisma are affected by the perceptions of their network ties.
Thus, the characteristics of leaders’ social networks play an important
role in how others perceive them as a leader and in their promotion to a
leadership role.
The social network approach has also focused on how one perceives
relations between people and the antecedents and consequences of
the accuracy of these perceptions – that is, cognitive networks (e.g.,
Krackhardt, 1990; Kilduff & Krackhardt, 1994; Krackhardt & Kilduff,
1999). For example, Kilduff and Krackhardt (1994) argued that ‘the
performance reputations of people with prominent friends will tend to
benefit from the public perception that they are linked to those friends’
(p. 89). Thus, they measured people’s perceptions of friendship ties in
the organization: ‘Who would this person consider to be a personal
friend? Please place a check next to all the names of those people who
that person would consider to be a friend of theirs’ (p. 91). They also
measured actual friendship ties: they asked participants to report their
personal friendship ties in the workplace. A friendship tie was indicated
when both parties of the reported friendship tie agreed that they were
friends. Study participants rated a focal participant’s performance reputation. Friends’ prominence was indicated both by network measures
(to what extent others asked him or her for advice) and formal status.
The results showed that the perceived prominence of the focal person’s
friend was related to the person’s own reputation. Interestingly, the
actual prominence of the person’s friend was not related to performance
reputation.
After this brief presentation about the relevance of the social network
approach in leadership research, I next focus on network methodology: network and research design; methods for sampling and collecting
network data and measuring social networks; and data analysis methods
(also see the Appendix for central network concepts and terms).
SCHYNS_9781785367274_t.indd 262
10/11/2017 15:20
A social network approach to examining leadership ­263
METHODOLOGY OF THE SOCIAL NETWORK
APPROACH
A central question for a researcher engaged in social network research is
whether to use whole or personal network design. Whole network design
refers to a bounded social environment, such as a school class or a department in an organization, in which all or most of the people report their
relations or connections to each other (e.g., Wasserman & Faust, 1994).
For example, a researcher asks all employees to indicate their relationships
with people at their department in terms of interactions, such as ‘adviceseeking’ (‘Who are the people you ask for advice?’) and social support
(‘Who are the most important people to you as a source of personal
support?’). Because all or most of the people in the focal unit report their
ties to each other, it is possible to define who relates to whom in the focal
network. The result of whole network design is a network of relationships
between all study participants in a given set (e.g., Borgatti, Everett, &
Johnson, 2013).
A personal network refers to a focal individual’s direct ties and her or
his perceptions of the relations among these ties in the network. These
networks are often called ‘ego networks’, and the focal individual who
reports his or her ties is called the ‘ego’ in network literature. Each
named network person or tie is an ‘alter’ (e.g., Wasserman & Faust,
1994). For example, a survey could ask study participants to name
persons with whom they have discussed important matters during the
last six months. These named persons (alters) would then represent the
participants’ personal networks. In addition, typical personal network
surveys also ask participants to report their perceptions about the extent
to which alters in the network are related to each other (alter–alter ties).
Thus, in personal network design, the aim is to collect data on each participant’s personal social environment, and there is typically no information about how study participants are connected to each other.
Whether whole or personal network design is being used has consequences for data collection, sampling, network measures and data analysis. Because whole network design aims to represent all ties among study
participants, it requires a higher response rate among participants than a
study based on personal network design (see, e.g., Costenbader & Valente,
2003). However, a whole network design enables a researcher to more fully
use social network analysis as a statistical method, and many important
network measures assume a whole network design (e.g., Wasserman &
Faust, 1994).
SCHYNS_9781785367274_t.indd 263
10/11/2017 15:20
264 Handbook of methods in leadership research
Sampling
Boundary specification
A sampling issue specifically related to whole network design is called
‘boundary specification’ (e.g., Wasserman & Faust, 1994). This specification defines who the actors are that can be part of a particular network;
that is, network boundaries, and what types of relations between these
actors are relevant to examine (Laumann, Marsden, & Prensky, 1983).
The research questions somewhat indicate how to define the study sample.
However, this sampling problem is highlighted in network research,
because network ties may not be limited according to formal boundaries, such as by organization, department, or workplace. For example,
a researcher may be interested in examining with whom people discuss
important work-related matters. Bounding network ties to include participants’ work groups in the workplace, for example, may leave many
network ties outside the study, because people may discuss their work with
many people outside their work group. Thus, defining whom to include in
the network study is important to obtain valid information about social
networks and related resources. Borgatti and colleagues (2013) suggest
that if the research question does not define clear boundaries regarding who could be possible network ties, a personal network design is an
option. Study participants are then free to name their network ties according to the research question and related study instructions. For example,
Carroll and Theo (1996) used a personal network design in their study
examining managers’ social networks. They were interested in discovering
how managers’ and non-managers’ social networks differ from each other.
They used data based on a General Social Survey that is a representative
sample of the US population. However, most of the studies on social
networks and leadership have capitalized on convenience samples (for a
review, see Carter et al., 2015).
In general, Laumann and colleagues (1983) suggest two basic strategies
for approaching boundary specification issues. The realist strategy for
boundary specification argues that social units or groups have boundaries that are often recognized by all of the members. Formal social units
such as an organization, workplace, or school class are examples of units
with clear boundaries for their members. Thus, a researcher sets network
boundaries according to the boundaries perceived by the members of
a social group or unit. A researcher using a nominalist strategy sets
boundaries for a network according to her or his theoretical approach
and related research questions; that is, the researcher defines the network
boundaries. For example, a researcher may be interested in examining
communication networks within work groups in an organization, thus
SCHYNS_9781785367274_t.indd 264
10/11/2017 15:20
A social network approach to examining leadership ­265
she or he may limit the network boundaries to include ties within work
groups.
Research Design
A main task of the research design is often to identify supporting evidence
for the causal claims between study variables; for example, the higher the
quality of working relationship between a leader and a follower is, the
higher the follower’s job satisfaction (e.g., Shadish, Cook, & Campbell,
2002). In other words, through the research design, the researcher tries
to minimize threats to his or her claim about the argued relationships
between study variables; that is, threats to internal validity. A basic classification between research designs is between cross-sectional, longitudinal, and experimental study designs. In a cross-sectional design, study
variables are measured at the same time. A study based on cross-sectional
design cannot make causal claims between study variables, because causal
claims require that the possible cause should precede the consequence in
time (ibid.). For example, Goodwin, Bowler, and Whittington (2008) used
a cross-sectional design and found that both the leader’s and the follower’s
network position was related to the quality of the working relationship
with the leader. In longitudinal research design, study variables are measured several times over time. A longitudinal design offers the possibility to
examine associations between variables over time. For example, Balkundi
and colleagues (2011) found that a leader’s central position in a social
network affects how others perceive their charisma as a leader. Because
they used a longitudinal design, they also argued that network position
has an effect on leader charisma rather than charisma explaining a central
position in the network. In addition, a longitudinal design is particularly
useful in modelling change over time, such as to whether the characteristics of a social network change over time (e.g., Snijders, Van de Bunt, &
Steglich, 2010). However, the problem with a longitudinal design is that it
cannot exclude the possibility of confounding variables that may explain
the relation between two variables.
Many scholars regard an experimental research design as the ‘golden
standard’ (West & Thoemmes, 2010) for research, because it can rule out
possible confounding variables for a causal relationship between variables
(Shadish et al., 2002). A typical experimental design first includes a baseline measurement for all of the study participants, then, study participants
are randomly assigned to experimental and control groups. Participants in
the experimental group receive a ‘treatment’ such as leadership training,
and the control group continues without treatment. Finally, a researcher
compares the experimental and control groups to determine whether
SCHYNS_9781785367274_t.indd 265
10/11/2017 15:20
266 Handbook of methods in leadership research
there are differences of interest between the groups according to the study
outcome such as leadership quality. For example, Lam and Schaubroeck
(2000) conducted a quasi-experimental field study in which they examined
whether service-quality leadership has an effect on unit-level service effectiveness and quality. They had two experimental groups and a control
group. The first experimental group included front-line employees who
received service-quality training; the second experimental group consisted
of employees who were perceived to be opinion leaders by their managers and who also received service-quality leadership training. For participants in the control group, no leadership training was provided. The
results showed that customers rated the group led by opinion leaders who
received service-quality leadership training as the most effective in service
delivery. Opinion leaders who act as service-quality leaders may have more
credibility when implementing new practices among employees than those
who are not perceived as opinion leaders. Network scholars further argue
that opinion leaders typically have a central position between network ties,
which enhances their social influence (e.g., Burt, 2005).
Unfortunately, the majority of research on social network and leadership is still based on cross-sectional research. Thus, causal claims about
the effects of social networks on leadership, or vice versa, need more
elaboration with stronger research designs. There are an increasing
number of longitudinal studies that offer better support for the role of
social networks in leadership and its outcomes (e.g., Balkundi et al., 2011).
However, scholars in the field have rarely capitalized on experimental
research design to improve the evidence regarding the causal processes
between social networks and leadership.
Network Methods for Data Collection
There are different methods for acquiring network data: surveys (for a
review, see Marsden, 2005), registers (e.g., Galunic et al., 2012), electronic sources such as email communication (Kossinets & Watts, 2006),
electronic tags (Ingram & Morris, 2007), and archives (e.g., Padget &
Ansell, 1993). Surveys are the most widely used method (e.g., Marsden,
2005). Below, examples of these methods are presented. I will present
these methods according to whole and personal network study designs,
although the same kind of network content instructions can be used
in both designs. However, a whole network design typically uses name
rosters of all study participants, since the researcher knows the network
boundaries. Instead, in a personal network design, the researcher does not
know the possible network ties beforehand.
SCHYNS_9781785367274_t.indd 266
10/11/2017 15:20
A social network approach to examining leadership ­267
Whole network design
In a whole network design, the sociometric method is the typical way to
collect data. In this method, a survey often provides a name roster of all
study participants, on which study participants indicate with whom they
interact or communicate. Specifically, a study participant sees a roster
including the names of all people in the given unit, such as an alphabetical
list of all employees in the department, and she indicates, for example, with
whom she ‘discusses important matters’ or whom ‘she asks for advice’.
The whole network instrument can also measure ties outside the organization, such as interorganizational collaboration among organizations. For
example, Figure 11.2 presents an example of a roster from a study in which
the leaders indicated interorganizational collaboration and communication ties of their organization (Jokisaari & Vuori, 2010). This survey asked
organizational leaders to note both their collaboration and communication with other organizations in the field. Other examples of instructions
used in whole network instruments are provided in Table 11.1.
There are also free-recall instruments in whole network design in which
participants are asked to report their network ties with open answers –
that is, no name roster is available. However, using a name roster makes
reporting easier for respondents by suggesting possible network ties and
reducing measurement error due to forgotten relations (Marsden, 2011).
That said, respondent fatigue can be a problem if the name roster includes
a high number of names. In that case, it could be wise to divide the roster,
for example according to department or hierarchy. For example, in a
study with 260 potential names for the name roster, the authors divided
the name roster according to departments and work groups to avoid
respondent fatigue (see Sparrowe & Liden, 2005).
Personal network design
A personal or ego network measurement survey typically involves three
parts: (1) a name generator; (2) name and relationship interpreters; (3)
questions related to ties between alters (network structure). First, the
name generator refers to instructions in a survey by which the researcher
encourages study participants to report their social relations, such as
‘With whom do you discuss important matters about work?’ There are
many options for the content of the name generator; that is, what types
of networks a researcher wants to examine, depending on the research
questions. In addition, a survey can include one or multiple name generators (e.g., Burt, 1997). Figure 11.3 presents an example of an ego network
survey, and Table 11.2 provides examples of name generators in leadership and management research. For example, Rodan and Galunic (2004)
examined the role of managers’ social networks in their job performance.
SCHYNS_9781785367274_t.indd 267
10/11/2017 15:20
268 Handbook of methods in leadership research
Instructions
Below you will find a name roster of organizations in alphabetical order. Please indicate
the organizations with which your organization has collaboration and/or regular
communication related to employment services. In other words, please answer
questions below by marking organizations that fit the questions, otherwise leave your
answer blank.
(A) Please indicate the employment offices with which your organization has or has had
collaboration in employment services. Collaboration may include shared projects,
services or development activities related to employment services.
(B) Please indicate the employment offices with which your office has or has had
regular communication regarding employment services. The communication may
include, for example, face-to-face discussions, meetings, or e-mails.
A
B
Collaboration
Communication
1 = Formal, e.g., based
1 = Daily
on contract
2 = Weekly
2 = Informal
3 = Monthly
3 = Multiplex
4 = Seldom
Organization:
Alajärvi
1
2
3
1
2
3
4
Alavus
1
2
3
1
2
3
4
Anjalankoski
1
2
3
1
2
3
4
Eno
1
2
3
1
2
3
4
Enontekiö
1
2
3
1
2
3
4
Espoo, Keskus
1
2
3
1
2
3
4
Espoo, Tapiola
1
2
3
1
2
3
4
Eura
1
2
3
1
2
3
4
Forssa
1
2
3
1
2
3
4
Figure 11.2
An example of whole network survey
They used a personal network design and four name generators to obtain
information about managers’ advice ties, innovative ties, buy-in ties and
confidant ties (see Table 11.2).
The position generator method asks directly whether a respondent
SCHYNS_9781785367274_t.indd 268
10/11/2017 15:20
A social network approach to examining leadership ­269
Table 11.1
Examples of instructions in whole-network designs
Types of relations
Instructions [in the column following each person’s name,
participants were asked the following:]
Advice
‘Check the names of individuals to whom you go for help
or advice about work-related matters’ (Sparrowe & Liden,
2005, p. 517)
‘Check the names of individuals you consider to be your
personal friends’ (Gibbons, 2004, p. 247)
‘Check the names of individuals to whom you go to discuss
confidential issues or problems at work’ (Sparrowe &
Liden, 2005, p. 517)
‘Which person is influential: he/she has clout in this
company?’ (Sparrowe & Liden, 2005, p. 517)
Friendship
Trust
Influence
knows people in certain occupations (e.g., Lin et al., 2001). For example,
‘Here is a list of jobs. Would you please tell me if you happen to know
someone (on a first-name basis) having each job?’ (Lin, 1999, p. 477). This
method has often been used to examine the role of social capital in career
success (for reviews, see Lin, 1999, 2001).
Second, a personal network measurement includes name and relationship interpreters that acquire information about the characteristics of each
named network person (alter) and the nature of the relationship between
the respondent (ego) and the named network person. A typical question
related to the characteristics of the alter concerns his or her occupational
status or position in the organizational hierarchy. In social capital theory,
the status of the alter indicates resources in the network: the higher the
status of the network connections, the more resources such as influence
and credentials are potentially available to the focal person. Relationshipquality questions typically relate to the strength of the ties between the
ego and the alters (Marsden & Campbell, 1984). A typical way to assess
tie strength is to ask about ‘closeness’ of the relationship between the ego
and the alter such as ‘How close do you feel to this person?’ In addition,
the frequency of contact between persons has also been an indicator of tie
strength, as well as the duration of the relationship, and some researchers
use a combination of these measures. The interpretation is that the closer
the relationship, and/or the higher the meeting frequency, the stronger the
tie strength. In contrast, researchers typically operationalize the number
of ‘weak ties’ by counting those ties that are rated to have low closeness
and/or meeting frequency. Furthermore, researchers often ask about the
content of the relationship between the ego and the alter, such as ‘What
SCHYNS_9781785367274_t.indd 269
10/11/2017 15:20
270
SCHYNS_9781785367274_t.indd 270
10/11/2017 15:20
Figure 11.3
1 2 3
PERSON 2
1 2 3
PERSON 3
1 2 3
PERSON 4
1 2 3
PERSON 5
1 2 3 4
1 2 3 4
1 2 3 4
1 2 3 4
1 2 3 4
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
1 2 3
PERSON 1
An example of personal network survey
4 = Other, what?
3 = Leader
2 = Supervisor
1 = Employee
4. His or her rank
3. How close you are?
(1 = Distant, 5 = Very close)
3 = Other, what?
2 = supervisor
1 = co-worker
2. What is your relationship
with this person (e.g.,
co-worker, supervisor)?
1. First name of the person?
People often discuss their important matters with others. If you think of the people in your workplace and look back over the
last few months, who are the people with whom you have most often discussed important matters related to your
work or workplace?
A social network approach to examining leadership ­271
Table 11.2
xamples of name-generating questions in personal network
E
designs
Name generator
Questions
Discuss important
matters
‘People often discuss their important matters with others.
If you think of the people in your workplace and look
back over the last few months, who are the people you
have most often discussed important matters related to
your work or workplace?’ (Jokisaari, 2013, p. 100)
‘Getting your job done on a daily basis as a manager often
requires advice and information from others. Who are
the key people you regularly turn to for information and
work-related advice to enhance your ability to do your
daily job?’ (Rodan & Galunic 2004, p. 549)
‘Over the last six months, are there any work-related
contacts from whom you regularly sought information
and advice to enhance your effectiveness on the job?’
(Podolny & Baron, 1997, p. 691)
‘Most people rely on a few select others to discuss sensitive
matters of personal importance, i.e., “confidants” on
whom they rely for personal support. Who are the
key people in your work environment that you regard
as your most important people as source of personal
support?’ (Rodan & Galunic, 2004, p. 549)
‘New ideas often require support from others without
which you cannot proceed. Who are the key people that
provide essential support to new initiatives?’ (Rodan &
Galunic, 2004, p. 549)
‘List anyone that you feel is a significant part of your
professional network. One way to identify these people is
to go through your address book, and ask “Is this person
significant in my professional network?”’ (Chua et. al.,
2008, p. 442)
Advice
Support
Buy-in
‘Professional ties’
is your relation to this person (e.g., co-worker, friend)?’ Answers to this
question can also be used to categorize strong (e.g., friend) and weak ties
(acquaintance). Table 11.3 provides examples of questions related to the
characteristics of the relationship and the alter.
Finally, a personal network survey includes questions about relations
between alters: participants are asked to evaluate the extent to which
named network persons interact with each other or know each other. These
questions about alter–alter ties enable a researcher to define the characteristics of the network structure. For example, ‘network density’ refers
SCHYNS_9781785367274_t.indd 271
10/11/2017 15:20
272 Handbook of methods in leadership research
Table 11.3
Examples of name and relationship interpreters
Name and relationship interpreters
Name interpreter
Relationship type
Status
Demographics
Relationship interpreters
Tie strength
‘What is your relationship with this person (e.g.,
co-worker, supervisor)?’
‘Is this person’s rank (1) higher than yours; (2) equal
to yours, (3) lower than yours?’
‘Person’s occupation:’
‘Gender of the person:’
‘Ethnicity of the person:’
‘How close are you to this person?’ ‘How often do
you meet?’
to how well people in a focal individual’s network are connected to each
other. The more they are connected to each other, the higher the network
density. Figure 11.4 gives an example of questions related to alter–alter ties.
It is also important to note that reporting personal networks can be time
consuming for the respondent depending on how many name generators
and related alter, relationship, and alter–alter ties questions a researcher
uses. For example, if a researcher asks respondents to name five alters,
this would mean that he or she has to evaluate ten alter–alter ties (i.e.,
N(N – 1)/2; N 5 number of ties); if a respondent names ten alters, there
would be 45 alter–alter ties to evaluate, and with 20 alters, the number of
alter–alter ties would be 190. Thus, the researcher must be careful when
planning the personal network design, because the time and cognitive
demands made on respondents can increase rapidly with an increasing
number of network ties. Furthermore, there have been discussions about
how well people remember relevant ties when asked to recall their network
ties (e.g., Adams & Moody, 2007; Kogovsek & Ferligoj, 2005). In general,
people tend to remember rather well their relations with those with whom
they interact regularly or who represent their important network ties (for
reviews, see Brewer, 2000; Marsden, 2011).
DATA MANAGEMENT AND ANALYSIS
Matrices and Network Diagrams
Social network analysis is an analytical technique that is used to represent
relations among actors and to explore the characteristics of networks and
SCHYNS_9781785367274_t.indd 272
10/11/2017 15:20
273
SCHYNS_9781785367274_t.indd 273
10/11/2017 15:20
Figure 11.4
1
2
3
1
1
2
2
3
3
Person 3: Jan
An example of question how to ask alter–alter ties
Person 4: Ann
Person 3: Jan
Person 2: Gretel
Person 1: Heather
Person 2: Gretel
1
1
1
2
2
2
Person 4: Ann
3
3
3
1
1
1
1
3
3
2
2
3
3
2
2
Person 5: Steve
Please indicate whether or not the persons in your network discuss important work-related matters with each other (in each pair;
e.g., ‘Do Person 1 and Person 2 discuss important matters with each other?’) (1 = don’t discuss or seldom; 2 = every now and
then; 3 = very often).
274 Handbook of methods in leadership research
actors’ positions in the networks (e.g., Knoke & Yang, 2008; Scott, 1991;
Wasserman & Faust, 1994). There are also social network analysis software programs for social network analysis, such as Multinet, NetMiner,
Pajek, SIENA, and UCINET (for a review, see Huisman & Van Duijn,
2005). Network scholars represent whole network data using specific data
matrices. In other words, in network analysis data are represented differently from traditional research data, in which rows represent observations
such as study participants and columns represent variables such as tenure
and job satisfaction for each participant. In network analysis, both rows
and columns represent nodes in a network, such as people, and the cells of
the matrix contain information about the ties or relations between nodes.
For example, if ties are dichotomous – that is, the tie between two nodes
exists or does not exist – the cells contain ‘1’ or ‘0’, respectively, depending
on the presence or absence of the tie. Furthermore, a data matrix typically
contains the same number of rows and columns, and the order of actors in
rows and columns is identical. Thus, the data matrix, or sociomatrix, contains information about all the possible ties between nodes. It is common
for a row to represent a focal actor and the column cells represent those
with whom she has ties or relations. In other words, a matrix can be read
as ‘who to whom’ (Monge & Contractor, 2003, p. 36) information about
networks. In the case of non-directional ties, such as who discusses with
whom, the matrix is symmetrical; that is, the value representing a tie from
node i to node j is the same as the value representing a tie from node j to
node i. As an example, in Figure 11.5, there is a symmetric data matrix
between 12 nodes; thus, the value of a tie from node i to node j is the
same as the value of a tie from node j to node i. In other words, the values
of the cells above and below the matrix diagonal are the same. In addition, the values in the diagonal are omitted from the network analysis,
because the diagonal represents a tie from a node to itself. The network
diagram of these data is shown in Figure 11.1.
It is quite common for a researcher to perform data transformation
with network data. That is, network software programs such as UCINET
(Borgatti, Everett, & Freeman, 2002) provide different procedures for
working with data matrices. The common transformation procedures are
symmetrizing, dichotomizing and combining data matrices (e.g., Borgatti
et al., 2013). By symmetrizing, a researcher can create a data matrix in
which all of the relations are reciprocated. For example, a researcher
has network data on friendship ties and defines that friendship tie as
existing only when both parties indicate that the other is a friend. By
symmetrizing the data matrix, a researcher can create a new data matrix
in which a friendship tie exists only if the friendship tie is reciprocated.
Dichotomizing includes a procedure in which the valued tie is transformed
SCHYNS_9781785367274_t.indd 274
10/11/2017 15:20
An example of a data matrix
Figure 11.5
275
SCHYNS_9781785367274_t.indd 275
10/11/2017 15:20
276 Handbook of methods in leadership research
to be dichotomous. For example, a researcher could ask study participants
to indicate how often they seek advice from their network ties on a scale of
‘1 5 seldom, 2 5 now and then, and 3 5 often’, and he or she might want
to include only those ties where advice seeking happens ‘often’. Thus, he
or she might dichotomize the data matrix by including only ties that have
a value greater than 2. Furthermore, researchers often ask about multiple
relations, such as friendship and advice ties, and may want to combine
these into one data matrix. For example, a researcher might want to
explore networks based on multiplex ties; that is, a tie including different
roles such as friendship and advice seeking. This is made possible by combining friendship and advice seeking data matrices into one data matrix.
Network diagrams, also called graphs and sociograms, are also a
common way to represent networks. A network diagram is a visualization
of a network showing relations between nodes (e.g., Wasserman & Faust,
1994). Nodes are typically labelled by their name, number or another
identifier. Lines represent a relation between two nodes. Arrows represent
directional ties. For example, if node A asks advice from node B, then there
is a line from node A with arrow pointing to node B in the graph. If node B
also asks advice from node A, there is another arrow from node B to node
A or a line with arrows at both ends. If a network diagram represents nondirectional ties, such as friendship ties, the line describing that tie would
be without an arrowhead or arrowheads at both ends. Figure 11.1 shows
a graph of 12 nodes and the relations or links between them. Because this
network diagram has only lines with arrowheads at both ends, it is a nondirectional graph. A researcher typically has leeway to decide how she or
he places nodes and lines in a graph. In addition, software programs for
network visualization such as Gephi, NetDraw, NetMiner and Pajek offer
many options that a researcher can use to visualize networks. In all, ‘the
precise placement of nodes and lengths of lines in a network diagram is
somewhat arbitrary, although some versions might be clearer than others.
Constructing insightful sociograms is as much an artistic as a scientific
activity’ (Knoke & Yang, 2008, p. 46).
Measures of Networks
The network approach has introduced specific measures for examining the
attributes and characteristics of social networks and actors’ positions in
these networks. Whole network concepts and related measures can relate
to the actor, dyadic, triadic, group, and whole network levels (Wasserman
& Faust, 1994). Personal network measures relate to the characteristics of
the focal actor’s network and can characterize ego–alter ties, alters, and
network structure.
SCHYNS_9781785367274_t.indd 276
10/11/2017 15:20
A social network approach to examining leadership ­277
Whole network design
Actor-level measures Network research on leadership has often concentrated on the actor level and measured how leaders’ positions in the networks are related to leadership outcomes and development. The concept
perhaps most often used to indicate position in the network is network
centrality and its indices (in-degree, out-degree, closeness, eigenvector, and
betweenness centrality; Borgatti et al., 2013; Freeman, 1979; Wasserman
& Faust, 1994). In-degree centrality and out-degree centrality refer to how
many ties a focal actor has in the network. Specifically, in-degree centrality relates to how many others in the network name or indicate that they
are related to the focal individual, such as those who ask for advice from
him or her. In turn, out-degree centrality indicates from how many nodes
a focal individual asks for advice, for example. Closeness centrality tells
how ‘reachable’ all others in the network are to a person – in other words,
how easily an individual can access resources or communicate, directly or
via intermediaries, with all others in the network. Eigenvector centrality
takes into account the centrality of the network nodes to whom a focal
actor is connected. In other words, eigenvector centrality sums the focal
actor’s ties to others by weighing them by the centrality of those others
(e.g., Borgatti et al., 2013). For example, Mehra, Dixon and others (2006)
found that leaders’ (eigenvector) centrality both in the friendship network
of leaders and among friendship networks of their work group are related
to group performance.
Betweenness centrality is also an important centrality measure that indicates brokerage roles in the whole network and takes into account both
direct and indirect ties; that is, the extent to which actors connect actors
that themselves are not connected to each other. For example, Galunic
and colleagues (2012) found that employees whose leaders showed high
betweenness centrality in networks – a brokerage role – were rated as being
more useful and helpful by others than employees whose leaders did not
have brokerage role. In addition, young professionals whose supervisors
were active in their work groups’ internal and external communication ties
showed higher promotion likelihood and less turnover than others (Katz
& Tushman, 1983). It is important to note that researchers often standardize the values of the indices of centrality so that they are comparable
between networks of different size (e.g., Knoke & Yang, 2008). This is also
an option available in network analysis software such as UCINET.
As an example, Table 11.4 shows the degree, closeness, betweenness
and eigenvector centrality measures (non-standardized) among persons in
the network shown in Figure 11.1. These were analysed using UCINET
(Borgatti et al., 2002).
As seen, Heather has the highest value for degree centrality (5). As this
SCHYNS_9781785367274_t.indd 277
10/11/2017 15:20
278 Handbook of methods in leadership research
Table 11.4
1 Jan
2 Gil
3 Heather
4 Chrissie
5 Gretel
6 Bernard
7 Teresa
8 Peter
9 Ed
10 Paul
11 Les
12 Ann
Network centrality measures of network in Figure 11.1
Degree
Closeness
Eigenvector
Betweenness
3.000
3.000
5.000
2.000
3.000
2.000
3.000
3.000
2.000
3.000
2.000
3.000
25.000
27.000
19.000
34.000
21.000
28.000
34.000
34.000
34.000
27.000
28.000
27.000
0.215
0.317
0.448
0.104
0.353
0.247
0.308
0.308
0.104
0.317
0.247
0.308
18.000
8.000
39.500
0.000
28.333
0.000
0.333
0.333
0.000
8.000
0.000
0.500
network is symmetrized, degree centrality is the same as the in-degree and
out-degree centrality measures. Thus, Heather has the highest number of
network ties. Heather also has the highest betweenness centrality value. As
seen in Figure 11.1, Heather is in a brokerage role between three groups
of people. For the closeness centrality measure, Chrissie, Teresa, Peter,
and Ed show the highest values, and Heather, Gretel, and Jan show the
lowest values. These values mean that Heather, Gretel, and Jan have best
opportunities to access others in their network, and Chrissie, Teresa, Peter
and Ed have rather peripheral positions in the network. In other words,
low closeness centrality values indicate ‘high centrality’ in the network, as
provided by the UCINET program. This can also be seen in Figure 11.1.
However, if a researcher uses normalized values of closeness centrality,
this would change the interpretation of closeness centrality values; that is,
high values would indicate high closeness centrality (e.g., Borgatti et al.,
2013).
Dyadic level At the dyadic level of networks, structural equivalence
is an often used measure to indicate the extent to which actors have
ties to similar others (Wasserman & Faust, 1994). Actors are said to be
structurally equivalent when they have the same relations to others in
the network, and they do not have to share relations with each other.
For example, in the network shown in Figure 11.1, Peter and Teresa are
structurally equivalent; that is, they have ties to the same people in the
network. Network theory argues that structurally equivalent actors in a
network are often dependent on the same social ties and related resources
SCHYNS_9781785367274_t.indd 278
10/11/2017 15:20
A social network approach to examining leadership ­279
and are motivated to monitor each other’s behaviour (e.g., Burt, 1987).
In leadership research, Sparrowe & Liden (2005) made innovative use of
the measure of structural equivalence to operationalize sponsorship in the
networks: sponsorship was indicated when the leader and the follower
had the same relationships in the networks of trusted relations. The trust
network included people with whom they ‘discuss confidential issues or
problems at work’; in other words, when a focal leader and a follower had
structurally equivalent positions in the trust network that indicated sponsorship. Furthermore, for a follower, sponsorship increases legitimacy and
assimilation into networks, as the leader shares his or her trusted network
ties with the follower (Sparrowe & Liden, 2005). The results showed that
sponsorship moderated the relation between leader–member exchange
and member influence in the workplace as perceived by others. If sponsorship was high, leader–member exchange was related to member influence,
but when sponsorship was low, leader–member exchange was not related
to member influence.
There are also other measures to indicate similar positions in the
network between two actors, such as regular equivalence. The regular
equivalence measure is not as strict a measure as structural equivalence
when indicating a similar position in the network (e.g., Wasserman &
Faust, 1994). For example, structural equivalence requires that two supervisors have relationships with the same subordinates, but regular equivalence requires only two supervisors who have subordinates.
Triadic level One way to analyse whole networks is to examine triads of
actors or triadic relationships. There are 16 possible ways in which three
actors can form triadic relationships (e.g., Wasserman & Faust, 1994).
Typically, a researcher examines the extent to which a network includes
transitive triads. For example, if person A names person B as a friend, B
names person C as a friend, and person A names person C as a friend, then
they form a transitive triad.
Particularly in research on cognitive networks, scholars have used
theory based on the transitivity principle. The transitivity principle states
that a perceiver will assume consistency in triadic relations; that is, assume
a transitive triad. A related concept is a balance schema that argues that
people tend to perceive positive relationships such as friendship or liking
as reciprocated by those involved in that relationship (e.g., Krackhardt &
Kilduff, 1999). If a focal person perceives that person A considers person
B to be a friend, he or she also assumes that person B will recognize person
A as a friend. In other words, both balance schema and transitivity state
that cognitive consistency is a prime motivation in perceptions of social
networks (e.g., Wasserman & Faust, 1994). For example, Krackhardt
SCHYNS_9781785367274_t.indd 279
10/11/2017 15:20
280 Handbook of methods in leadership research
and Kilduff (1999) asked study participants from four organizations to
evaluate who they thought were friends in their workplace. In other words,
participants were instructed to name friendship ties between people in
their workplace as they perceived them: ‘Who would this person consider
to be a personal friend? Please place a check by the names of those who
that person would consider to be a friend of theirs’ (perceived friendship
network; Krackhardt & Kilduff, 1999, p. 773). In addition, participants
were instructed to indicate their own friends in the workplace (actual
friendship network). An actual friendship tie was indicated when two
people both named each other as a friend. The results showed that the
actual friendship network was a poor predictor of perceived network ties.
The results further showed that study participants tended to perceive their
close and distant ties as being balanced.
(Sub)group level Often within a social network are subgroups of people
who spend more time with each other or are otherwise more connected
to each other compared to non-group members in the network. For
example, a workplace holds subgroups of people who socialize. In the
network literature a cohesive subgroup – that is, a clique – is characterized by the following attributes: mutuality or reciprocity of ties between
people, reachability or closeness of all group members, high frequency of
ties between people, and a higher relative density of the network within
the subgroup compared to the rest of the network (Wasserman & Faust,
1994). In a cohesive group, people typically share opinions, attitudes,
and views, and group norms are easy to maintain. For example, a study
found that subordinates’ proximity in networks related to similarity in the
­perceptions of their leader’s charisma (Pastor et al., 2002).
Whole network level Whole network measures capitalize on the characteristics of the complete measured network. The most common measures
at the whole network level are density and centralization. Density refers
to how many actual ties are present in the network compared to all possible ties in the network. For example, a meta-analysis showed that teams
with high network density showed better performance and willingness to
stay together than teams with low network density (Balkundi & Harrison,
2006). Network centralization concerns the extent to which the centrality of actors varies within the network. If there are few central actors in
the network – that is, network relations are concentrated among a few
­individuals – the whole network shows high centralization. When individual centrality scores are rather evenly distributed among actors in the
network, the whole network is decentralized (Wasserman & Faust, 1994).
Sparrowe and colleagues (2001) examined individuals’ network central-
SCHYNS_9781785367274_t.indd 280
10/11/2017 15:20
A social network approach to examining leadership ­281
ity and work group centralization effects on individuals’ job performance
and on group performance. The results indicated that individuals’ advice
network centrality related to their job performance, whereas group-level
network centralization negatively related to group performance. The
authors reasoned that at the group level, networks that are concentrated
among a few individuals hinder cooperation among people in the group so
that information is not equally shared (see also, Mehra, Smith, Dixon, &
Robertson, 2006). Thus, network structure may also hinder collaboration,
and group performance may then suffer. Researchers adopting the social
network approach have also argued for the use of multilevel analysis to
combine different levels of network research into multilevel models in order
to examine how separate levels may relate to each other (e.g., Monge &
Contractor, 2003). However, earlier research has rarely capitalized on multilevel network models, since: ‘most network data are either transformed to
a single level of analysis (e.g., the actor or the dyadic level), which necessarily loses some of the richness in the data, or are analysed separately at
different levels of analysis, thus precluding direct comparisons of theoretical influences at different levels’ (Contractor, Wasserman, & Faust, 2006,
p. 684).
Personal network measures
Personal network measures can be divided into three categories: measures
of named network persons (measures of alters), characteristics of the
relationships between the ego and alters (measures of relationship), and
characteristics of ties between named network persons (alter–alter ties,
measures of network structure).
Measures of the alters can indicate network-based resources. In social
capital theory, the alter’s social status or position in the organizational or
social structure indicates network-based resources that an ego can access
through his or her network ties (e.g., Lin, 2001). Social status can be indicated by the alter’s educational level, socioeconomic status, or organizational rank, for example. Lin (2001) used three indicators of network-based
resources based on the alter’s social status. ‘Upper reachability’ indicates
the resources at the highest position or status that the ego can reach
through his or her network. ‘Resource heterogeneity’ indicates the variation between highest and lowest social status or position in the network.
‘Extensity of resources’ indicates how many different social statuses or
positions a person can access through his or her social ties. ‘Network
range’ is also used to indicate diversity in the network in terms of different
statuses or group memberships of the network ties. In addition, depending on the research question, other alter characteristics, such as ethnicity
or gender, can be used. A researcher can then calculate the proportion or
SCHYNS_9781785367274_t.indd 281
10/11/2017 15:20
282 Handbook of methods in leadership research
number of alters in each category. Alternatively, a researcher can evaluate
the variation or dispersion of categories in the network.
Measures of relationships typically refer to the relationship type and
quality between the ego and alters. The relationship type between the ego
and alter typically refers to the proportion or number of ties of a given
type, such as the number of leaders, colleagues or friends in the network.
Some researchers also count multiplex ties, such as whether a relation
type is both a friend and a colleague. For example, Carroll and Teo
(1996) examined managers’ personal networks and used a classification
related to their relations with named network contacts. For example, the
number of co-workers was a count of co-workers named as network ties.
The proportion of co-worker ties was calculated by dividing the number
of co-­workers by the total number of the network ties (network size).
Relationship quality between the ego and an alter is typically indicated by
the strength of the tie (Granovetter, 1973). Tie strength can be indicated
by counting the type of ties, such as the number of weak ties, or by averaging the tie strength between an ego and all of her alters. For example,
Carroll and Teo (1996) indicated tie strength by the number of close ties,
which was measured by asking respondents to indicate whether they were
close to each of the people they named. The proportion of close ties was
­measured by dividing the number of close ties by the network size.
Finally, network density and brokerage role in the network are basic
measures for network structure based on personal network design. As
noted above, network density indicates the number of alter–alter ties
divided by the total number of possible alter–alter ties in the network.
Thus, it is typically calculated by excluding ego–alter ties. For example, the
relationship between a pair of named network persons can be coded 0 if
participants report that these persons do not or seldom discuss things with
each other, 0.5 if participants indicate that they discuss things with each
other every now and then, and 1 if participants report that they discuss
things with each other often (e.g., Jokisaari, 2013). Network density is
then the mean of the strength of ties between all named network persons –
namely, the average level of interconnection between named network ties.
For example, Parker and Welch (2013) found that the density of collaboration networks was negatively related to scientists’ leadership position in
academia. They argued that a low-density network indicates that network
ties are from different social groups, which helps to acquire additional
resources that support a promotion to a leadership role.
A set of measures has been used to indicate a brokerage role in the personal network. As mentioned above, a person in a brokerage role connects
people in the network who are not themselves connected to each other. In
other words, a person who is in the brokerage role has structural holes in
SCHYNS_9781785367274_t.indd 282
10/11/2017 15:20
A social network approach to examining leadership ­283
Table 11.5
etwork constraint and density measures of network in
N
Figure 11.1
1 Jan
2 Gil
3 Heather
4 Chrissie
5 Gretel
6 Bernard
7 Teresa
8 Peter
9 Ed
10 Paul
11 Les
12 Ann
Network constraint
Network density
0.611
0.611
0.382
1.125
0.333
1.125
0.840
0.840
1.125
0.611
1.125
0.840
0.333
0.333
0.200
1.000
0.000
1.000
0.667
0.667
1.000
0.333
1.000
0.667
his or her network (e.g., Burt, 2005). A brokerage role is assumed to offer
the following benefits for a person: heterogeneous information and point
of views, opportunity for control of what information one shares with
others, and early access to new opportunities (Burt, 1992). Perhaps the
most often used measure to indicate brokerage role – that is, structural
holes, based on personal networks – is the network constraint measure (for
a review, see Burt, 2005). In fact, the network constraint measure indicates
network closure – in other words, a lack of brokerage opportunities or
structural holes (Burt, 1992). There is a lack of brokerage opportunities
when a personal network is dense (high network density) or when network
ties are connected to each other through a central mutual contact (ibid.).
As an example, Table 11.5 presents values for both network density and
network constraints for the people in Figure 11.1. As seen, Gretel (0.33)
and Heather (0.38) have the lowest value for network constraints as well as
the lowest network density values (Gretel: 0.0, Heather: 0.2) and thus the
best opportunities for having a brokerage role in the network. In contrast,
Les, Ed, Chrissie, and Bernard have the highest values for both measures.
For example, they have the highest network density value (1.0), which
indicates that all of their network ties are connected to each other, as can
be seen in Figure 11.1.
Earlier research has also shown the benefits of brokerage roles. For
example, Burt (2007) examined supply-chain managers’ brokerage opportunities using their personal networks. In the network procedure, managers were asked to name persons with whom they most often had discussed
SCHYNS_9781785367274_t.indd 283
10/11/2017 15:20
284 Handbook of methods in leadership research
matters related to supply-chain issues. Then they were asked to indicate
the perceived discussion between each pair in their personal network
about supply-chain matters; that is, how often each pair discussed supplychain issues (‘often’, ‘sometimes’, or ‘rarely’). The less that the pairs of
network contacts discussed with each other – that is, low network constraints – the more likely the focal manager was to play a brokerage role.
The results showed that managers whose networks were characterized
by high network constraint had a lower salary and lower performance
evaluations than managers who had brokerage opportunities (low network
constraint) in their networks. In addition, Rodan and Galunic (2004)
found that structural holes in managers’ personal networks related to their
performance.
I have presented above measures of social networks based on both
whole and personal network designs. Although network scholars typically
use different measures depending on whether they used a whole or personal network design, there is a debate among network scholars regarding to what extent measures of network that were developed based on a
whole network design are suitable to characterize personal networks (e.g.,
to what extent the measure of betweenness centrality gives similar results
based on whole and personal network data, Marsden, 2002).
Statistical Inference
Besides providing descriptive statistics for the characteristics of social
networks and actors’ positions in these networks, the social network
approach offers methods for statistical inference based on social network
data. These statistical procedures include quadratic assignment procedure
(QAP), exponential random graph models (ERGMs, also referred to as p*
models), and actor-oriented models (e.g., Borgatti et al., 2013; Wasserman
& Faust, 1994). It is important to note that the assumptions of general or
traditional statistical analyses, such as the independence of observations,
are not valid when statistical inference is based on whole network data.
The main reason is that whole network data consist of non-independent
observations, and thus traditional significance tests do not apply for
testing the statistical significance of estimates (e.g., Wasserman & Faust,
1994). Instead, statistical significance tests related to social networks are
often based on non-parametric permutation tests – that is to say, the use of
randomized samples (ibid.). The observed estimate, such as a correlation,
is compared to estimates based on the distribution of simulated random
samples in order to conclude whether its occurrence is more likely than
one would expect to observe by chance. Statistical inference based on QAP
correlation and regression analysis is based on these permutation tests –
SCHYNS_9781785367274_t.indd 284
10/11/2017 15:20
A social network approach to examining leadership ­285
that is, randomized samples (e.g., Krackhardt, 1987). These tests provide
statistical information to examine relations among networks or between
networks and individual attributes. For example, a researcher may be
interested in examining whether people are more likely to ask for advice
from friends than one would expect by chance. He or she could examine
this research question using QAP regression – that is, whether advice
seeking and friendship network matrices are related to each other. QAP
regression can be used when the dependent variable is a binary variable
(e.g., whether people share a relation: yes/no; e.g., Borgatti et al., 2013).
For example, Gibbons (2004) examined the role of friendship and advice
networks in professional values. The research questions focused on, for
example, the extent to which people seek advice from people whose professional values converge with their own values over time and the extent to
which friends’ values converge over time. The study data included whole
network data based on friendship and advice-seeking ties in workplaces,
as well as information on the study participants’ professional values.
She examined study hypotheses using QAP regression, and the results
showed that friendship networks were related to changes in professional
values and that changes in professional values were related to changes in
advice-seeking ties. General programs for social network analysis such
as UCINET provide procedures for QAP correlations and regression
analysis.
The general rationale behind using ERGMs is that ‘the observed
network is seen as one particular pattern of ties out of a large set of possible patterns. In general, we do not know what stochastic process generated
the observed network, and our goal in formulating a model is to propose a
plausible and theoretically principled hypothesis for this process’ (Robins,
Pattison, Kalish, & Lusher, 2007, p. 175). In other words, ERGMs offer
tools for a researcher to examine what kind of social processes as indicated
by the substructures of a network could explain the current patterns of
the whole observed network. Thus, the question is often whether certain
network properties or characteristics in the network are likely to occur
more than one would expect by chance (e.g., Robins et al., 2007). For
example, a researcher can ask whether the level of triads or reciprocity in
the network is higher than one would expect by chance in a given network.
For example, Lazega and Pattison (1999) examined substructures within
collaboration networks in an organization and asked, among other
questions, whether resource exchanges among employees show regular
interaction patterns that are not limited to the dyadic level. They found,
for example, that network triads are more likely to occur than one could
expect based on chance. The estimation of an ERGM requires specific
software, such as PNet and Statnet.
SCHYNS_9781785367274_t.indd 285
10/11/2017 15:20
286 Handbook of methods in leadership research
Actor-driven or actor-oriented models provide tools to examine
network changes and changes in covariates over time (e.g., Snijders, 2005).
With these models, both the role of network characteristics and actor
choices (initiation of a new tie, dissolving a tie) in network changes can
be modelled. For example, a researcher can examine whether an advice
network develops towards increased reciprocity over time. Furthermore,
network characteristics and actor attributes are often interdependent, and
it is important to model how network characteristics and actor attributes
co-evolve over time. Furthermore, it is possible to analyse models on how
changes in social networks and changes in actors’ attributes and behaviour
are related to each other (Snijders et al., 2007). For this purpose, SIENA
(Simulation Investigation for Empirical Network Analysis) software is
available for statistical analyses to examine network changes over time
(e.g., Snijders, 2005). For example, Emery (2012) examined whether an
actor’s attributes (i.e., emotional abilities) are related to others’ perceptions of her leadership characteristics over time. In the study, participants
were asked to nominate who they considered to be leaders among their
group members. According to these nominations, a leadership network
matrix – that is, who nominates whom as a leader – was constructed.
In addition, the study participants evaluated their own emotional abilities. These measures were assessed three times during the study period.
By using actor-oriented models it was possible to examine both network
effects (e.g., reciprocity, transitivity) and actor effects (emotional abilities)
on the emergence of others’ perceptions of leadership over time. There
have also been studies on network changes in the organizational context
related to how/why people have certain network positions (e.g., Kossinets
& Watts, 2006; Lee, 2010), changes in work group ties (Schulte, Cohen,
& Klein, 2012) and network changes during transition (Jonczyk, Lee,
Galunic, & Bensaou, 2016), among others (for a review, see Tasselli et al.,
2015).
Case Example of Collecting and Analysing Whole Network Data1
This case example focuses on leaders’ interorganizational collaboration
networks (Jokisaari & Vuori, 2010) to illustrate how to apply network
measures with a whole network design. Specifically, the network measures in the case example include direct ties in the network, structural
equivalence, and brokerage roles. This example also illustrates how social
influence plays an important role in the decision making of organizational
representatives, such as leaders. For example, when leaders have to make
adoption decisions about a new organizational or work practice – that
is, the adoption of an innovation – they have not yet experienced the
SCHYNS_9781785367274_t.indd 286
10/11/2017 15:20
A social network approach to examining leadership ­287
consequences and fit of the innovation in the focal environment. The
social network approach argues that in this kind of situation, leaders use
information about the innovation available through their network ties,
such as those with colleagues (for a review, see Burt, 2005). Specifically,
the literature on social networks and the diffusion of innovations argues
that the responsiveness of the representatives of the organization to a new
practice is in many ways related to their social proximity to other actors in
the field and their adoption behaviour. For example, earlier research has
shown that exposure to the adoption behaviour of others through network
ties influences the focal actor’s adoption of new practices (for a review, see,
e.g., Burt, 2005).
Jokisaari and Vuori (2010) examined the role of both relational (direct
ties) and positional (structural equivalence, brokerage positions) characteristics of interorganizational networks in the adoption of a new practice
over time. In the network questionnaire, leaders of the organizations were
first asked to identify organizations with which they have (1) collaborated
and (2) regularly communicated about matters related to employment
services. Information on collaboration was elicited as follows: ‘Please
indicate the employment offices with which your office has or has had
collaboration in employment services. Collaboration may include shared
projects, services or development activities related to employment services’. Regarding communication, participants were asked the following:
‘Please indicate the employment offices with which your office has or has
had regular communication regarding employment services. The communication may include, for example, face-to-face discussions, meetings, or
e-mails’. A roster with the names of all employment offices in Finland in
alphabetical order was provided to indicate both collaborative and communication partners. This name roster is shown in part in Figure 11.2. In
addition, participants were asked to indicate the time of the relationship
and whether it was still ongoing. Communication ties were almost identical to collaboration ties, which is why we decided to use only collaboration
ties in the analyses.
Adoption among direct ties was measured by the percentage of organizations among the focal organization’s collaborative ties that had previously adopted the new innovation (group training programme: ‘The
Työhön Job Search Programme’); that is, during the previous month or
earlier. Adoption among structurally equivalent ties indicated the percentage of structurally equivalent organizations that had previously adopted
the innovation. Structural equivalence was examined using the blockmodelling procedure (CONCOR) of the UCINET 6 program (Borgatti et
al., 2002). This method identifies groups of actors with similar ties based
on the correlations between the ties and divides them into blocks. The
SCHYNS_9781785367274_t.indd 287
10/11/2017 15:20
288 Handbook of methods in leadership research
number of partitions in this hierarchical clustering method is examined.
The goal is to obtain blocks that show highly correlated patterns of ties
between actors within a block and low correlations with actors outside
the block (Wasserman & Faust, 1994). We ran CONCOR procedures
with three, four and five partitions. When three partitions were used, the
procedure created seven blocks with an average within-block density of
0.18. When four partitions were used, 13 blocks were constructed, and the
average within-block density was 0.75. When five partitions were used,
24 blocks were created, and the average within-block density was 0.87.
However, 25 per cent of the blocks included only single or dyadic actors.
The existence of blocks with a single or two actors is often an unstable
solution, and partitions with such blocks should be avoided (Wasserman
& Faust, 1994). Consequently, in this data set, four partitions gave
the most appropriate solution, which was then used in the subsequent
analyses.
The brokerage position in the network at the whole network level was
indicated by betweenness centrality (Freeman, 1979) and at the local level
by network constraint (Burt, 1992). As noted above, betweenness centrality measures the extent to which an organization is directly connected only
to the organizations that are not directly connected to each other in the
network (Freeman, 1979). It takes into account both direct and indirect
ties in the network and reflects the actor’s brokerage position in the whole
network. The square root of betweenness centrality was used in the analyses because of the non-normality of this measure. The network constraint
accounts for the immediate network of an organization, and it measures
the extent to which the organization has network ties with other organizations that are connected with one another or indirectly connected via a
central actor. In other words, a low network constraint means a greater
likelihood of a local brokerage role.
The analyses were conducted using discrete-time survival analysis
to estimate ‘whether’ and ‘when’ (Singer & Willett, 2003) the adoption
of a new practice occurs. This analysis takes into account the timesensitive nature of the data – in other words, whether the values of
variables may vary with time. The results showed that during the early
phase of the diffusion process, the adoption behaviour of collaborative organizations and local-level brokerage position were related to
adoption among employment offices in the diffusion of the job search
programme. Furthermore, the results showed that adoptions among
structurally equivalent organizations contributed to differentiation –
that is, non-adoption.
SCHYNS_9781785367274_t.indd 288
10/11/2017 15:20
A social network approach to examining leadership ­289
CONCLUSION
The social network approach argues that to understand leadership development, behaviour and outcomes, we have to see them as essentially
depending on the social environment. The social network approach provides both the theory and methodology to examine the characteristics of
the social environment of leadership (Balkundi & Kilduff, 2005; Carter et
al., 2015; Sparrowe, 2014; Sparrowe & Liden, 1997). Specifically, theories
related to the social network approach such as social capital theory argue
that the social environment provides both opportunities and constraints
for leadership, its outcomes, and its development. The methodology of the
social network approach provides means to measure and analyse this social
environment in detail in order to empirically examine the role of social networks in leadership, and there are many methods to gather network data,
such as survey and registers, and different ways to define the content of
networks (e.g., advice seeking, social support, friendship), depending on
the research questions. Furthermore, there are many network measures
to characterize social networks from the individual level, such as network
centrality, to the whole network level, such as network centralization.
Finally, social network analysis as a statistical method provides methods
to examine the characteristics of social networks.
NOTE
1. This example is adapted from Jokisaari & Vuori (2010) with the permission of Oxford
University Press
REFERENCES
Adams, J., & Moody, J. (2007). To tell the truth: Measuring concordance in multiply
reported network data. Social Networks, 29(1), 44–58.
Adler, P., & Kwon, S.-W. (2002). Social capital: Prospects for a new concept. Academy of
Management Review, 27(1), 17–40.
Balkundi, P., & Harrison, D.A. (2006). Ties, leaders, and time in teams: A strong ­inference about
network structure’s effects on team viability and performance. Academy of Management
Journal, 49(1), 41–68.
Balkundi, P., & Kilduff, M. (2005). The ties that lead: A social network approach to leadership. The Leadership Quarterly, 16(6), 941–961.
Balkundi, P., Kilduff, M., & Harrison, D. (2011). Centrality and charisma: Comparing how
leader networks and attributions affect team performance. Journal of Applied Psychology,
96(6), 1209–1222.
Borgatti, S.P., Everett, M., & Freeman, L. (2002). UCINET 6 for Windows: Software for
social network analysis. Harvard, MA: Analytic Technologies.
SCHYNS_9781785367274_t.indd 289
10/11/2017 15:20
290 Handbook of methods in leadership research
Borgatti, S.P., Everett, M.G., & Johnson, J.C. (2013). Analyzing social networks. London,
UK: Sage.
Brass, D.J. (1995). A social network perspective on human resources management. Research
in Personnel and Human Resources Management, 13, 39–79.
Brass, D.J., & Buckhardt, M.E. (1993). Potential power and power use: An investigation of
structure and behavior. Academy of Management Journal, 36(3), 441–470.
Brewer, D.D. (2000). Forgetting in the recall-based elicitation of personal and social networks. Social Networks, 22(1), 29–43.
Burt, R.S. (1987). Social contagion and innovation: Cohesion versus structural equivalence.
American Journal of Sociology, 92(6), 1287–1335.
Burt, R.S. (1992). Structural holes: The social structure of competition. Cambridge, MA:
Harvard University Press.
Burt, R.S. (1997). A note on social capital and network content. Social Networks, 19(4),
355–373.
Burt, R.S. (2005). Brokerage and closure: An introduction to social capital. Oxford: Oxford
University Press.
Burt, R.S. (2007). Second-hand brokerage: Evidence on the importance of local structure for
managers, bankers, and analysts. Academy of Management Journal, 50(1), 119–146.
Carroll, G.R., & Teo, A.C. (1996). On the social networks of managers. Academy of
Management Journal, 39(2), 421–440.
Carter, D.R., DeChurch, L.A., Braun, M.T., & Contractor, N.S. (2015). Social network
approaches to leadership: An integrative conceptual review. Journal of Applied Psychology,
10(3), 597–622.
Chua, R., Ingram, P., & Morris, M. (2008). From the head and the heart: Locating cognition-and-affect-based trust in managers’ professional networks. Academy of Management
Journal, 51, 436–452.
Coleman, J.S. (1988). Social capital in the creation of human capital. American Journal of
Sociology, 94(Supplement), S95–S120.
Contractor, N.S., Wasserman, S., & Faust, K. (2006). Testing multitheoretical, multilevel
hypotheses about organizational networks: An analytical framework and empirical
example. Academy of Management Review, 31(3), 681–703.
Costenbader, E., & Valente, T. (2003). The stability of centrality measures when networks
are sampled. Social Networks, 25(4), 283–307.
Cross, R., & Parker, A. (2004). The hidden power of social networks: Understanding how work
really gets done in organizations. Boston, MA: Harvard Business School Press.
Day, D., Fleenore, J., Atwater, L., Sturm, R., & McKee R. (2014). Advances in leader and
leadership development: A review of 25 years of research and theory. The Leadership
Quarterly, 25(1), 63–82.
Dinh, J.E., Lord, R.G., Gardner, W.L., Meuser, J.D., Liden, R.C., & Hu, J. (2014).
Leadership theory and research in the new millennium: Current theoretical trends and
changing perspectives. The Leadership Quarterly, 25(1), 36–62.
Emery, C. (2012). Uncovering the role of emotional abilities in leadership emergence. A longitudinal analysis of leadership networks. Social Networks, 34(4), 429–337.
Fernandez, R.R. (1991). Structural bases of leadership in intraorganizational networks.
Social Psychology Quarterly, 54(1), 36–53.
Freeman, L.C. (1979). Centrality in social networks: Conceptual clarification. Social
Networks, 1(3), 215–239.
Galunic, C., Ertuk, G., & Gargiulo, M. (2012). The positive externalities of social
capital: Benefiting from senior brokers. Academy of Management Journal, 55(5),
1213–1231.
Gibbons, D.E. (2004). Friendship and advice networks in the context of changing professional values. Administrative Science Quarterly, 49(2), 238–262.
Goodwin, V.L., Bowler, W.M., & Whittington, J.L. (2008). A social network perspective
on LMX relationships: Accounting for the instrumental value of leader and follower networks. Journal of Management, 35(4), 954–980.
SCHYNS_9781785367274_t.indd 290
10/11/2017 15:20
A social network approach to examining leadership ­291
Granovetter, M.S. (1973). The strength of weak ties. American Journal of Sociology, 78(6),
1360–1380.
Granovetter, M.S. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91(3), 481–510.
Huisman, M., & Van Duijn, M.A. (2005). Software for social network analysis. In
P.J. Carrington, J. Scott, & S. Wasserman (Eds.), Models and methods in social network
analysis (pp. 270–316). New York: Cambridge University Press.
Ingram, P., & Morris P. (2007). Do people mix at mixers? Structure, homophily, and the ‘life
of the party’. Administrative Science Quarterly, 52(4), 558–585.
Jokisaari, M. (2013). The role of leader–member and social network relations in newcomers’
role performance. Journal of Vocational Behavior, 82(2), 96–104.
Jokisaari, M., & Vuori J. (2010). The role of reference groups and network position in the
timing of employment service adoption. Journal of Public Administration Research and
Theory, 20(1), 137–156.
Jonczyk, C.D., Lee, Y.G., Galunic, C.D., & Bensaou, B.M. (2016). Relational changes
during role transitions: The interplay of efficiency and cohesion. Academy of Management
Journal, 59(3), 956–982.
Kaiser, R.B., Hogan, R., & Graig, S.B. (2008). Leadership and the fate of organizations.
American Psychologist, 63(2), 96–110.
Katz, R., & Tushman, M. (1983). A longitudinal study of the effects of boundary spanning supervision on turnover and promotion in research and development. Academy of
Management Journal, 26(3), 437–456.
Kilduff, M., & Brass, D.J. (2010). Organizational social network research: Core ideas and
key debates. Academy of Management Annals, 4(1), 317–357.
Kilduff, M., & Krackhardt, D. (1994). Bringing the individual back in: A structural analysis
of the internal market for reputation in organizations. Academy of Management Journal,
37(4), 87–108.
Kilduff, M., Tsai, W., & Hanke, R. (2006). A paradigm too far? A dynamic stability reconsideration of the social network research program. Academy of Management Review, 31(4),
1031–1048.
Knoke, D., & Yang, S. (2008). Social network analysis (2nd ed.). Thousand Oaks, CA:
Sage.
Kogovsek, T., & Ferligoj, A. (2005). Effects on reliability and validity of egocentered
network measurements. Social Networks, 27(3), 205–229.
Kossinets, G., & Watts, D.J. (2006). Empirical analysis of an evolving social network.
Science, 311(5757), 88–90.
Krackhardt, D. (1987). QAP partialling as a test of spuriousness. Social Networks, 9(2),
171–186.
Krackhardt, D. (1990). Assessing the political landscape: Structure, cognition, and power in
organizations. Administrative Science Quarterly, 35(2), 342–369.
Krackhardt, D., & Kilduff, M. (1999). Whether close or far: Social distance effects on perceived balance in friendship networks. Journal of Personality and Social Psychology, 76(5),
770–782.
Lam, S.N. & Schaubroeck, J. (2000). A field experiment testing frontline opinion leaders as
change agents. Journal of Applied Psychology, 85(6), 987–995.
Laumann, E.O., Marsden, P.V., & Prensky, D. (1983). The boundary specification problem
in network analysis. In R. Burt & M. Minor (Eds.), Applied network analysis: A methodological introduction (pp. 18–34). Beverly Hills, CA: Sage.
Lazega, E., & Pattison, P. (1999). Multiplexity, generalized exchange and cooperation in
organizations: A case study. Social Networks, 21(1), 67–89.
Lee, J.J. (2010). Heterogeneity, brokerage, and innovative performance: Endogenous formation of collaborative inventor networks. Organization Science, 21(4), 804–822.
Lin, N. (1982). Social resources and instrumental action. In P. Marsden & N. Lin
(Eds.), Social structure and network analysis (pp. 131–145). Beverly Hills, CA: Sage
Publications.
SCHYNS_9781785367274_t.indd 291
10/11/2017 15:20
292 Handbook of methods in leadership research
Lin, N. (1999). Social networks and status attainment. Annual Review of Sociology, 25(1),
467–487.
Lin, N. (2001). Social capital: A theory of social structure and action. New York: Cambridge
University Press.
Lin, N., Fu, Y.-C., & Hsung, R.-M. (2001). The position generator: Measurement techniques
for investigations of social capital. In N. Lin, K. Cook, & R.S. Burt (Eds.), Social capital:
Theory and research (pp. 57–81). Hawthorne, NY: Aldine de Gruyter.
Marsden, P. (2002). Egocentric and sociocentric measures of network centrality. Social
Networks, 24(4), 407–422.
Marsden, P. (2005). Recent developments in network measurement. In P.J. Carrington,
J. Scott, & S. Wasserman (Eds.), Models and methods in social network analysis (pp. 8–30).
New York: Cambridge University Press.
Marsden, P. (2011). Survey methods for network data. In J. Scott and P.J. Carrington (Eds.),
The Sage handbook of social network analysis (pp. 370–388). London: Sage Publications.
Marsden, P., & Campbell, K.E. (1984). Measuring tie strength. Social Forces, 63(2),
482–501.
McEvily, B., Soda, G., & Tortoriello, M. (2014). More formally: Rediscovering the missing
link between formal organization and informal social structure. Academy of Management
Annals, 8(1), 299–345.
Mehra, A., Dixon, A.L., Brass, D.J., & Robertson, B. (2006). The social network ties of group
leaders: Implications for group performance and leader reputation. Organization Science,
17(1), 64–79.
Mehra, A., Smith, B.R., Dixon, A.L., & Robertson, D. (2006). Distributed leadership in
teams: The network of leadership perceptions and team performance. The Leadership
Quarterly, 17(3), 232–245.
Monge, P., & Contractor, C. (2003). Theories of communication networks. Oxford: Oxford
University Press.
Oh, H., Chung, M.-H., & Labianca, G. (2004). Group social capital and group effectiveness:
The role of informal socializing ties. Academy of Management Journal, 47(6), 860–875.
Oh, H., Labianca, G., & Chung, M.-H. (2006). Multilevel model of group social capital.
Academy of Management Review, 31(3), 569–582.
Padget, J., & Ansell, C. (1993). Robust action and the rise of the Medici, 1400–1434.
American Journal of Sociology, 98(6), 1259–1319.
Parker, M., & Welch, E.W. (2013). Professional networks, science, ability, and gender
determinants of three types of leadership in academic and engineering. The Leadership
Quarterly, 24(2), 332–348.
Pastor, J.C., Meindl, J.R., & Mayo, M.C. (2002). A network effects model of charisma attributions. Academy of Management Journal, 45(2), 410–420.
Podolny, M., & Baron, J. (1997). Resources and relationships: Social networks and mobility
in the workplace. American Sociological Review, 62, 673–693.
Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction to exponential
random graph (p*) models for social networks. Social Networks, 29, 173–191.
Rodan, S., & Galunic, C. (2004). More than network structure: How knowledge heterogeneity influences managerial performance and innovativeness. Strategic Management Journal,
25(6), 541–562.
Schulte, M., Cohen, N.A., & Klein, K.J. (2012). The coevolution of network ties and perceptions of team psychological safety. Organization Science, 23(2), 564–581.
Scott, J. (1991). Social network analysis: A handbook. London: Sage.
Shadish, W.R., Cook, T.D., & Campbell, D.T. (2002). Experimental and quasi-experimental
designs for generalized causal inference. Belmont, CA: Wadsworth.
Singer, J.D., & Willett, J.B. (2003). Applied longitudinal data analysis: Modeling change and
event occurrence. New York: Oxford University Press.
Snijders, T.A.B. (2005). Models for longitudinal network data. In P. Carrington, J. Scott, &
S. Wasserman (Eds.), Models and methods in social network analysis (pp. 215–247). New
York: Cambridge University Press.
SCHYNS_9781785367274_t.indd 292
10/11/2017 15:20
A social network approach to examining leadership ­293
Snijders, T.A., Van de Bunt, G.G., & Steglich, C.E. (2010). Introduction to stochastic actorbased models for network dynamics. Social Networks, 32(1), 44–60.
Snijders, T.A.B., Steglich, C.E.G., & Schweinberger, M. (2007). Modeling the co-evolution
of networks and behavior. In K. van Montfort, H. Oud, & A. Satorra (Eds.), Longitudinal
models in the behavioral and related sciences (pp. 41–71). Mahwah, NJ: Erlbaum.
Sparrowe, R.T. (2014). Leadership and social networks: Initiating a different dialog. In
D. Day (Ed.), The Oxford handbook of leadership and organizations (pp. 434–454). New
York: Oxford University Press.
Sparrowe, R.T., & Liden, R.C. (1997). Process and structure in leader–member exchange.
Academy of Management Review, 22(2), 522–552.
Sparrowe, R.T., & Liden, R.C. (2005). Two routes to influence: Integrating leader–
member exchange and social network perspectives. Administrative Science Quarterly,
50(4), 505–535.
Sparrowe, R.T., Liden, R.C., Wayne, S.J., & Kraimer, M.L. (2001). Social networks and the
performance of individuals and groups. Academy of Management Journal, 4(2), 316–325.
Tasselli, S., Kilduff, M., & Menges, J.I. (2015). The microfoundations of organizational
social networks: A review and an agenda for future research. Journal of Management,
41(5), 1361–1387.
Tichy, N.M., Tushman, M.L., & Fombrun, C. (1979). Social network analysis for organizations. Academy of Management Review, 4(4), 507–519.
Venkataramani, V., Green, S.G., & Schleicher, D.J. (2010). Well-connected leaders: The
impact of leaders’ social network ties on LMX and members’ work attitudes. Journal of
Applied Psychology, 95(6), 1071–1084.
Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications.
Cambridge, UK: Cambridge University Press.
Wellman, B. (1988). Structural analysis: From method and metaphor to theory and substance. In B. Wellman & S.D. Berkowitz (Eds.), Social structures: A network approach (pp.
19–61). Cambridge, UK: Cambridge University Press.
West, S.G., & Thoemmes, F. (2010). Campbell’s and Rubin’s perspectives on causal inference. Psychological Methods, 15(1), 18–37.
Zaccaro, S.J., & Klimoski, R.J. (2001). The nature of organizational leadership: An introduction. In S.J. Zaccaro & R.J. Klimoski (Eds.), The nature of organizational leadership
(pp. 3–41). San Francisco, CA: Jossey-Bass.
SCHYNS_9781785367274_t.indd 293
10/11/2017 15:20
294 Handbook of methods in leadership research
APPENDIX: CONCEPTS AND TERMS RELEVANT TO
A SOCIAL NETWORK APPROACH1
Actors: Individuals, social groups, organizations or other units that are
part of a network.
Alter: An actor named as a member of a focal actor’s network.
Brokerage: A role in a network in which an actor connects others who are
not directly connected to each other (see structural hole).
Centrality: A network concept that indicates different central positions
in a network. Degree centrality indicates an actor’s number of network
ties (with in-degree referring to the number of ties to the actor from other
actors and out-degree referring to the number of ties from the actor to
other actors). Betweenness centrality measures the extent to which a focal
actor brokers or connects actors who are not connected to each other.
Eigenvector centrality indicates the extent to which a focal actor is related
to central actors in a network. Closeness centrality reveals how well an
actor can reach all other actors in a network.
Centralization: An indicator of the extent to which network ties are concentrated around a small number of actors.
Closure: A network in which actors are connected to each other. Network
density is a typical indicator of network closure.
Density: A network measure that indicates the extent to which alters in a
network are connected to each other. In particular, density is calculated
by dividing the number of network ties between alters by the maximum
number of possible ties between alters.
Egocentric network: An actor’s direct ties in a network and relations
between these ties – that is, the actor’s personal network (cf. whole
network).
Homophily: Actors’ tendency to have relations with those similar to themselves with respect to personal and social attributes, such as ethnicity,
organizational rank, and/or socioeconomic status (SES).
Multiplexity: A relation between two actors that includes different
types of connections. For example, A and B could be both friends and
co-workers.
Personal network: An actor’s direct ties in a network and relations between
these ties, also known as the actor’s egocentric network.
SCHYNS_9781785367274_t.indd 294
10/11/2017 15:20
A social network approach to examining leadership ­295
Reciprocity: A relation between two actors in which both actors indicate that
the relation exists. For example, A asks B for advice, and B asks A for advice.
Social capital: An actor’s network-based resources, such as information,
advice and recommendations that advance goal attainment. At the group
level, social capital is often indicated by network closure that enhances
norms and trust among group members.
Strength of tie (weak or strong ties): A concept used to characterize a relation between two actors based on attributes such as emotional closeness or
intimacy, meeting frequency, and reciprocity. Weak ties, such as acquaintance relations, are often characterized by low intimacy, low meeting frequency, and/or low reciprocity. Strong ties are characterized by emotional
closeness, high meeting frequency, and reciprocal services; for instance,
friendship relations tend to be strong ties.
Structural equivalence: The extent to which two actors have similar
network positions – that is, have ties to the same actors in a network.
Structural hole: A missing relation between two actors; a third actor can
play a broker role between these two actors.
Transitivity: A concept used to describe network triads. For example, if
A names B as a friend, B names C as a friend, and A names C as a friend,
then a transitive triad exists.
Upper reachability: An indicator of social resources in a network. Upper
reachability is often indicated by the highest occupational prestige or
socioeconomic status of network ties (e.g., Lin, 2001).
Whole network: A network derived from reports from all or most of the
actors in a given unit indicating these actors’ ties with each other (cf. personal network).
NOTE
1. See also Brass (1995) and Kilduff & Brass (2010).
SCHYNS_9781785367274_t.indd 295
10/11/2017 15:20
12. Diary studies in leadership
Sandra Ohly and Viktoria Gochmann
PREVIEW
Diary studies have increasingly been used in different areas of management research (Figure 12.1). In this chapter we will review recent research
using diary studies on leadership to illustrate the opportunities of this
methodology. We will describe the adequate study design (sample size, frequency of daily entries, event-contingent vs interval-contingent responses)
for diary studies on leadership. Furthermore, we will discuss additional
research topics (e.g., effects of daily leader behaviors and of leadership
style, dynamics of leadership) that might benefit from this research methodology. Throughout this chapter, we will provide two types of examples:
studies treating leadership as a stable characteristic or as a transient state.
Research on leadership has often relied on cross-sectional or longitudinal designs with two measurements. An underlying assumption in these
studies is that leadership has some stability over time. Leadership at one
point in time is used to predict behavior or attitudes at the same or a later
point in time (see Dulebohn, Bommer, Liden, Brouer, & Ferris, 2012; Ilies,
Nahrgang, & Morgeson, 2007 for details; Judge, Piccolo, & Ilies, 2004;
Mackey, Frieder, Brees, & Martinko, 2017). Rarely is the development or
change in leadership examined (see Nahrgang, Morgeson, & Ilies, 2009 for
an exception). The assumption of stability is problematic because there are
theoretical reasons to believe that the behavior, the perception of leaders,
and the relationship quality change over time (Klaussner, 2014) or across
situations (Dóci, Stouten, & Hofmans, 2015; see also Shamir, 2011).
Several researchers have argued that a better understanding of the causal
dynamics associated with leadership requires more precise theorizing
and measurement (Gooty, Connelly, Griffith, & Gupta, 2010; Hoffman
& Lord, 2013; Yukl, 2012). Hoffmann and Lord argued that this better
measurement can be developed in diary studies. Furthermore, a one-time
measurement of relationship quality might be systematically distorted
by liking, previous judgments or expectations guided by implicit leadership theories, or by an overemphasis on salient performance episodes
(Hoffman & Lord, 2013). For both these reasons, leadership research
might benefit from conducting diary studies. Finally, diary studies allow
the examination of research questions that cannot be examined otherwise,
296
SCHYNS_9781785367274_t.indd 296
10/11/2017 15:20
297
SCHYNS_9781785367274_t.indd 297
10/11/2017 15:20
0
1
2
3
4
5
9
8
19
0
9
19
Figure 12.1
Number of Studies
Number of studies reviewed in this chapter employing diary methodologies
Year of Publication
91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15
19 19 19 19 19 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20
298 Handbook of methods in leadership research
for example, the effects of daily leader behavior on followers’ attitudes and
behaviors. In this chapter, we will (1) review research on leadership that
has employed a diary design and (2) describe the general approach of diary
studies and the decisions faced by researchers when conducting them. (3)
We will discuss exemplary research questions in leadership research that
could be tested using diary studies.
DEFINITION AND CHARACTERISTICS OF A DIARY
STUDY
Diary methods refer to a class of methods using one or more daily assessments (Bolger, Davis, & Rafaeli, 2003). Related terms for this methodology include experience sampling studies (Csikszentmihalyi & LeFevre,
1989), event sampling (Reis & Gable, 2000), ecological momentary assessment (Beal & Weiss, 2003), and day reconstruction method (Kahneman,
Krueger, Schkade, Schwarz, & Stone, 2004). Whereas experience sampling focuses on thoughts, feelings, and behavior (flow experiences,
performance-related behavior) at various times across the day at fixed
or variable times, event sampling is contingent on the occurrence of predefined events, such as an interaction with a supervisor. The experiences
and behaviors are then reported after the event occurred. In both types of
studies, leader behaviors and employee reactions could be assessed. Day
reconstruction focuses on a complete assessment of daily activities and
associated thoughts and feelings retrospectively, typically at the end of
the day. In a typical day-reconstruction study, the activities are analysed
in terms of the experienced happiness. The term ecological momentary
assessment implies a higher validity of data compared to traditional
methods, and includes both event sampling and experience sampling
of experiences and behaviors. Data on momentary states might also be
assessed using physiological sensors (heart rate or skin conductance). This
type of study is called ambulatory assessment (Trull & Ebner-Priemer,
2013). Because of the similarities of these methodologies, and in line with
others (Bolger et al., 2003) we prefer to use the broader term diary study.
Diary studies can be either qualitative, quantitative or both. Qualitative
diaries contain open reports, such as diaries in a common sense and can
be used for theory building. For example, Seele (2016) used a qualitative approach to study the role of met expectations in the development
of newly formed leader–subordinate dyads. Diary studies can be purely
quantitative by employing short questionnaires. Miner, Glomb, and Hulin
(2005), for example, asked their 41 participants twice a day how strongly
they experienced an emotion on a scale ranging from 0 to 3 and if certain
SCHYNS_9781785367274_t.indd 298
10/11/2017 15:20
Diary studies in leadership ­299
events occurred. They analysed this data using hierarchical linear modeling. Both methods can also be combined and qualitative data can be
quantified by using categories rated by independent judges (e.g., Amabile,
Schatzel, Moneta, & Kramer, 2004).
The advantage of diary studies is the opportunity to examine research
questions in naturalistic work environments and using immediate assessment (Beal, Weiss, Barros, & MacDermid, 2005) instead of relying on
aggregated judgments. In questionnaire surveys evaluation can be biased
because raters rely on implicit theories, schemas, or individual differences because raters cannot correctly recall the focal leader behavior
(Hansbrough, Lord, & Schyns, 2015). For instance, a follower will rate
his or her leader as trustworthy, charismatic, or transformational when
liking this person (Brown & Keeping, 2005). Thus, the rating is distorted
by the general attitude towards the leader. When asked to report specific
instances of behavior, this bias is less likely to occur. Diary studies also
allow the study of behavioral and emotional patterns and their interdependences. For example, when an employee comes to work and receives
an inappropriate remark he or she might react emotionally and, as a
consequence, might even perform worse. What effect does this inappropriate remark have on the overall performance or the relationship with the
leader? How many such occurrences are needed before there is a change
in the general judgment about the leader? Or is it the overall judgment
that buffers the effects so that the follower of a highly charismatic leader
might think: “Well, my boss got up on the wrong side of the bed” without
a change in the overall evaluation? Knowledge about these patterns can
be useful to fill the black box of prior established effects. Throughout
this chapter, we will provide examples of how diary studies in leadership
research can contribute to a better understanding of leadership and leader
effectiveness.
Thus daily diaries can be used to study (1) the occurrence and temporal
pattern of transient states and behaviors; (2) relationships between transient states such as experiences and behaviors; and (3) relationships of
stable variables (person or situation characteristics) with transient states,
experiences, or behaviors (Ohly, Sonnentag, Niessen, & Zapf, 2010).
Our review of previous diary studies in leadership research is organized
accordingly.
PREVIOUS RESEARCH: LEADERSHIP DIARIES
With regard to leadership research, these research questions refer to (1)
the occurrence of leader behaviors and their temporal patterns (e.g., how
SCHYNS_9781785367274_t.indd 299
10/11/2017 15:20
300 Handbook of methods in leadership research
much time does a leader spend in interactions with followers?); (2) leader
behaviors as predictors or outcomes of transient states (e.g., do interactions with leaders lead to lower positive affect?); (3) relationships of leadership styles to employee behavior (e.g., do followers of transformational
leaders show more daily creativity?). Below, we will briefly review research
on each of these issues using diary studies. We will focus on methodological issues here.
Daily Leadership Behavior
Based on the assumption that leader behavior varies across time and situations, researchers have examined both the occurrence of leader activities,
and their predictors and outcomes.
Occurrence and temporal patterns of transient states and behaviors:
Leader behavior
In a study on daily activities in different domains, Camburn, Spillane, and
Sebastian (2010) examined 48 school principals for five days and three
periods over the course of three years. Their results revealed that principals spend more time on student affairs than on leadership activities such
as personnel, instructional leadership, building operations, or finances. To
validate the daily logs, both experience sampling and observations were
used and yielded similar results. The authors conclude that daily logs were
useful to study school principals’ activities in a more economical way than
structured observations, that intensive contact with participants helped
them achieve a high response rate, and that engagement in instructional
leadership (as an important task of school principals) was lower than is
desired based on professional standards. This study suggests that leaders
rarely interact with their followers.
Our own research stresses this issue. In a series of diary studies,
employees were asked if they had an interaction with their supervisor on
a particular day. If true, they were asked to shortly describe this event.
These descriptions were later used to develop and test taxonomy of leader
behaviors (Gochmann & Ohly, 2017). Across the four studies, 50.8 percent
of the employees did not report a daily interaction with their supervisor.
In a study comparing daily ratings of transformational leadership
over the course of one work week and general ratings of 143 employees,
Hoption (2016) found that average daily ratings are significantly lower
than general ratings of transformational leadership. Hoption argues that
these findings support the assumption that general ratings are not only
based on the frequency of actual daily leader behavior. Leadership behavior during specific instances such as crisis or reaching milestones might
SCHYNS_9781785367274_t.indd 300
10/11/2017 15:20
Diary studies in leadership ­301
have a disproportionately greater influence than everyday behavior. Thus,
although rare, some leader–follower interactions might have a profound
impact on general ratings of leadership, but also on follower motivation
and behavior. To our knowledge, no study to date has examined the temporal patterns of leader behaviors.
Relationships between transient states: Leader behavior as an outcome
Nielsen and Cleal (2011) examined situational antecedents of transformational leadership behaviors in a sample of 58 middle managers.
Participants were invited to rate their behavior on average eight times a
day, at random times with a minimum of 30 minutes in between, resulting
in 3072 momentary assessments in total. Twenty-six percent of the times
the participants were beeped to participate in the momentary assessment,
they were in contact with subordinates, and thus rated their own transformational leadership behavior using a measure adapted to the situational
self-report measurement. Results revealed that participants rated their
transformational leadership behavior as higher in situations when they
engaged in problem-solving activities and felt in control of the situation.
Transformational leadership behavior ratings were also related to perceptions of the work environment.
In a study of 53 managers, the level of daily transformational, considerate, and abusive leadership behaviors were related to stable leadership
identity (collective, relational, and individual; Johnson, Venus, Lanaj,
Mao, & Chang, 2012). The daily leadership ratings were aggregated to
an average frequency score for each manager in this study. Using a diary
methodology in this case had the advantage of resulting in more accurate
ratings of daily leadership behaviors (see above) than a conventional
cross-sectional study, and a reduction of common source bias because
leader identity and leader behavior were not assessed in a single survey at
the same point in time.
In a unique approach that matched leaders’ and subordinates’ ratings,
88 supervisors provided ratings of sleep quality and self-rated ego depletion over a period of two weeks; subordinates rated daily abusive
supervisor behavior and work engagement (Barnes, Lucianetti, Bhave,
& Christian, 2015). Daily subordinate ratings were included in further
analysis when they had at least a moderate amount of contact with their
supervisor on this specific day. The ratings were aggregated to the level
of the work unit based on analysis of their shared variance in work unit
members’ ratings. The results provide support for the theoretical model
linking sleep quality, ego depletion, abusive supervision, and work unit
work engagement, indicating that leader behavior is dependent on the
leader’s momentary well-being.
SCHYNS_9781785367274_t.indd 301
10/11/2017 15:20
302 Handbook of methods in leadership research
Taken together, these studies suggest that leader behaviors are dependent on internal states (depletion), identity, or contextual conditions.
Relationships between transient states: Leadership behavior as predictor of
employee outcomes
Research indicates that employees feel lower levels of positive affect
when interacting with their leader as compared to interactions with colleagues or customers (Bono, Foldes, Vinson, & Muros, 2007). In this
study, 57 employees participated and gave ratings of their daily affect and
their interaction partner roughly every two hours over the course of ten
working days. In total, 1983 ratings were collected.
Daily transformational (but not transactional) leader behavior was
linked to daily work engagement in a sample of 61 naval cadets who
rated their eight leaders’ behavior during 34 days on board (Breevaart
et al., 2014). Daily transformational leadership ratings, provided by 42
employees on five consecutive workdays, was linked to higher levels of
optimism, which in turn was linked to higher employee work engagement
(Tims, Bakker, & Xanthopoulou, 2011). Daily supervisor coaching rated
by 42 employees over the course of five consecutive work days of a fast
food restaurant (total ratings k 5 210) was related to self-rated daily work
engagement and financial returns, as indicated by the branch manager
(Xanthopoulou, Bakker, Demerouti, & Schaufeli, 2009). On days when
supervisors provided more coaching, employees experienced higher levels
of work engagement, and the fast food branch had higher levels of
­financial returns than on days when supervisor coaching was low.
As indicated above, daily abusive supervisor behavior is linked to
lower work group work engagement (Barnes et al., 2015). Aggregated
daily transformational leadership and aggregated daily abusive leadership
behaviors were related to leader effectiveness, as rated by subordinates
and colleagues (Johnson et al., 2012). Interestingly, also the consistency
of these behaviors across situations (operationalized as a low standard
deviation) was related to effectiveness, indicating the usefulness of multiple assessments of leader behavior. The authors conclude that “a person
who acts in a transformational manner today does not necessarily do so
tomorrow” (p. 1268).
Daily leader behaviors were assessed in diaries using open questions
in a study by Amabile and colleagues (2004) and categorized using the
managerial practice survey (Yukl, Wall, & Lepsinger, 1990). Daily behaviors including monitoring, consulting, socioemotional support, networking, and recognizing were positively related to daily perceived supervisor
support for creativity, as rated by the 139 participants. Negative instances
of not (competently) problem-solving, not monitoring, not motivating or
SCHYNS_9781785367274_t.indd 302
10/11/2017 15:20
Diary studies in leadership ­303
inspiring, or not recognizing were related to lower perceived supervisor
support. A total of 7194 daily observations were analysed. This study is
an example of combining qualitative and quantitative analyses using diary
data.
Social conflicts with supervisors (assessed after work) and negative
affect of employees at bedtime were linked (Volmer, 2015). In total, 98
employees provided 438 after-work surveys and 482 bedtime surveys.
Analyses revealed that when employees experienced a social conflict with
their supervisor during work time, they had higher levels of negative affect
at bedtime, also when controlling for negative affect in the beginning of
the day. This approach helps to rule out alternative explanations and
­provides some evidence of a causal effect.
Taken together, these studies suggest that leader behaviors have an
impact on their followers’ affect, motivation and behavior.
Relationship of Stable Variables with Transient States: Leadership Style
and Follower Reaction
In studies examining leadership style (or other more stable leader characteristics), the diary is combined with an introductory questionnaire on
these characteristics. For example, in a study on managers’ empathy (the
tendency to react compassionately to employees), the managers rated
their level of empathy before the employees filled out the diaries (Scott,
Colquitt, Paddock, & Judge, 2010). Sixty employees supervised by one
of 13 managers completed a daily survey for two weeks, resulting in 436
observations. Results revealed that employees with empathic managers
report lower levels of somatic complaints, but not better affective wellbeing. Notably, this study also showed that manager empathy moderated the within-individual relationship between goal progress and state
positive affect. Perceptions of daily goal progress were more strongly
associated with positive affect for groups of employees with empathic
managers. This study is an example of examining a cross-level interaction
effect.
Employees who rate their supervisor as being generally high on transformational leadership experience higher levels of daily positive affect
(Bono et al., 2007). Again, transformational leadership (as a more stable
characteristic) was rated in an introductory questionnaire, and positive
affect was rated several times daily, contingent on a signal emitted by an
electronic device. The advantage of using a diary methodology in this case
lies in the more valid assessment of positive affect (no recall or retrospective bias in the report of affect). These studies show that leadership styles
are related to employees’ affect.
SCHYNS_9781785367274_t.indd 303
10/11/2017 15:20
304 Handbook of methods in leadership research
HOW TO CONDUCT A DIARY STUDY IN
LEADERSHIP RESEARCH
When conducting a diary study, a number of decisions have to be made. In
this section, we address practical issues such as the sample size required in
diary studies, compliance, design and measures, devices, and approaches
to data analysis.
Sample Size
When planning a study, the question arises of how many participants are
required. A popular rule of thumb for multilevel studies (Scherbaum &
Ferreter, 2009) recommends at least 30 units on the higher level, and 30
units on the lower level. This might be hard to achieve in a diary study,
and only few examples of the research cited above would meet this recommendation. In this regard, it is important to note that there are at least two
sample sizes in experience sampling studies: number of participants N and
total number of incidents on the lower level (days, events) k. The required
sample size to achieve a certain power depends on the primary aim of the
study.
When researchers are interested in the relationship between daily behaviors and daily reactions, the total number of assessments k (equals N *
assessments sampled per person) is relevant to determining power. Studies
of this kind have employed between k 5 210 and k 5 7194 momentary/
daily assessments (Amabile et al., 2004; Xanthopoulou et al., 2009).
If researchers are interested in temporal patterns of behavior, an even
smaller number of participants might be sufficient, but more daily assessments are needed, for example 20 participants but up to 75 momentary/
daily assessments (see Fuller et al., 2003; Teuchmann, Totterdell, &
Parker, 1999 for examples outside the leadership area). When diary data
is aggregated to the person level, or when the focus is on stable predictors
(leadership style), sample size requirements are the same as in conventional survey studies, and N is the relevant sample size. Diary studies have
examined 53 (Johnson et al., 2012) or even 13 leaders (Scott et al., 2010),
which can be considered a very small sample. Maas and Hox (2005) concluded that standard errors can be biased when N < 50. Small samples
further entail the risk of being underpowered, and results might not be
generalizable to the population of leaders.
Researchers might be interested in interaction effects between individual differences and momentary assessments of behavior on momentary outcomes (cross-level interaction) (see Johnson et al., 2012). When
testing such cross-level interactions, power issues need specific attention
SCHYNS_9781785367274_t.indd 304
10/11/2017 15:20
Diary studies in leadership ­305
(Aguinis, Gottfredson, & Culpepper, 2013). Scherbaum and Ferreter
(2009) reviewed simulation studies on this issue and concluded that an
increasing number of participants often have a greater effect on power
than an increasing number of daily/momentary ratings. It is important to
note that there is oftentimes a dropout of participants between the introductory survey and the diary part, resulting in a smaller sample (Ohly et
al., 2010). This issue was also evident in some of the studies we reviewed
above.
Nevertheless, the generalizability of findings also depends on the
sample size, and the days sampled also need to be representative of individuals’ work life. Not all potential participants may be willing to take
part in a diary study, and not all relevant events or daily ratings will be
reported, resulting in selective samples on both levels (see Ohly & Fritz,
2010). In the Ohly and Fritz (2010) study, individuals under generally
high levels of time pressure were less likely to provide daily ratings.
Because time pressure was a focal variable in this study, we discussed how
the selective dropout might have affected study findings. It is advisable
to report the amount and reasons for missing data on both levels. In this
context it is important to keep in mind that subordinates might not have
contact with their supervisors on a daily basis. Thus, researchers might
strive to oversample the number of days to have a sufficient number for
the analysis.
Because modern statistical analyses such as multilevel analyses are flexible in handling missing data, it is not necessary for participants to provide
equal numbers of momentary assessments. It is common for researchers to
limit the analysis to participants who provide a certain minimum number
of momentary assessments to ensure the generalizability of findings. When
only few momentary entries are provided, there is a danger that these are
not representative of individuals’ everyday experience. For example, an
employee might be more motivated to provide daily ratings of his or her
supervisor on days when this person’s behavior was exceptionally friendly
or unfriendly than on days when the behavior was like every other day.
There is no definite recommendation for cut-offs because the decision
when to include or exclude participants in the analysis also depends on
the focal research question, and the researchers’ estimate of the likelihood
that missing data occurs not completely randomly (see Schafer & Graham,
2002 for more on the issue of missing data and ways to handle it). Stone
and Shiffman (2002) discuss reasons for missing data in diary studies, and
recommend that researchers report the missing data rates (and dropouts
and compliance rates) in detail so that others are able to assess the validity
of findings.
SCHYNS_9781785367274_t.indd 305
10/11/2017 15:20
306 Handbook of methods in leadership research
Recruitment and Compliance to Study Protocol
Recruiting the appropriate number of participants and ensuring compliance is the next challenge for researchers conducting diary studies. To
encourage participants’ collaboration and ensure compliance with the
protocol of when to fill in the diary, experts’ recommendations include
building an alliance with potential participants by explaining in detail
why the often strenuous and intrusive data collection process (Uy, Foo, &
Aguinis, 2010) is warranted, and what both researcher and participant can
gain (Stone & Shiffman, 2002). In our own experience, some participants
even enjoy taking part in diary studies because it helps them reflect on
their daily behavior. Researchers can support this reflection by providing
a detailed individualized feedback report at the end of the study, thereby
also building an alliance. In previous research, participants were also paid
contingent on their participation (Ilies, Scott, & Judge, 2006), but this
procedure entails the risk that participants will produce fake data to be
eligible for their remuneration (Green, Rafaeli, Bolger, Shrout, & Reis,
2006). Compliance to study protocol is important to ensure good data
quality, and instructing participants accordingly is recommended (Stone
& Shiffman, 2002).
Study Design and Measures
The frequency of assessing daily behaviors, events, or experiences also
needs to be determined in advance when planning a diary study. Higher
frequencies of assessments entail a number of advantages but these must
be weighed against the burden on participants. First, multiple assessments provide a more accurate picture of individuals’ experiences of
work. Furthermore, with multiple daily measurements, lagged effects of
leadership on outcomes can be tested, which facilitates causal inferences
(Judge, Simon, Hurst, & Kelley, 2014). Finally, retrospective bias can be
minimized when arranging multiple daily assessments. For example, the
description of a leadership event might be more accurate when p
­ articipants
are asked to describe each event as soon as it occurs.
This rating scheme is called event contingent, and can be distinguished
from interval-contingent and signal-contingent rating schemes. Intervalcontingent ratings are given at fixed times or in fixed intervals (e.g., at
noon or every two hours), and signal-contingent ratings at random times
during waking hours. Reis and Gable (2000) give recommendations for
the use of different recording protocols. They also recommend that filling
out the daily questionnaire should not exceed five to seven minutes, but
others recommend even shorter durations (Sonnentag & Geurts, 2009;
SCHYNS_9781785367274_t.indd 306
10/11/2017 15:20
Diary studies in leadership ­307
Table 12.1
I tems used in Zacher & Wilden (2014) to assess ambidextrous
leadership
Opening behaviors
Closing behaviors
Today, my supervisor allowed different
ways of accomplishing a task
Today, my supervisor encouraged
experimentation with different ideas
Today, my supervisor gave possibilities
for independent thinking and action
Today, my supervisor gave room for
own ideas
Today, my supervisor established
routines
Today, my supervisor took corrective
action
Today, my supervisor controlled
adherence to rules
Today, my supervisor paid attention
to uniform task accomplishments
Note: Items were rated on a five-point scale from 1 5 “not at all” to 5 – “frequently, if not
always”
Uy et al., 2010) not to compromise response rates, compliance, and data
quality. Therefore, abbreviated scales are used. For example, Zacher and
Wilden (2014) used the four highest-loading items of two scales of leadership behavior. Specifically, the items that were used for two facets of
ambidextrous leadership, opening and closing behaviors)1 are shown in
Table 12.1.
These items show a strong focus on daily behavior. Adapting established
leadership-style instruments to the daily level requires that the respective
behavior can be shown on a daily basis – for example, articulating a vision
as a sample behavior of transformational leadership. Hoption (2016) used
the following item: “My leader emphasized a collective sense of a mission
today.” Care should be taken when adapting scales to a different time
frame to ensure construct validity. For example, a daily assessment of
leader–member exchange quality might not fully capture the quality of the
relationship in the sense of the theoretical conception (Bono, 2013).
Devices
The advantage of electronic devices such as smartphones, portable or
stationary computers is that compliance with the study protocol (e.g., predetermined times to answer the questions) can be tracked via an electronic
time stamp. Entries that do not comply with the study protocol can be
identified and eliminated.
Furthermore, electronic devices can be programmed to automatically
remind participants of the questionnaire (Ohly et al., 2010; Sonnentag &
Geurts, 2009; Uy et al., 2010). For a detailed discussion of compliance
SCHYNS_9781785367274_t.indd 307
10/11/2017 15:20
308 Handbook of methods in leadership research
problems with paper-and-pencil diaries see Green and colleagues (2006).
With increasing smartphones and software available for different devices,
we predict an increasing number of diary studies using these devices.
Software is available from multiple sources.2 The available options differ
in terms of support for different devices, the available rating formats, and
their design. From our experience, apps will not perform equally well for
users of android and iOS devices, and some apps are developed for only
one type of operating system.
Data Analysis
Data gathered with diary methods require specific data analysis techniques. As noted above, common research hypotheses are concerned
with relationships between transient variables (e.g., leader behavior and
employee affect), and relationships of stable characteristics (leadership
style with transient variables (employee affect).
Because day-level data are nested within-person, multilevel approaches
are needed to test interrelations because they take the interdependence
of data into account (Raudenbush & Bryk, 2002; Snijders & Bosker,
1999; see also Chapter 10 by Yammarino and Gooty in this handbook).
For example, reports of daily affect might be more similar when the
same person reports them repeatedly, as compared to reports by different persons. The daily observations constitute level 1 data, and the
stable person characteristics constitute level 2 data. In leadership diaries,
this issue is complicated further: daily entries (level 1) might be nested
in persons (level 2), who are nested in teams with one leader (level 3),
resulting in a three-level data structure (Scott et al., 2010). The amount
of dependence present in the data at each level can be determined by calculating ICC (intraclass correlation coefficient) values (Bliese & Ployhart,
2002). When ICC values are low, there may not be a need to conduct multilevel analysis, or a specific level (e.g., the team level) could be neglected.
Relationship between two level 1 variables or between a level 2 variable and a level 1 variable can be analysed. Furthermore, the relationship
between two level 1 variables (e.g., leader behavior and employee affect)
might be dependent on a level 2 variable (e.g., leader’s empathy). This
is called cross-level interaction. When researchers are interested in the
trend of transient variables over time, additional statistical techniques are
required, for example time series analysis or latent growth analysis (see
Chapter 13 by Hall in this handbook). Research questions on dynamics
of leadership, such as on the effect of different subsequent supervisor–
employee interactions, require even more sophisticated analyses such as
sequence analysis (Biemann & Datta, 2014).
SCHYNS_9781785367274_t.indd 308
10/11/2017 15:20
Diary studies in leadership ­309
Diary data could also be aggregated to a higher level, for example daily
ratings of leader behaviors could be used to determine the typical leader
behavior across time (cf. Johnson et al., 2010). The disadvantage of this
aggregation is that fluctuations across days are neglected in the analyses.
The appropriate analysis technique must be chosen depending on the
research question. A complete treatment of multilevel modeling and the
alternative approaches is beyond the scope of this chapter. For an accessible treatment of multilevel analysis with a focus on analysing cross-level
interactions see Aguinis and colleagues (2013).
Diary data might also be analysed using qualitative methods. In our
own research (Gochmann & Ohly, 2017) the daily reports of leader–
employee interactions were sorted into categories of possible interaction
types, for example goal-oriented behavior, positive feedback, or participation. The reliability of this categorization was examined by calculating
interrater agreement. Information from the daily reports was further used
to refine the categories. For more information on qualitative analysis, see
Chapter 14 by Schilling in this handbook.
FUTURE RESEARCH
In this part of the chapter, we will develop ideas on how to employ diary
studies in future research on leadership.
Leadership Style Developing Over Time
A leadership style develops over time and across multiple events. Klaussner
(2014) describes the emergence of abusive supervision as an escalating
process of subordinate–supervisor interactions. This process starts when
a subordinate perceives he or she is being treated unfairly, reacts to this
perception in a dysfunctional way, which in turn stimulates supervisor
behavior towards the subordinate. Klaussner argues that the “perceptions of supervisor injustice are proposed to accumulate over reoccurring
trigger events when they remain unresolved” (p. 320). A diary study that
examines the respective behaviors and perceptions multiple times across
a longer time range would be ideally suited to studying this process. The
accumulation of injustice perceptions could be studied by combining an
event- and an interval-based approach in that participants are instructed
to report important events as soon as they occur, but also give ratings of
injustice perceptions at regular intervals.
Development of LMX can be thought of as mutual reinforcements
(Dienesch & Liden, 1986) in which a leader delegates important interesting
SCHYNS_9781785367274_t.indd 309
10/11/2017 15:20
310 Handbook of methods in leadership research
tasks based on a first impression of the subordinate, and this delegation
is reinforced when the subordinate shows loyalty and good performance.
Previous research has examined the role of supervisor or subordinate
characteristics in the development of LMX (e.g., Nahrgang, et al., 2009)
but has largely neglected the role of transient states and events (see
Cropanzano et al., 2017, for theoretical arguments).
To study the development of a leadership style over time, the leadership
style would be assessed in a pre- and a post-test questionnaire, and the
relevant events or experiences would be reported in the diary (from the
leader’s or the employee’s perspective, or both). Again, a combination of
event- and interval-contingent methods seems ideal to capture the dynamic
of mutual reinforcements or the escalating process. Because the focus of
this research is on the development of a stable leadership characteristic in
both cases (LMX or abusive supervision), a large number of participants
is required for robust results, but less daily assessments might be adequate
based on the observation that leader–subordinate interactions do not take
place on a daily basis.
Based on the observation that participants sometimes like to take part
in diary studies, one could use the diary as part of an intervention. Diaries
that provide an opportunity to reflect on past behavior and to prepare for
the future, for example, by using open-ended questions, can serve as an
intervention in itself (see also Burt, 1994). Diaries as an accurate picture
of leader behavior could also be used to enhance leadership interventions.
For example, diary reports on critical leadership incidents could serve as
the basis of leadership coaching. Finally, the diaries could also serve to
evaluate the effectiveness of leadership training (Hammer, Kossek, Anger,
Bodner, & Zimmerman, 2011) because the daily reports of what individuals do might be a more accurate report than general leadership ratings
(Hoffman & Lord, 2013).
Leadership Behavior and Fluctuating States and Behaviors
Temporal patterns of leadership behavior
Affect has been shown to follow a distinctive temporal pattern. Some
leadership behavior might be affect driven in that leaders behave more
or less favorably towards their subordinates based on their own affective
state. Previous research linking abusive supervision and ego depletion
supports this view (Barnes et al., 2015). Furthermore, external temporal
cues such as end of month or end of year can trigger specific behavior.
For example, a leader in an accounting firm would show more “closing”
leadership behavior shortly before the end of the month deadline. Based
on these arguments, one would argue that leadership behavior follows a
SCHYNS_9781785367274_t.indd 310
10/11/2017 15:20
Diary studies in leadership ­311
specific temporal pattern (for more on temporal lenses on behavior see
Shipp & Cole, 2015). Identification of this pattern would help to develop
novel interventions to enhance leadership effectiveness. Furthermore, the
time of day or day of the week might be a third variable that sometimes
underlies the relationship between momentary states and behaviors (Liu
& West, 2015). Researchers are thus advised to carefully evaluate if there
might be temporal patterns in any of the momentary states or behaviors they are interested in and to control for time if necessary (see as an
example Daniels, Boocock, Glover, Hartley, & Holland, 2009).
Leadership style and fluctuating states and behaviors
Recent research has shown that employees feel more motivated when
their leaders show inspirational leadership (Antonakis, D’Adda, Weber,
& Zender, 2014). By examining the effects of specific leader behaviors in
a diary study, the sustainability of this effect could be determined. For
example, one could examine how specific rhetoric influence attempts are
linked to daily work motivation over the course of the following work
week.
Gooty and colleagues (2010), after reviewing the literature on affect
and leadership, give recommendations for future research in this area. For
example, they identify the impact of follower behaviors on leaders’ style
and leader behavior as a gap in the literature. One could study this effect
using a diary approach and having leaders rate their subordinates and
vice versa on a daily basis (see Barnes and colleagues’ (2015) approach
on matching daily leader and subordinate ratings). Leaders could also
be asked to provide ratings of their leadership style in a pre- and postdiary questionnaire to test for longer-term changes in leadership style.
When the focus is on how subordinates affect leaders’ daily behavior, a
large number of matching daily entries is required. When the focus is on
the effect of followers’ behavior on change in leadership style, a larger
number of ­supervisor–subordinate dyads is needed (see “Leadership Style
Developing Over Time” above). Furthermore, Gooty and colleagues call
for more research integrating cognitive appraisal theories with research on
leadership and affect. It would be interesting to examine how the appraisal
of certain events such as a conflict (both from the leader’s and from the
follower’s view) affects their reaction to the event, and the subsequent
behavior. The appraisal is ideally assessed directly after the event happens
to reduce retrospective bias. An event-contingent scheme could be used so
that participants are instructed to rate their appraisal as soon as the event
has occurred. Another research area in which diaries seems ideal is the
“function of emotions in instigating or suppressing moral and ethical leadership behavior” (p. 1000). Discrete emotions might be best captured using
SCHYNS_9781785367274_t.indd 311
10/11/2017 15:20
312 Handbook of methods in leadership research
an interval-based scheme where participants are contacted at irregular
intervals to report on their momentary state. A general recommendation
is to better match the levels of analysis in theory, methods, and inferences.
The research questions discussed above are concerned with relationships of momentary states and momentary behaviors. Johnson and colleagues (2012) argued that individual differences such as trait self-control,
agreeableness and emotional stability might buffer the relationship
between low sleep quality and abusive supervision. This idea calls for
more research examining cross-level interactions of individual differences
and daily experiences and behaviors. As described above, these studies
need to sample a large number of leaders. The sampling protocol for
the daily assessment depends on the focal variable. When examining the
relationship between sleep quality and daily leader behaviors, an intervalcontingent rating scheme and one daily rating is appropriate. When
focusing on the relationship between specific events such as conflicts with
subordinates and behaviors an event-based approach might be useful.
This might result in a longer study period to sample a sufficient number
of events. Signal-contingent response and multiple daily assessments
would be advisable when examining daily leader affect and daily leader
behavior.
LIMITATIONS OF DIARY STUDIES
Although diaries provide some advantages over common one-time survey
research, including more precise measurements, they are not without
problems. One issue with diary research is that causal conclusions are
often not warranted. Although the design and the analyses eliminate concerns that stable individual differences underlie the relationships in a diary
study, correlations might be due to a third state variable such as context
or affect when answering the questionnaire (Bono, 2013). It might be that,
due to repeated measurements, participants guess the aim of the study and
provide their responses accordingly, creating spurious results. Therefore,
extra measures must be undertaken to rule out alternative explanations of
significant results. In the studies reviewed above, these measures include
the temporal separation of predictor and outcome, the use of independent
sources of data, or the use of state affect as a control variable. Researchers
might also want to test for reverse causality by examining the relationship
of their outcome at one point in time and the predictor at a later point in
time (cf. Ilies, Scott, & Judge, 2006). It might also be that diary studies
increase reflection on behavior, thus changing behavior over the time of
the study. Changes in behavior might be checked by testing for time trends
SCHYNS_9781785367274_t.indd 312
10/11/2017 15:20
Diary studies in leadership ­313
(cf. Daniels et al., 2009). These methods are also recommended for future
research using diary studies in leadership research.
CONCLUSIONS
This chapter reviews leadership research employing diary methodologies
for a variety of research questions. By reviewing this research, discussing
important decisions researchers face when conducting diary studies, and
by providing examples of future research questions, we hope to inspire
other researchers to employ this methodology to capture leadership “live
as it is lived” (Bolger et al., 2003, p. 579).
NOTES
1. Opening behaviors encourage followers to think independently, experiment with new
ways of doing things, embrace error learning, take risks, and challenge the status quo.
Closing behaviors encourage followers to create routines, monitor goals, avoid errors,
take corrective action, and adhere to rules and standards (Rosing et al., 2011).
2. For example, http://www.expimetrics.com/; http://ilumivu.com/; see also http://www.
saa2009.org/?page_id557; all accessed July 21, 2017.
REFERENCES
Aguinis, H., Gottfredson, R.K., & Culpepper, S.A. (2013). Best-practice recommendations for estimating cross-level interaction effects using multilevel modeling. Journal of
Management. doi: 10.1177/0149206313478188
Amabile, T.M., Schatzel, E.A., Moneta, G.B., & Kramer, S.J. (2004). Leader behaviors and
the work environment for creativity: Perceived leader support. The Leadership Quarterly,
15(1), 5–32.
Antonakis, J., D’Adda, G., Weber, R., & Zender, C. (2014). Just words? Just speeches?
On the economic value of charismatic leadership. (Working Paper). Department of
Organizational Behavior, University of Lausanne.
Barnes, C.M., Lucianetti, L., Bhave, D.P., & Christian, M.S. (2015). “You wouldn’t like me
when I’m sleepy”: Leaders’ sleep, daily abusive supervision, and work unit engagement.
Academy of Management Journal, 58(5), 1419–1437.
Beal, D.J., & Weiss, H.M. (2003). Methods of ecological momentary assessment in organizational research. Organizational Research Methods, 6(4), 440–464.
Beal, D.J., Weiss, H.M., Barros, E., & MacDermid, S.M. (2005). An episodic process model
of affective influences on performance. Journal of Applied Psychology, 90(6), 1054–1068.
Biemann, T., & Datta, D.K. (2014). Analyzing sequence data: Optimal matching in management research. Organizational Research Methods, 17(1), 51–76.
Bliese, P.D., & Ployhart, R.E. (2002). Growth modeling using random coefficient models:
Model building, testing, and illustrations. Organizational Research Methods, 5(4), 362–387.
Bolger, N., Davis, A., & Rafaeli, E. (2003). Diary methods: Capturing life as it is lived.
Annual Review of Psychology, 54(1), 579–616.
SCHYNS_9781785367274_t.indd 313
10/11/2017 15:20
314 Handbook of methods in leadership research
Bono, J.E. (2013). New developments in within-person research. Paper presented at the
Academy of Management Conference, Boston, MA.
Bono, J.E., Foldes, H.J., Vinson, G., & Muros, J.P. (2007). Workplace emotions: The role of
supervision and leadership. Journal of Applied Psychology, 92(5), 1357–1367.
Breevaart, K., Bakker, A., Hetland, J., Demerouti, E., Olsen, O.K., & Espevik, R. (2014).
Daily transactional and transformational leadership and daily employee engagement.
Journal of Occupational and Organizational Psychology, 87(1), 138–157.
Brown, D.J., & Keeping, L.M. (2005). Elaborating the construct of transformational leadership: The role of affect. The Leadership Quarterly, 16(2), 245–272.
Burt, C.D. (1994). Prospective and retrospective account-making in diary entries: A model of
anxiety reduction and avoidance. Anxiety, Stress and Coping, 6(4), 327–340.
Camburn, E.M., Spillane, J.P., & Sebastian, J. (2010). Assessing the utility of a daily log
for measuring principal leadership practice. Educational Administration Quarterly, 46(5),
707–737.
Cropanzano, R., Dasborough, M.T., & Weiss, H.M. (2017). Affective events and the development of leader–member exchange. Academy of Management Review, 42, 233–258.
Csikszentmihalyi, M., & LeFevre, J. (1989). Optimal experience in work and leisure. Journal
of Personality and Social Psychology, 56(5), 815–822.
Daniels, K., Boocock, G., Glover, J., Hartley, R., & Holland, J. (2009). An experience sampling study of learning, affect, and the demands control support model. Journal of Applied
Psychology, 94(4), 1003–1017.
Dienesch, R.M., & Liden, R.C. (1986). Leader–member exchange model of leadership: A
critique and further development. Academy of Management Review, 11(3), 618–634.
Dóci, E., Stouten, J., & Hofmans, J. (2015). The cognitive-behavioral system of leadership:
Cognitive antecedents of active and passive leadership behaviors. Frontiers in Psychology,
6. Retrieved from https://doi.org/10.3389/fpsyg.2015.01344.
Dulebohn, J.H., Bommer, W.H., Liden, R.C., Brouer, R.L., & Ferris, G.R. (2012). A metaanalysis of antecedents and consequences of leader–member exchange: Integrating the past
with an eye toward the future. Journal of Management, 38(6), 1715–1759.
Fuller, J.A., Stanton, J.M., Fisher, G.G., Spitzmueller, C., Russell, S.S., & Smith, P.C.
(2003). A lengthy look at the daily grind: Time series analysis of events, moods, stress, and
satisfaction. Journal of Applied Psychology, 88(6), 1019–1033.
Gochmann, V., & Ohly, S. (2017). What’s going on? An event perspective of daily leader–follower interactions. Paper presented at the Academy of Management Meeting, Atlanta,
GA.
Gooty, J., Connelly, S., Griffith, J., & Gupta, A. (2010). Leadership, affect, and emotions: A
state of the science review. The Leadership Quarterly, 21(6), 979–1004.
Green, A.S., Rafaeli, E., Bolger, N., Shrout, P.E., & Reis, H.T. (2006). Paper or plastic? Data
equivalence in paper and electronic diaries. Psychological Methods 11(1), 87–105.
Hammer, L.B., Kossek, E.E., Anger, W.K., Bodner, T., & Zimmerman, K.L. (2011).
Clarifying work–family intervention processes: The roles of work–family conflict and
family-supportive supervisor behaviors. Journal of Applied Psychology, 96(1), 134–150.
Hansbrough, T.K., Lord, R.G., & Schyns, B. (2015). Reconsidering the accuracy of follower
leadership ratings. The Leadership Quarterly, 26(2), 220–237.
Hoffman, E.L., & Lord, R.G. (2013). A taxonomy of event-level dimensions: Implications
for understanding leadership processes, behavior, and performance. The Leadership
Quarterly, 24(4), 558–571.
Hoption, C. (2016). It does not add up: Comparing episodic and general leadership ratings.
Leadership, 12(4), 491–503.
Ilies, R., Nahrgang, J.D., & Morgeson, F.P. (2007). Leader–member exchange and citizenship behaviors: A meta-analysis. Journal of Applied Psychology, 92(1), 269–277.
Ilies, R., Scott, B.A., & Judge, T.A. (2006). The interactive effects of personal traits and experienced states on intraindividual patterns of organizational citizenship behavior. Academy
of Management Journal, 49(3), 561–575.
Johnson, R.E., Venus, M., Lanaj, K., Mao, C., & Chang, C.-H. (2012). Leader identity as
SCHYNS_9781785367274_t.indd 314
10/11/2017 15:20
Diary studies in leadership ­315
an antecedent of the frequency and consistency of transformational, consideration, and
abusive leadership behaviors. Journal of Applied Psychology, 97(6), 1262–1272.
Judge, T.A., Piccolo, R.F., & Ilies, R. (2004). The forgotten ones? The validity of consideration and initiating structure in leadership research. Journal of Applied Psychology, 89(1),
36–51.
Judge, T.A., Simon, L.S., Hurst, C., & Kelley, K. (2014). What I experienced yesterday is
who I am today: Relationship of work motivations and behaviors to within-individual
variation in the five-factor model of personality. Journal of Applied Psychology, 99(2),
199–221.
Kahneman, D., Krueger, A.B., Schkade, D.A., Schwarz, N., & Stone, A.A. (2004). A survey
method for characterizing daily life experience: The day reconstruction method. Science,
306(5702), 1776–1780.
Klaussner, S. (2014). Engulfed in the abyss: The emergence of abusive supervision as an escalating process of supervisor–subordinate interaction. Human Relations, 67(3), 311–332.
Liu, Y., & West, S.G. (2015). Weekly cycles in daily report data: An overlooked issue.
Journal of Personality, 84(5), 560–579.
Maas, C.J.M., & Hox, J.J. (2005). Sufficient sample sizes for multilevel modeling.
Methodology, 1, 86–92.
Mackey, J.D., Frieder, R.E., Brees, J.R., & Martinko, M.J. (2017). Abusive supervision: A
meta-analysis and empirical review. Journal of Management, 43(6), 1940–1965.
Miner, A.G., Glomb, T.M., & Hulin, C. (2005). Experience sampling mood and its correlates
at work. Journal of Occupational & Organizational Psychology, 78(2), 171–193.
Nahrgang, J.D., Morgeson, F.P., & Ilies, R. (2009). The development of leader–member
exchanges: Exploring how personality and performance influence leader and member
relationships over time. Organizational Behavior and Human Decision Processes, 108(2),
256–266.
Nielsen, K., & Cleal, B. (2011). Under which conditions do middle managers exhibit
transformational leadership behaviors? An experience sampling method study on the
predictors of transformational leadership behaviors. The Leadership Quarterly, 22(2),
344–352.
Ohly, S., & Fritz, C. (2010). Work characteristics, challenge appraisal, creativity and proactive behavior: A multilevel study. Journal of Organizational Behavior, 31(4), 543–565.
Ohly, S., Sonnentag, S., Niessen, C., & Zapf, D. (2010). Diary studies in organizational
research: An introduction and some practical recommendations. Journal of Personnel
Psychology, 9(2), 79–93.
Raudenbush, S.W., & Bryk, A.S. (2002). Hierarchical linear models: Applications and data
analysis methods. Thousand Oaks, CA: Sage.
Reis, H.T., & Gable, S.L. (2000). Event sampling and other methods for studying everyday
experience. In H.T. Reis & C.M. Judd (Eds.), Handbook of research methods in social and
personality psychology (pp. 190–222). New York: Cambridge University Press.
Rosing, K., Frese, M., & Bausch, A. (2011). Explaining the heterogeneity of the leadership–innovation relationship: Ambidextrous leadership. The Leadership Quarterly, 22(5),
956–974.
Schafer, J.L., & Graham, J.W. (2002). Missing data: Our view of the state of the art.
Psychological Methods, 7(2), 147–177.
Scherbaum, C.A., & Ferreter, J.M. (2009). Estimating statistical power and required sample
sizes for organizational research using multilevel modeling. Organizational Research
Methods, 12(2), 347–367.
Scott, B.A., Colquitt, J.A., Paddock, E.L., & Judge, T.A. (2010). A daily investigation of the
role of manager empathy on employee well-being. Organizational Behavior and Human
Decision Processes, 113(2), 127–140.
Seele, H. (2016), Die Wirkung von enttäuschten Mitarbeitererwartungen an Personalführung
– Attributionstheoretische Effekte und Handlungskonsequenzen [The effects of disappointed employee expectations on personnel management – attribution theory and action
sequences] (Dissertation). Universität Kassel, Springer Gabler, forthcoming.
SCHYNS_9781785367274_t.indd 315
10/11/2017 15:20
316 Handbook of methods in leadership research
Shamir, B. (2011). Leadership takes time: Some implications of (not) taking time seriously in
leadership research. The Leadership Quarterly, 22(2), 307–315.
Shipp, A.J., & Cole, M.S. (2015). Time in individual-level organizational studies: What is
it, how is it used, and why isn’t it exploited more often? Annual Review of Organizational
Psychology and Organizational Behavior, 2(1), 237–260.
Snijders, T.A.B., & Bosker, R.J. (1999). Multilevel analysis: An introduction to basic and
advanced multilevel modeling. Thousand Oaks, CA: Sage.
Sonnentag, S., & Geurts, S. (2009). Methodological issues in recovery research. In
S. Sonnentag, P.L. Perrewé & D.C. Ganster (Eds.), Current perspectives in job-stress
recovery. Research in occupational stress and well-being (pp. 1–36). Bingley, UK: Emerald.
Stone, A.A., & Shiffman, S. (2002). Capturing momentary, self-report data: A proposal for
reporting guidelines. Annals of Behavioral Medicine, 24(3), 236–243.
Teuchmann, K., Totterdell, P., & Parker, S.K. (1999). Rushed, unhappy, and drained: An
experience sampling study of relations between time pressure, perceived control, mood,
and emotional exhaustion in a group of accountants. Journal of Occupational Health
Psychology, 4(4), 37–54.
Tims, M., Bakker, A.B., & Xanthopoulou, D. (2011). Do transformational leaders enhance
their followers’ daily work engagement? The Leadership Quarterly, 22(1), 121–131.
Trull, T.J., & Ebner-Priemer, U. (2013). Ambulatory assessment. Annual Review of Clinical
Psychology, 9(2), 151–176.
Uy, M.A., Foo, M.-D., & Aguinis, H. (2010). Using experience sampling methodology to
advance entrepreneurship theory and research. Organizational Research Methods, 13(1),
31–54.
Volmer, J. (2015). Followers’ daily reactions to social conflicts with supervisors: The moderating role of core self-evaluations and procedural justice perceptions. The Leadership
Quarterly, 26(5), 719–731.
Xanthopoulou, D., Bakker, A.B., Demerouti, E., & Schaufeli, W.B. (2009). Work engagement and financial returns: A diary study on the role of job and personal resources. Journal
of Occupational & Organizational Psychology, 82(1), 183–200.
Yukl, G. (2012). Effective leadership behavior: What we know and what questions need more
attention. Academy of Management Perspectives, 26(4), 66–85.
Yukl, G.A., Wall, S., & Lepsinger, R. (1990). Preliminary report on validation of the managerial practices survey. In K.E. Clark & M.B. Clark (Eds.), Measures of leadership (pp.
223–237). Greensboro, NC: Center for Creative Leadership.
Zacher, H., & Wilden, R.G. (2014). A daily diary study on ambidextrous leadership and selfreported employee innovation. Journal of Occupational and Organizational Psychology,
87(4), 813–820.
SCHYNS_9781785367274_t.indd 316
10/11/2017 15:20
13. Modeling leadership-related change with
a growth curve approach
Rosalie J. Hall
A variety of leadership and followership theories are dynamic in the sense
that they imply change in the focal variables over time. In particular, one
thinks of leadership theories that involve the learning and development
of leadership skills and behaviors (e.g., Day, Fleenor, Atwater, Sturm,
& McKee, 2014), but they also include theories about processes such as
the assimilation/adaptation that occurs as a result of workplace socialization or the development of leader–member exchange relationships and
trust. Indeed, recent considerations of leadership and followership from a
dynamic perspective are yielding new insights.
For example, Day and Sin’s (2011) theoretically based empirical study
focuses on changes in ratings of perceived leadership effectiveness at four
points in time as study participants took part in term-long action learning projects focused on team-building and the development of leadership.
Among other results, Day and Sin demonstrated between-participant
variability in ratings of perceived leadership effectiveness at the study
start, indicating that first-year university students differ significantly in
this leadership quality. In addition, they found differences in the shape
of the individual change trajectories for leadership effectiveness over the
course of the project. Specifically, the majority of the sample showed a
drop in leadership effectiveness ratings from the initial measurement time,
which then plateaued or showed a very slight upturn, while a smaller
group showed a linear, increasing trend in ratings of leadership effectiveness across the four measurement times. This suggests differential benefits
from the leadership development initiative, with one group benefiting and
(at least in the short run) the other group either not benefiting or possibly
even showing negative effects.
In addition, Day and Sin (2011) found several important contingent
relationships. One of them was that study participants who more strongly
identified as a leader at a specific point in time also tended to have higher
leadership effectiveness ratings at that time. They also found that different
forms of goal orientation differentially related to initial ratings of leadership effectiveness and/or the pattern of change in effectiveness over time.
Another example of a dynamic study related to leadership and
317
SCHYNS_9781785367274_t.indd 317
10/11/2017 15:20
318 Handbook of methods in leadership research
f­ollowership is Jokisaari and Nurmi’s (2009) study of changes in role
clarity, work mastery, and job satisfaction as a function of perceived
supervisor support, using measurements collected from organizational
newcomers on four occasions over about a year-and-a-half. In general,
levels of perceived supervisor support for these newcomers tended to
decline over time, in keeping with previous works on the “honeymoon”
period of high positive evaluations that often characterizes the start of
interpersonal relationships, and is typically followed by a return to more
realistic levels (Fichman & Levinthal, 1991). Another interesting finding
from this study was that newcomers who experienced a steeper decline
over time in their perceptions of supervisor support also had steeper
declines in their levels of role clarity and job satisfaction, as well as lower
salary increases. Longitudinal studies such as the two just described are
particularly helpful in understanding the direction and pattern of change
(if any) of leadership- and followership-related variables and the rates at
which processes such as leader development and follower socialization
occur.
To test theories of change typically requires the collection of longitudinal data consisting of three or more repeated measurements on the
focal units of analysis over a time span (e.g., hours, days, weeks, months,
years, etc.) that is appropriate for the research question being addressed.
The data collection can be done either in a field or a laboratory context
(e.g., see Rietzschel, Wisse, and Rus on laboratory studies, Chapter
3 in the current volume). And, the repeated observations may be of
persons, dyads, groups, or any other type of entity. Once the data have
been collected, then a suitable analytic method must be applied. There
are a variety of available choices for the analysis of longitudinal data.
The current chapter deals with a general approach that may be useful
in many ­circumstances – growth curve modeling (GCM). The models
are also sometimes referred to as latent curve models (e.g., McArdle &
Nesselroade, 2014) or latent growth curve models. In general, such growth
models are used to “estimate between-person differences in within-person
change” (Curran, Obeidat, & Losardo, 2010, p. 122). This highly flexible
technique can be implemented using either a multilevel modeling (MLM)/
random coefficients regression framework or a structural equation modeling (SEM) framework.
This chapter provides a conceptual description of the underlying logic
of growth curve modeling, an overview of multilevel and SEM approaches
to specifying growth models, some tips for data collection and analysis
strategy, and a discussion of considerations and limitations related to the
use of the technique.
SCHYNS_9781785367274_t.indd 318
10/11/2017 15:20
Leadership-related change with a growth curve approach ­319
OVERVIEW OF GROWTH CURVE MODELING
General Considerations for Modeling Change
To begin, it might be helpful to first consider in the abstract the features a
useful method for analysing change should have. Suppose that you wished
to model a process of leadership-related change or development over time,
what characteristics of your data should you be able to demonstrate or test
with an ideal analytic method? It seems that most fundamentally the analysis should allow us to separate meaningful change in leadership-related
variables (in either a positive or negative direction) from random fluctuation. That is, we would wish to determine whether the observed variability
that appears in the data is consistent and meaningful (i.e., implies systematic change), or does it represent something more banal such as unreliability in our measurement instrument or small and inconsistent changes due
to nuisance factors such as fatigue, mood, and so on?
Next, we would want to be able to characterize the pattern of changes
that occur. For example, for a particular developmental process, do
we see a general increase or decrease in the level of some relevant variable, such as increased self-efficacy for leadership following a training
program or developmental experience (e.g., Hannah, Avolio, Walumba,
& Chan, 2012; Lester, Hannah, Harms, Vogelgesang, & Avolio, 2011), or
a decreased focus on an individual-level leader self-identity accompanied
by an increase in relational or collective identity as leaders gain expertise
in their role (e.g., Lord & Hall, 2005)? If positive or negative change is
present, its rate might stay consistent over time, so that the pattern could
be described as following a linear trajectory. Alternatively, there might be
a non-linear pattern of change, such as a pattern of acceleration in which
change occurs at a faster rate as time progresses, or a deceleration or plateauing effect in which the rate of change is initially high and then slows
down. Such patterns are often referred to in the literature as growth trajectories (Singer & Willet, 2003), and can be described by parameters that
capture the average direction and rate of change in the sample.
Beyond this, the analytic approach should allow us to determine whether
change or growth occurs rather uniformly in the sample as a whole, or
whether there is individual variability in the pattern or rate of change. For
example, because of a variety of experiential, personality and ability-based
factors, we might expect that some persons could quickly acquire critical
leadership skills while for others the process might take longer (e.g., Day
& Dragoni, 2015). Thus, an ideal analytic approach would allow us to
quantify the extent to which there is significant variability across leaders in
the rate and pattern of change. Relatedly, this approach should also allow
SCHYNS_9781785367274_t.indd 319
10/11/2017 15:20
320 Handbook of methods in leadership research
us to determine whether all persons are starting at the same, or different,
initial levels of leadership skill or performance, and what implications this
has, if any, for the pattern and rate of change. For example, we might
want to investigate whether persons receiving a leadership intervention
start at very different skill levels, and whether, as a result, they benefit differentially from the intervention, with some advancing rapidly and others
struggling or plateaued, depending upon their starting skill level.
Finally, once the pattern of change has been described, it would be
advantageous to have some means to predict various aspects of that
pattern via other independent variables. (The GCM literature often refers
to such predictors as “covariates.”) For example, suppose we want to
know whether males and females start out at the same level of leadership
performance, and whether they develop at the same rate, as studied in a
sample of military cadets described in Lord, Hall, and Halpin (2010). In
the example just given, gender is what the growth curve literature (e.g.,
Singer & Willet, 2003) refers to as a time-invariant variable. That is, its
value does not change over time, and we could model its effects directly
onto growth curve parameters such as the intercept and slope coefficients
that describe the pattern of change. The goal orientation effects found
in Day and Sin (2011) are also examples of the effects of time-invariant
variables.
We might also want to determine the effects of time-varying variables
that can change value from one measurement occasion to another, such
as whether a leader’s current level of mood or self-efficacy influences his
or her performance at a specific point in time. For example, Day and Sin
(2011) found that the level of a participant’s leader identity at a particular
point in time was associated with his or her rated leader effectiveness at
that particular point in time. Thus, time-varying variables are modeled
as directly influencing the value of the dependent variable at a specific
point in time, and would normally be variables that we expect to vary
within-person – either randomly or systematically – over the timespan of
the study.
We might also want to consider models that allow us to determine
whether the growth trajectory of one variable relates to the growth trajectory of a second variable. For example, we might investigate whether
the initial status and rate of change over time in leadership identity relate
to the initial status and rate of change in leadership efficacy? Another
example of this type of model, sometimes called a parallel process latent
growth curve model (e.g., Wickrama, Lee, Walker O’Neal, & Lorenz,
2016), comes from the Jokisaari and Nurmi (2009) study, in which they
showed relationships of the dynamic pattern of change in perceived supervisor support with rates of changes in socialization outcomes and salary
SCHYNS_9781785367274_t.indd 320
10/11/2017 15:20
Leadership-related change with a growth curve approach ­321
increases over time. In addition, recently there has been additional work
done on the development of techniques allowing the empirical identification of categories of persons with very different patterns of change in
heterogeneous populations. The growth curve modeling approach offers
ways to address all of these objectives.
Key Characteristics of Growth Models
Although foundational works were also published earlier, the most recent
roots of growth curve modeling include works from the 1980s and 1990s
that cut across multiple analytic approaches and scholarly disciplines. The
underlying statistical approaches for manifest (i.e., observed or raw) variable longitudinal models of change over time, using maximum likelihood
estimation techniques, include latent curve analysis that has most typically been implemented within a structural equation modeling framework
(e.g., McArdle, 1988; McArdle & Epstein, 1987; Meredith & Tisak, 1990),
random coefficients regression (Laird & Ware, 1982), and multilevel modeling (Bryk & Raudenbush, 1987; Goldstein, 1995). The different growth
curve modeling literatures vary in their underlying statistical justifications
for the approaches, and may also differ in some aspects such as the extent
to which unbalanced or missing data can be accommodated, the variety of
estimators available, the fit indices available and the choices for modeling
residual terms.
In addition, these somewhat different approaches may influence preferences for certain kinds of software tools. For example, those using the
SEM-based approach are likely to prefer structural equation modeling
software such as LISREL (Jöreskog & Sörbom, 2015), Mplus (Muthén
& Muthén, 1998–2015), EQS (Bentler, 2006) or similar SEM packages. In
contrast, those taking a multilevel approach are likely to prefer software
such as HLM (Bryk & Raudenbush, 1992) or MLwiN (Goldstein et al.,
1988; Rabash, Steele, Browne, & Goldstein, 2016), while those who use
a random coefficients approach may employ general purpose statistical
packages such as SAS’s PROC MIXED (SAS, n.d.) or Stata’s “xt” routines such as xtreg and xtmixed (StataCorp, 1996–2016). However, across
a very broad range of GCMs (i.e., all linear models and many non-linear
models), the different methods of implementing the analysis should yield
essentially the same results (e.g., Curran, 2003; Ferrer, Hamagami, &
McArdle, 2004). In this chapter, the emphasis will be on the multilevel
modeling and the SEM approaches, as they currently seem to be most used
in the leadership literature.
Applications of GCM are quite popular in many research areas, including developmental, social, and personality psychology, business and
SCHYNS_9781785367274_t.indd 321
10/11/2017 15:20
322 Handbook of methods in leadership research
­ anagement, and education. In part, this popularity may be because
m
GCMs allow one to start with a simple model describing change over time,
and then incorporate additional complexities including covariates, relationships between GCMs, and identification of heterogeneous patterns of
change. Our starting point in this chapter, however, will be a very simple
growth curve model that has two key variables: an independent variable
that reflects time (or some variable related to time, such as age), and
repeated measurements of the focal dependent variable at different occasions, for example, repeated measurements of leader self-efficacy or performance. We’ll consider some particulars for each of these variables next.
Considerations for the independent time variable
In GCM, even though observations occur at specific points in time and
may be indicated with just a few discrete values, time in general is assumed
to be a continuous independent variable. As will be described in more
detail a bit later, a coded time variable is typically employed in the multilevel approach to GCM, in order to indicate when a particular observation has been made. This is in contrast to some other approaches such as
repeated measures ANOVA in which the independent variable indicating
time is treated as categorical. The treatment of time as continuous has
the advantage of also allowing the analysis of datasets in which not every
person is measured at exactly the same time. (Such datasets may have a
large number of different observation times, and tend to be better handled
with the multilevel approach than the SEM approach.)
For example, imagine that you were studying a large number of supervisors who took part in a day-long leadership training program. The
training program needs to be offered at multiple times, spaced out over
a four-week time period. Suppose that you wanted to collect repeated
measures of a variable such as leadership effectiveness that you believe
will change over time as a result of the training, using a self-report survey
method and collecting data at the very start of the training program and
at three later points in time. However, for practical reasons, you must
distribute the post-training surveys at each time period to all of the trainees at once. If the first survey is sent out a week after the last group has
received its training, the time that has elapsed since training is +1 week
for the last group that was trained, but it is +5 weeks for the first group to
receive training. There will be a similar kind of variability in time elapsed
for the remaining two measurement periods. The training-related change
in leadership effectiveness might well depend upon how much time has
elapsed since training (as leadership researchers and practitioners, we
hope that it is a positive change, and that it continues to increase over
time!) This kind of variability can be accommodated in the growth curve
SCHYNS_9781785367274_t.indd 322
10/11/2017 15:20
Leadership-related change with a growth curve approach ­323
modeling analysis by the coding scheme adopted to indicate the time of
measurement.
When data are collected from all persons at exactly the same points in
time (or, in some developmental studies, at the same ages), the dataset is
said to be “balanced on time” or simply, balanced. This term also applies
to those datasets where the original plan was to collect observations from
all persons at the same times, but some individuals are missing responses
for one or more of these response times, in a pattern that is believed to be
missing at random or missing completely at random. Data collections in
which people are observed at different points in time, potentially even with
no two people sharing the same times of observation, are called unbalanced. In general, the multilevel approach to GCM handles unbalanced
datasets more readily than does the SEM approach.
Considerations for the dependent variable
In the models considered in this chapter, the repeated dependent variable is also assumed to be continuous, as well as normally distributed.
(However, methods for similar analyses of ordered categorical dependent variable models exist; for example, see Rabe-Hesketh and Skrondal,
2012 as well as syntax examples on the Mplus website at statmodel.com.)
It is also important that the dependent variable measure has the same
metric or scaling across all measurement occasions, and that the construct
underlying the measure maintains the same meaning (Kline, 2016; Singer
& Willett, 2003). These latter two requirements are necessary so that any
observed change can be attributed to processes occurring over the passage
of time, and not simply to changes in the measurement instrument used
or changes to what it means to participants as they develop. This is often
accomplished by simply using exactly the same instrument for all measurement occasions, but in some circumstances there might be reasons to vary
the content of the dependent variable measurement instrument from one
time to another. For example, if implicit measures involving word fragments were used as the dependent variable (see Chong, Djurdjevic, and
Johnson on implicit measures in Chapter 2 of this volume), the same set
of word fragments should not be repeated from one time to the next, in
order to avoid familiarity effects. In cases like this, it might be possible to
identify equivalent or equated instruments in order to proceed with GCM.
Finally, the SEM approach to GCM has the additional option of modeling the dependent variable as a latent factor, thus potentially increasing
its reliability and construct validity.
The typical maximum likelihood estimation procedure used in growth
curve analysis allows for the accommodation of missing data on the
dependent variable side (Curran et al., 2010). This is a very convenient
SCHYNS_9781785367274_t.indd 323
10/11/2017 15:20
324 Handbook of methods in leadership research
feature, as in practice it can be quite difficult to get a full set of data points
from every individual (or other entity) participating in a longitudinal
study. However, in order for estimation to be unbiased in the presence of
missing data, the missing values should be at least missing at random (i.e.,
missingness is not contingent upon the level of the dependent variable; see
Graham, 2009 or Shafer and Graham, 2002 for a good general overview
of modern methods for dealing with missing data).
Parameter estimation in GCMs
Both fixed and random effects are typically estimated in growth models.
The fixed effects include an intercept parameter that indicates the mean
population level of the dependent variable at the measurement occasion
with a time code of “0” (often chosen to be the initial measurement occasion), and one or more additional parameters that describe the mean
population pattern of change in the dependent variable over time (for
linear models this is often a slope parameter). The key random effects
parameters describe the extent of variability across persons in the coefficients that describe individual growth curve trajectories.
Although in many other statistical applications we are not especially
interested in values of variances, in the GCM context these random
effects can be quite interesting because they tell us whether people tend
to have the same growth trajectories or not. For example, if most participants in a leadership study have almost the same value of intercept for
their individual-level growth trajectories for leadership effectiveness, the
variance of the fixed effect intercept parameter will be small, and we might
conclude that all participants have begun the study with the same level of
effectiveness. However, if intercepts vary widely in value from person to
person, the variance associated with the fixed effect intercept parameter
will be large, suggesting that there is substantial variability in initial levels
of leadership effectiveness. We can similarly look at the estimated variance in slope parameters, to determine whether the rate of change is likely
to be constant or varying. For example, although we may have intuitions
that some people acquire leadership skills more rapidly than others (i.e.,
that there is substantial variability in slope coefficients), GCM combined
with a thoughtful data collection effort could help us to more precisely
determine whether our intuition is correct and if so, more precisely what
the actual extent of variability is.
Short Overview of the Multilevel Approach to GCM
The multilevel model analytic approach builds on the idea that repeated
measurements of the dependent variable are clustered or nested within
SCHYNS_9781785367274_t.indd 324
10/11/2017 15:20
Leadership-related change with a growth curve approach ­325
a higher-level entity such as a person (e.g., Bryk & Raudenbush, 1987;
Rogosa & Willett, 1985). For example, a study might involve ratings of
leader effectiveness, collected every three months over a period of a year.
Thus, the resulting dataset has four effectiveness ratings (taken at months
1, 4, 7, and 10) for each leader included in the study. As can be inferred
from this example, a key difference between the growth curve model and
a more general multilevel model is that for GCM datasets the clustered
observations at the lower level are ordered with respect to time. This
means that time will need to be explicitly treated as a predictor variable in
the multilevel GCM data analysis.
The Level 1 model
The most basic multilevel approach takes the form of a two-level model.
The lowest level (Level 1) specifies the individual growth model –
­describing how an individual changes over time – as shown in the example
of Equation 13.1. This model describes the value of the dependent variable
as depending upon three terms: a constant intercept coefficient, a second
coefficient that is multiplied by time, and a residual term. The coefficients
on the right-hand side of the model can potentially be different for every
person in the dataset. Thus, it captures the shape of the within-individual
growth trajectory for any specific person in the dataset. Another way of
saying this is that the Level 1 model captures the intra- (within-) individual
effects of time on the dependent variable:
Level 1: Yij = p0i + p1iTimeij + eij(13.1)
In this model, Yij is the value of the dependent variable for a given individual (i) at a specific time (j). For example, in a study of change in leader
effectiveness over time, Y13 would be the leader effectiveness rating for
person 1 at the third measurement occasion.
The πs of Equation 13.1 are growth parameters describing change over
time at an individual level, and are estimated from the data. They can be
thought of as analogous to coefficients in a standard regression model, with
p0i representing an intercept term, and p1i representing a slope coefficient
that captures the effect of time on the dependent variable. The values of
Timeij are supplied by the researcher, to indicate the time at which, for a particular individual, a dependent variable measurement was taken. To help
scale the value of the intercept estimate, one of the Timeij values is set at zero.
So, for example, if there are four equally spaced measurement occasions, the
values of Timeij could be coded as 0, 1, 2, and 3. In this example, we would
expect only those values of time to be used, but in datasets that do not have
this balanced structure, individuals could vary in their values of Timeij.
SCHYNS_9781785367274_t.indd 325
10/11/2017 15:20
326 Handbook of methods in leadership research
Finally, eij is a residual term that reflects errors of prediction in the
individual-level growth trajectory. In other words, at each relevant measurement occasion, there will likely be some difference between the actual,
observed value of the dependent variable and the value that is predicted
based on the intercept and slope coefficients for that individual. The
residuals are assumed to be independent and normally distributed, with
a mean of zero. The version of the Level 1 model shown in Equation 13.1
could be used to fit any linear pattern of change, regardless of whether it is
slow or rapid, or involves an increase or a decrease in values over time. If
desired, additional πs could be included in the model to introduce higherorder terms that allow testing for curvilinear effects, such as a quadratic
(squared) effect of time. The effects of additional time-varying predictors
could also be incorporated in this model, such as a measure of experienced
stress at each point in time. Finally, alternative assumptions about residuals could be incorporated in the model, such as whether they are heteroscedastic over time (i.e., variances are unequal) in various patterns, and/
or non-independent.
The Level 2 model
In growth curve analysis, one or more models are also specified at a higher
level. While the Level 1 model describes how an individual changes over
time, Level 2 models concern potential between-persons differences (i.e.,
inter-individual differences) in the values of the growth parameters of the
Level 1 model. These parameters are typically – at least initially – assumed
to randomly vary across individuals. Continuing on with our leadership
effectiveness example, we might believe that potentially both the intercept
and slope parameters can vary meaningfully between individuals. In other
words, the initial value of leadership effectiveness might be relatively low
for some individuals, while others have moderate or high initial values of
the dependent variable. And some leaders might have a relatively rapid
rate of linear change in their effectiveness over time (perhaps as they
benefit from developmental training or experiences), while others change
slowly or not at all. These ideas are captured in the following two Level 2
models:
p0i = g00 + u0i(13.2a)
p1i = g10 + u1i(13.2b)
The model in Equation 13.2a describes individual leaders’ intercept
parameters (p0i) as a function of a latent mean population intercept
value (g00), and (u0i), a term that reflects the deviation of the individual’s
intercept value from the mean intercept value. Similarly, Equation 13.2b
SCHYNS_9781785367274_t.indd 326
10/11/2017 15:20
Leadership-related change with a growth curve approach ­327
describes an individual slope parameter (p1i) as a function of a latent
mean population slope parameter (g10) and a deviation of the individual
slope parameter from the mean slope (u1i). The values of g00 and g10 are
estimated as fixed effects, and describe the aggregate pattern of growth or
change over time. More complex versions of Level 2 equations can also
include additional terms on the right-hand side of the equation representing potential predictors of the values of individual intercepts and slopes.
For example, a potential predictor for individual values of the intercept
for leader effectiveness is the number of years of supervisory experience
that a particular leader has. More specifically, we might expect that there
is a positive relationship between years of supervisory experience and
leader effectiveness. In this example, supervisory experience functions as a
time-invariant predictor, as it has the same value across all measurement
times. If this new predictor variable is added to the intercepts equation, it
now looks like Equation 13.3 below:
p0i = g00 + g01Experiencei + u0i(13.3)
The coefficient for the supervisory experience variable (g01) can be tested
to determine whether it is significantly greater than zero. Similarly, the
previous equation for slopes (Equation 13.2b) could also have an added
term if we believe that prior supervisory experience not only influences
the intercept value but also affects the linear rate of change in leader
effectiveness.
Finally, although they might not at first glance look especially interesting, the values of u0i and u1i from Equations 13.2a and 13.2b can give
researchers valuable information about the homogeneity or heterogeneity of the values of the individual growth parameters. These two random
effects variables are typically reported on output in the form of two variances and a covariance. The two variances, t00 and t11, give an estimate of
the extent to which there is variability across different individuals in the
estimates of the intercept and slope growth parameters, respectively. The
estimated values of these two variances can be tested to determine whether
they are significantly different from zero. Suppose, for example, that in
our study of changes in leadership effectiveness over time, the variance
around the intercept is relatively small while the variance around the slope
is relatively large. This would suggest that while most individuals were
similar in their level of leadership effectiveness at the start of the study,
there was substantial variability in the extent (and perhaps direction) of
their changes in effectiveness over time.
In addition, the covariance between the individual intercept and slope
values, t01, indicates the extent to which individual intercept and slope
SCHYNS_9781785367274_t.indd 327
10/11/2017 15:20
328 Handbook of methods in leadership research
values relate to each other, and its estimate can also be tested to determine
whether it is significantly different from zero. For example, in our illustration, a positive, non-zero covariance would indicate that leaders with
higher initial levels of effectiveness also tend to improve at a faster rate
than those with a low initial level of effectiveness, as might be expected
if a third variable such as leadership motivation or readiness to learn
influenced both one’s initial level of leader effectiveness and one’s rate of
increase in effectiveness. A negative covariance might occur, on the other
hand, if leaders at very high initial levels of effectiveness did not have
much room to improve further so had low rates of change, while leaders at
low initial levels of effectiveness could make easy changes in behavior that
rapidly changed their levels of effectiveness.
Finally, note that the separate Level 1 and Level 2 models are sometimes
combined into a single equation, by substitution (see, for example, Bryk &
Raudenbush, 1992), and the interpretation of output from some analytic
programs can be easier if you are familiar with this single equation expression of the GCM. Also, note that for most GCMs taking a multilevel
approach, two levels such as have just been described are sufficient. But in
some cases, an additional level of nesting is appropriate. For example, one
might be looking at changes in followers over time, and those followers
might in turn be nested in different work groups. In that case, a three-level
model (with work group at the highest level) would be desirable. This type
of situation is one where the multilevel modeling approach has an advantage over the SEM approach, as it is possible to specify and estimate such
three-level models fairly easily.
Short Overview of the SEM Approach to GCM
As illustrated in the path diagram of Figure 13.1, the structural equation
modeling approach to latent growth curve modeling essentially involves
a specialized application of factor analysis, using means and covariance
analysis (e.g., Meredith & Tisak, 1990; Willett & Sayer, 1994). (See Kline,
2016, Chapter 15, for an introduction to working with means structures in
SEM.) In the GCM factor model, one or more common factors that represent change over time are specified, using the repeated measurements of the
dependent variable as multiple indicators of the latent factors. Assuming
that we are modeling linear growth, two latent factors would be specified:
an intercept factor and a slope factor, labeled as “FI” and “FS” respectively in the path diagram of Figure 13.1. Sometimes these are referred to as
chronometric factors. The latent means of these factors – illustrated in the
path diagram by the paths leading from the triangle above the factors – are
estimates of the population mean intercept and slope values that corre-
SCHYNS_9781785367274_t.indd 328
10/11/2017 15:20
Leadership-related change with a growth curve approach ­329
1
FI
1
FS
1
1
1
0
1
2
3
Y1
Y2
Y3
Y4
e1
e2
e3
e4
Note: The rectangles towards the bottom of the figure represent the repeated
measurements of the dependent variable Y at four different measurement occasions.
Towards the top of the figure, the circles labeled “FI” and “FS” are the intercept and slope
latent factors respectively. These two latent factors have freely estimated error variances,
and are allowed to freely covary. A pattern of fixed factor loadings with values of “1” is
used to specify the intercept factor. A pattern of fixed factor loadings with values of 0–3
is used to specify the slope factor. The triangle near the two latent chronometric factors
indicates that their means are estimated. Finally, each of the measured dependent variables
has a latent residual term, e1–e4.
Figure 13.1
ath diagram depicting an SEM model specifying a linear
P
growth trajectory
spond to the g00 and the g10 in the Level 2 equations of the multilevel model
approach that was previously described. In addition, the variances of these
factors provide estimates of what were termed t00 and t11 in the multilevel
context, and the covariance between the two factors estimates t01.
SCHYNS_9781785367274_t.indd 329
10/11/2017 15:20
330 Handbook of methods in leadership research
The manner in which the factor model for GCMs is specified differs
from a standard CFA model in that it tends to have a larger number of
fixed factor loadings. These fixed loadings help to define the chronometric factors in a pre-specified manner that describes the desired pattern of
change, for example, constant, linearly increasing/decreasing, quadratic,
or non-linear change. Depending upon the particular form of growth
trajectory that is expected, the values of these fixed loadings will differ
somewhat. However, to achieve model identification, at least one loading
for each chronometric factor must be fixed to a pre-specified value, rather
than freely estimated, and for each factor except the intercept factor, one
loading must be fixed to zero (McArdle & Nesselroade, 2014).
For example, suppose you want to fit a linear growth trajectory for balanced data with four measurement occasions, spaced a month apart for all
participants in the study. In Figure 13.1, the rectangles along the bottom
of the diagram represent the repeated values of the focal dependent variable (e.g., leader effectiveness), labeled as “Y,” with a subscript to denote
the time of measurement. Each of the Y variables has a latent residual
term (i.e., e1–e4). Note that the factor loadings for the intercept factor have
all been fixed to a value of “1,” as the intercept retains a constant value
across all four measurement occasions. In contrast, the factor loadings
for the slope factor represent a time multiplier for the value of the slope,
analogous to the values of the Timeij variable in the multilevel approach.
In the illustration, at the first measurement occasion (t1), the factor
loading is fixed to a value of “0.” Because we have equally spaced times
of observation for all participants in this example, at Times 2, 3, and 4,
respectively, the fixed values of the factor loadings are “1,” “2,” and “3.”
With this set of fixed factor loadings (sometimes called basis weights), the
intercept factor mean refers to the estimated population mean value of the
dependent variable on the Time 1 measurement occasion (i.e., the occasion
coded “0”), and the slope factor mean reflects the change in the level of the
dependent variable for a one-unit change in time.
Further considerations in coding for time
Depending upon the specifics of one’s study, it might be useful to employ
alternative weights for the slope factor loadings. For example, continuing with the example introduced in the previous paragraph, suppose that
an intervention was made at the second measurement occasion, so that
you wanted the estimated intercept value to reflect the mean level of the
dependent variable at that point in time. In that case, you might prefer
to use values of –1, 0, 1 and 2 for the fixed factor loadings on the slope
factor. Alternatively, suppose that you wanted the estimated slope coefficient to be interpretable as the change from Time 1 to Time 4. To do
SCHYNS_9781785367274_t.indd 330
10/11/2017 15:20
Leadership-related change with a growth curve approach ­331
that, you could use proportional values, making the difference between
the factor loadings for the first and last measurement occasions equal to
one unit, with the loadings for Times 2 and 3 falling proportionally in
between, thus you could choose fixed loading values of 0, 1/3, 2/3, and 1.
(Although described here in the section on the SEM approach, the same
logic can be applied in choosing values for the Timeij variable if the multilevel approach is used.)
Note that any of the three different sets of fixed factor loadings just
described would result in the same value for the overall fit of the GCM.
However, these choices will affect the values of some of the estimated
model parameters. More specifically, when different values of the fixed
factor loadings are used, the slope mean and variance parameters will not
change, and neither will the error variances. However, the intercept mean
and variance will change and so will the covariance of the slope and intercept (McArdle & Nesselroade, 2014).
The previous examples of values for fixed factor loadings were for data
collected at equally spaced intervals. Yet sometimes there may be good
reasons to collect data at unequally spaced intervals. In such situations it
may be worth considering whether a set of factor loadings that reflects the
unequal spacing might be of use. For example, Boswell, Shipp, Payne, and
Culbertson’s (2009) study of honeymoon and hangover effects provides a
nice illustration of the application of GCM to the study of job satisfaction
in the context of work socialization. In their study, they collected data on
newcomer job satisfaction at four points in time, specifically: (a) Time 1,
the first day on-the-job; (b) Time 2, three months after Time 1; (c) Time 3,
six months after Time 1; and (d) Time 4, a year after Time 1. Notice that
the time interval between Times 3 and 4 is twice as large as the interval
between Times 2 and 3. Importantly, they had an a priori rationale for this
data collection schedule, based on both previous socialization research
and input from knowledgeable organization members. Although their
published results suggest that they likely used a 0, 1, 2, 3 coding for time
in their analyses, which indeed may be quite appropriate, they could also
have considered values that reflect the unequal time intervals. One such
fixed factor loading pattern would be 0, 1, 2, 4. (Astute readers might also
notice that one way of interpreting such a loading pattern is that there is a
“missing data collection occasion” halfway in between Time 3 and Time 4,
for which no person in the sample has data.)
To give another example of this issue that allows some more elaboration
of the implications of choice of the fixed factor loading values, consider the
following schedule of data collection at four points in time, with unequal
intervals: Time 2 data were collected at two weeks following Time 1, Time
3 data were collected at six weeks following Time 1, and Time 4 data were
SCHYNS_9781785367274_t.indd 331
10/11/2017 15:20
332 Handbook of methods in leadership research
collected at 12 weeks following Time 1. The choice of fixed loadings on
the slope factor could accommodate this and reflect the differences in the
time intervals between observations. If you used values of 0, 2, 6, and 12 as
your fixed loadings, the estimated slope coefficient would reflect the mean
population change in the dependent variable for a time unit of one week.
Or, you could use values of 0, 1, 3, and 6, in which case the slope coefficient now provides an estimate of the change over a two-week period. Or,
you might alternatively prefer to use values of 0, 0.17, 0.50, and 1. This
latter coding would make the slope coefficient reflect the mean population
change over the time period spanning from Time 1 to Time 4 – a period of
three months. Notice that in determining these values, it does not matter
whether the actual units of time are minutes, days, months, years, or any
other unit. The key idea is to reflect the spacing between measurement
intervals.
Specification of curvilinear, non-linear and other alternative growth
trajectories
Two factors – specifying a linear growth trajectory – may be sufficient
to describe the pattern of change over time in your data. However, it is
not unusual for there to be a second- or even higher-order component
to the growth trajectory. A curvilinear (i.e., polynomial) growth pattern
that includes a quadratic effect can be specified by adding a third chronometric factor, and fixing the loadings from that factor to values equal to
the square of the corresponding linear factor value. For example, factor
loadings on the quadratic factor for Y1 to Y4 would be 0, 1, 4, and 9 (i.e.,
02, 12, 22, 32), if the linear factor (FL) had loadings of 0, 1, 2, and 3. (A
similar approach to specifying a quadratic term can be taken if you are
using the multilevel approach, by adding another term to the Level 1
equation, consisting of p2iTimei2j. Also, a corresponding additional Level
2 equation could be added if you wish to determine variability around
this component.) In addition, you might want to investigate an alternative
re-parameterization of the quadratic model developed by Cudeck and Du
Toit (2002), which allows for the estimation of the quadratic function’s
minimum and maximum values, instead of the more familiar slope and
quadratic components.
A cubic effect could also be specified following a similar strategy in
which the fixed factor loadings are the slope coefficients, taken to a
power of three. However, although the quadratic, and sometimes the
cubic, models have been used fairly extensively by researchers who want
to accommodate deviations from linearity in their models, they often
imply an unrealistic pattern of growth if used to predict the value of the
dependent variable at time points beyond the final time of observation.
SCHYNS_9781785367274_t.indd 332
10/11/2017 15:20
Leadership-related change with a growth curve approach ­333
For example, a quadratic function might fit a growth trajectory that rises
rather quickly initially but then slows down substantially, but its extension into future time periods might imply that the level of the dependent
variable at some point decreases over time, a condition that is probably
not true for variables such as leadership identity, effectiveness, and so on.
That drawback might make other – non-linear as opposed to polynomial –
functions more attractive, even though they may be somewhat more difficult to implement.
Indeed, many processes that at least partly have biological ­underpinnings
– such as learning, or some of the biometric indicators discussed by Dixon,
Webb, and Chang in Chapter 7 of this volume – are likely to change in a
non-linear manner (e.g., Grimm, Ram, & Hamagami, 2011). Non-linear
forms include exponential and logistic functions, as well as other possibilities. If you wish to fit growth trajectories that you believe have a nonlinear – rather than a curvilinear – form, you have some reading ahead of
you as they will not be covered in detail in this chapter, but the investment
of time could be very rewarding! As a starting point, you may want to see
Grimm et al. (2011) for an excellent general overview.
Two additional options might be considered when fitting complex
growth trajectories. The first of these is the piecewise latent growth model
(e.g., Bryk & Raudenbush, 1992). The piecewise model is especially useful
when the nature of your sample is such that you might expect an abrupt
change in slope at some point in time. Often, such changes can occur when
your measurement period spans a transition of some sort. For example,
perhaps you have a series of measurements of leadership identity over
time from a group of managers. The first several measurement times occur
before a promotion, and the remaining measurement times follow the
promotion. We might expect a moderately high but flat or only slowly
increasing level of leadership identity before the promotion, as these
managers have been functioning in their current leader roles for a period
of time. Following promotion, there may be a sudden rapid change in
leader identity as the managers engage in cycles of identity claiming and
granting with new subordinates and peers (e.g., DeRue & Ashford, 2010).
This type of model can be fit by having two slope factors, rather than one.
Pre-promotion identity measures would load on the first slope factor and
post-promotion identity measures would load on the second, with the
promotion as the point of inflection for the piecewise growth trajectory.
For a published example of the application of this type of model, see Li,
Duncan, Duncan, and Hops (2001).
The second alternative model that can be helpful to consider is the
fully latent curve model (McArdle, 1988; Meredith & Tisak, 1990). In this
model, a subset of the fixed factor loadings is freed so that they can be
SCHYNS_9781785367274_t.indd 333
10/11/2017 15:20
334 Handbook of methods in leadership research
estimated. The resulting estimates can be compared to the values of fixed
loadings that would specify growth curves of specific forms, to indicate the
extent of variability from those functional forms. Specifying such models
so that they achieve identification has some subtleties, you may wish to
consult Ghisletta and McArdle (2001) for an example.
Finally, it should be at least briefly mentioned that one important
advantage of the SEM approach to GCM is that it is relatively easy to
specify models in which the dependent variable (e.g., leader effectiveness,
leader identity, etc.) is latent, rather than measured. The GCMs that
have been considered in this chapter so far have had manifest (measured)
dependent variables, thus they fall into the category of “first order latent
growth curve models.” When the dependent variable is latent, the GCM is
frequently called a “second-order latent growth curve model” or a “curve
of factors model” (McArdle, 1988). In such models, the dependent variables are latent factors with multiple indicators, all measured at the appropriate points in time. An advantage of using latent dependent variables is
that reliability is increased because measurement error can be separated
from true variance. Greater reliability might improve the ability to model
the change and to find statistical significance for covariate relationships.
Another advantage of latent dependent variables is that you can directly
test the measurement invariance of the dependent variable across time,
using a multiple group analysis.
A Quick Note about Residual Structures
Residual terms (e1–e4 in the SEM approach or eij in the multilevel
approach) represent the variance in the Yt variables that is not explained
by any of the chronometric factors (and any other variables that are
modeled as having direct effects on Y at a given time, such as time-varying
covariates). An advantage of using an SEM approach to estimating latent
growth curves is that the residual terms can be flexibly modeled and
tested. The default assumption in the specification of the GCM discussed
so far in this chapter has been that the residuals are homoscedastic (i.e.,
of equal magnitude across time) and independent (i.e., uncorrelated with
each other) once the growth component of the model has been properly
specified. These assumptions are often unrealistic with longitudinal data.
Researchers using the SEM approach may place additional – or relax
existing – constraints on the error terms. For example, in most software
packages by default the covariances among the residuals are all fixed to
zero (i.e., independent/uncorrelated residuals), however, some of these
restrictions might be relaxed, allowing adjacent error terms to covary,
as would be implied by an autoregressive error structure. The multilevel
SCHYNS_9781785367274_t.indd 334
10/11/2017 15:20
Leadership-related change with a growth curve approach ­335
modeling approach to GCM also allows for the investigation of autoregressive and heterogeneous error structures (e.g., Curran & Bollen, 2001),
although not as flexibly as in the SEM approach. As mentioned earlier in
the chapter, once the best fit to a functional form has been established, it
is important to test alternative error structures. For an introduction to
this issue, see Singer and Willett (2003). For further study, Wu, West, and
Taylor (2009) have a good – if somewhat technical – discussion of sources
of misspecification in GCMs and the variety of fit indices that may be
employed to help determine the sources of misfit in one’s model.
Conditional GCMs: Adding Covariates
As already indicated in the section on the multilevel modeling approach,
once the general form of the growth trajectory is successfully modeled,
then additional predictor variables can be added to the model. These variables are typically mean-centered before being included in the analysis, to
aid in the interpretation of the resulting parameter estimates. Figure 13.2
shows, in path diagram form, a latent growth curve model that includes
generic time-invariant and time-varying predictor variables. The effects of
time-invariant covariates can be tested for statistical significance to determine whether they influence intercepts and rates of change (i.e., a linear
slope, quadratic term, etc.), while time-varying covariates can be tested
to determine whether they affect the values of the dependent variable
directly. For example, in the Day and Sin (2011) article, leader identity
at each measurement occasion was a significant time-varying predictor of
ratings of perceived leader effectiveness, while various types of goal orientation predicted intercept and slope values.
PRACTICAL CONSIDERATIONS
Planning the Data Collection
In general, longitudinal data collection involves substantial forethought
and planning. Decisions need to be made about the optimal number of
participants, as well as about how frequently and for how many occasions
data should be collected. In making such decisions, you will likely want
to balance requirements based on theory, a knowledge of what previous
researchers have done and statistical requirements, with competing considerations of cost and accessibility. The desirable sample size depends upon
several factors, including the functional form of the growth curve being estimated (more complex forms will require larger samples) and the number of
SCHYNS_9781785367274_t.indd 335
10/11/2017 15:20
336 Handbook of methods in leadership research
Time-invariant
Covariate
1
FI
1
Time-varying
covariate
(shown for T1)
FS
1
1
1
0
1
2
3
Y1
Y2
Y3
Y4
e1
e2
e3
e4
Note: A single time-varying covariate effect is also depicted on the dependent variable
at Time 1 (typical models would also include time-varying covariates at the remaining
measurement times as well).
Figure 13.2
ath diagram for linear growth trajectory with freely
P
estimated time-invariant covariate effects directly on
intercept and slope parameters
measurement occasions (to reach a given level of statistical power, you need
more persons in the study if the number of measurement occasions is small).
In general, maximum likelihood estimators require larger sample sizes, and
also the statistical power to detect effects increases with a greater number
of persons. Statistical power will be decreased when there are missing data.
SCHYNS_9781785367274_t.indd 336
10/11/2017 15:20
Leadership-related change with a growth curve approach ­337
For further specifics, you might wish to consult Zhang and Wang (2009) for
an example of how to implement a power analysis via SAS macros.
When it comes to the issue of choosing how many measurement occasions to have, a minimum of three is advisable even for relatively simple
growth models. Although technically you could fit a linear growth trajectory with only two measurement occasions, having only two measurement
periods does not provide the opportunity to disconfirm a linear form,
much less compare it to a more complex functional form, as a straight line
will fit any two points perfectly. To demonstrate deviations from linearity,
you must have at least three measurement occasions. In addition, if you
expect to fit a polynomial trajectory (e.g., quadratic, cubic), you should
always have at least one more measurement occasion than the highestpowered term in the equation for your functional form. For example, if
you are fitting a quadratic form, the highest power would be 2, so you
would need an absolute minimum of 2 + 1 5 3 measurement occasions,
and it would be preferable to have more than three measurement occasions
in order to help disconfirm a quadratic form if it is not the correct one.
Also, as the number of measurement occasions is increased, the precision
of the estimated growth parameters (e.g., intercept and slope coefficients)
increases. Yet, a desire for precision might need to be balanced with practical concerns. For example, too many measurement occasions might lead
to participant fatigue and dropout or careless responding, and will certainly increase the costs and effort required.
How to space occasions of measurement depends upon the particular
phenomenon you are trying to model. Some processes, such as the development of leader–member exchange relationships might be expected to
occur fairly rapidly, and then remain relatively stable over time. The measurement occasions for such processes probably span days or weeks. In
contrast, the development of certain leadership skills might take months
or years, thus measurement occasions for these variables should be spaced
much further apart over time. Other aspects being equal, you also might
want to consider whether longer intervals could lead to greater dropout
from the study or whether shorter intervals might result in too much carry­
over in responding from the previous measurement occasion. Finally, it is
critical to be certain that you have designed your data collection procedure
to allow you to link responses from the same participant across all measurement occasions.
Structuring the Dataset for Analysis
Depending upon your choice of statistical analysis packages, your dataset
will need to be either in one of two different forms, described by Singer
SCHYNS_9781785367274_t.indd 337
10/11/2017 15:20
338 Handbook of methods in leadership research
I llustration of two dataset forms: (a) person-level dataset,
one data record per individual; (b) person-period dataset, one
data record per measurement occasion per individual
Table 13.1
(a) Person-level dataset
Person
Repeated dependent variable Time-invariant
covariate
Efficacy1 Efficacy2 Efficacy3
1
2
3
etc.
4
3
3
...
5
3
4
...
6
4
5
...
Sex
Time-varying covariate
NegAff1 NegAff2 NegAff3
1
2
2
...
2
3
4
...
1
3
1
...
2
2
3
...
(b) Person-period dataset
Person
Time
1
1
1
2
2
2
3
3
3
etc.
1
2
3
1
2
3
1
2
3
...
Dependent
variable
Time-invariant
covariate
Time-varying
covariate
Efficacy
Sex
NegAff
4
5
6
3
3
4
3
4
5
...
1
1
1
2
2
2
2
2
2
...
2
1
2
3
3
2
4
1
3
...
Note: The same values are displayed in each dataset.
and Willett (2003) as either a person-level dataset or a person-period
dataset (Table 13.1). As can be seen in the table, the person-level dataset
is more similar to the datasets typically used for other types of analysis,
in that it has a “wide” or horizontal structure in which each person in the
dataset has a single data record. The dependent data measures from the
different time points are saved with different variable names. For example,
you might name multiple measures of a leadership self-efficacy measure
taken at different points in time as “Efficacy1,” “Efficacy2,” “Efficacy3,”
and so on, to make clear which measurement occasion each one is associated with. Time-invariant covariates, such as gender or supervisory experience are indicated with a single variable for each. In contrast, time-varying
covariates must be saved as multiple variables, in a manner similar to that
SCHYNS_9781785367274_t.indd 338
10/11/2017 15:20
Leadership-related change with a growth curve approach ­339
used for the dependent variable. For example, if you wanted to treat negative affect as a time-varying covariate, you would need to have a set of variables such as “NegAff1,” “NegAff2,” “NegAff3,” and so on. Person-level
datasets are more typical when a structural equation modeling approach
is used to estimate the GCM.
In contrast, as illustrated in Table 13.1, person-period datasets have
multiple data records for each person in the dataset. Because this typically
results in a dataset with many lines of data, it is sometimes called a “long
form” dataset. Each line of data contains the values for one individual,
for one specific measurement occasion. For example, if a balanced study
had data collected from 100 persons at four points in time, there would be
400 data records in the dataset. In contrast to the person-level dataset, this
data form typically has a variable that explicitly indicates time (e.g., the
measurement occasion, age at which measure was taken, etc.).
Ideally, you would check on whether a person-level or person-period
dataset is required for your analytic package before entering your data
into a dataset, and then input the data accordingly. However, if you end
up with your data in the wrong form, it can generally be easily transposed.
Most broad purpose data analysis packages such as SPSS, SAS, or Stata
have a procedure that allows you to move from one form to the other.
Indeed, even if you plan to use a more specialized software package for
the GCM analyses, it is often easier to use a more general analysis package
such as one of those mentioned above for those data tasks that need to
take place before estimating the GCMs, such as screening for outliers,
creating scale scores from survey items, and assessing reliability.
Preliminary Data Steps
Among the preliminary analysis steps to undertake, you should try to get
a sense of the shapes of individual growth trajectories to see if the function (linear, curvilinear, non-linear, etc.) that you intend to fit is even
plausible for your data. One way in which this is often done is to produce
graphic displays of the individual data points, nested within individuals.
An example of this is shown in Figure 13.3, which displays the pattern
of data points for six different individuals who all have measurements
taken at five points in time. The variability in patterns of the data points
over time for different individuals shown in this figure is fairly typical.
Suppose one wanted to fit a linear growth trajectory to these data.
Although none of the figures shows a strictly linear pattern of change,
this form would not be dismissed out of hand as several of the plots
have a somewhat linear form, and all appear to show a general positive
(upward) trend. A variation on this type of individual plot adds a fitted
SCHYNS_9781785367274_t.indd 339
10/11/2017 15:20
ID = 4
6
5
4
3
2
1
0
1
2
3
4
5
Mean Leader Effectiveness Score
Mean Leader Effectiveness Score
340 Handbook of methods in leadership research
ID = 10
6
5
4
3
2
1
0
1
Measurement Occasion (Time)
4
3
2
1
0
1
2
3
4
5
Measurement Occasion (Time)
ID = 23
6
5
4
3
2
1
0
2
3
4
Measurement Occasion (Time)
Figure 13.3
5
Mean Leader Effectiveness Score
5
3
4
5
ID = 16
6
5
4
3
2
1
0
1
2
3
4
5
Measurement Occasion (Time)
Mean Leader Effectiveness Score
Mean Leader Effectiveness Score
Mean Leader Effectiveness Score
ID = 12
6
1
2
Measurement Occasion (Time)
ID = 29
6
5
4
3
2
1
0
1
2
3
4
5
Measurement Occasion (Time)
lot of data over time by individual, showing varied patterns
P
of individual growth or change
(linear) regression line to the data points in each plot, rather than simply
connecting adjacent values.
Figure 13.4 shows another way in which this issue can be explored. It
depicts a “spaghetti plot” in which all (or for large datasets, a sizeable sample
SCHYNS_9781785367274_t.indd 340
10/11/2017 15:20
Leadership-related change with a growth curve approach ­341
6
Mean Leader Effectiveness Score
5
4
3
2
1
0
1
2
3
4
5
Measurement Occasion (Time)
Note: Alternative versions of such plots may simply connect values of observations or
may fit non-linear functions to each individual’s data points.
Figure 13.4
paghetti plot showing fitted linear regression lines for
S
multiple individuals on the same axes
chosen at random) of the individual growth trajectories are displayed using
the same set of axes. Again, there can be variations on this type of figure in
whether different trajectories are fitted to the individual lines, or whether
raw data points are simply connected. In addition to graphic displays, you
can also simply estimate linear regression models for each individual’s data
(using time as a predictor variable) and inspect the resulting coefficients, to
get a sense of the range of intercept and slope values that would result from
fitting a linear function. For more detail on producing and interpreting plots
and other preliminary analysis procedures, see Singer and Willett (2003).
Assessment of Model Fit and Parameter Estimates
Interpretation of the results from the estimation of GCMs involves both
an assessment of the adequacy of model fit, and significance tests of
SCHYNS_9781785367274_t.indd 341
10/11/2017 15:20
342 Handbook of methods in leadership research
specific estimated parameters. The initial focus should be on fitting the
growth trajectory, without including any of the intended covariates in the
model. In the multilevel approach, the assessment of overall model fit has
tended to emphasize comparisons of alternative models. SEM approaches
typically also involve comparisons of alternative models, but in addition tend to consult a larger number of model fit indices, including the
likelihood ratio test statistic (LRT, also commonly called the chi-square
goodness of fit statistic), the RMSEA, CFI and TLI. Comparisons of
nested models can be made with the LRT, and non-nested models can be
compared with the AIC or BIC. An initial step in data analysis is often to
fit two simple models for comparison purposes. These allow assessing: (1)
whether there is sufficient variance to justify a multilevel analysis and (2)
whether there is evidence for any form of growth or change. Singer and
Willett (2003, Chapter 4) term these the unconditional means model and
the unconditional growth model.
The unconditional means model is extremely simple. It does not
involve a time predictor at all, but merely partitions the total variance
in the dependent variable into two portions – variability associated with
differences across time points within a person, and variability associated
with differences in mean levels across individuals. If variance at either
of these levels is zero or very close to zero, it does not make sense to try
to predict the outcome at that level. The unconditional growth model is
slightly more complex – it does include a linear time predictor in the Level
1 model, but the Level 2 models do not have any additional predictors
(i.e., they are simply in the form of Equations 13.2a and 13.2b). If the
linear slope parameter is not statistically significant, it may mean that
(on average) there is no change over time in the dependent variable, or
that the change takes a complex non-linear form such as an oscillator. A
comparison of results from these two models, along with results from any
more complex forms of growth trajectory (such as polynomial forms) that
are deemed plausible can lead to a determination of the best fitting functional form. This should be followed by some tests of alternative models
to determine whether the error structure is properly specified. Estimations
of models involving covariate effects should only be attempted after these
preliminary models are satisfactorily specified. There is considerably
more detail that should be attended to in this process, but it is beyond the
scope of this overview chapter. If you plan to use GCM, you should read
further in Singer and Willett (2003) or one of the other many excellent
sources on the topic (see mention of some of these in the final section of
this chapter).
SCHYNS_9781785367274_t.indd 342
10/11/2017 15:20
Leadership-related change with a growth curve approach ­343
CONSIDERATIONS, LIMITATIONS, AND
CONCLUDING REMARKS
The goal of this chapter has been to make at least a mention of many
aspects of growth curve modeling and to encourage you to apply it to your
own leadership research interests. However, if you want to become proficient in this technique, you should devote some additional time to more
study of the finer details of the technique. There are many excellent books
with very readable descriptions, and in many cases, they provide sample
syntax for various statistical packages. For example, the Singer and Willett
(2003) book cited in this chapter also has a companion website with many
examples that include both multilevel and SEM approaches. Other helpful
books for newcomers include Duncan, Duncan, and Strycker (2006), and
Wickrama et al. (2016). Persons specifically interested in the multilevel
modeling approach would do well to refresh their acquaintance with relevant sections of Bryk and Raudenbush (1992). It is also helpful to carefully read other research studies that have applied GCM to see how their
authors formulated their research questions and then went about addressing them analytically.
Although many of the methodological and statistical issues have
received emphasis in this chapter, it is critical to remember that a key
ingredient for a top-quality GCM study is a thorough grounding in
theory. Even if there are some exploratory aspects to your empirical investigation, your ability to interpret results depends upon your understanding
not only of the analytic technique but how those results fit in with a body
of literature. For example, the Day and Sin (2011) study described at the
start of this chapter was firmly grounded in theories of leadership development. And, an important idea underlying the Jokisaari and Nurmi (2009)
study that was described is Fichman and Levinthal’s (1991) idea of the
honeymoon in interpersonal relationships. Theory may also inform decisions about how frequently measurements should be made and how many
might be necessary to address your research issue.
This chapter described both the multilevel and the SEM approaches to
GCM. For many applications, either would be a reasonable choice, and
which one is chosen might simply be based on researcher preferences and
familiarity with a particular software. However, as has been discussed by
various authors, including Lindenberger and Ghisletta (2004), there may
be factors that make one approach preferable to the other. For example,
the multilevel approach better handles datasets with unbalanced data and
also those where there are a large number of patterns of missing data.
The SEM approach offers a wider variety of fit indices, some of which
are sensitive to sources of misfit that cannot be specifically identified with
SCHYNS_9781785367274_t.indd 343
10/11/2017 15:20
344 Handbook of methods in leadership research
the multilevel approach (e.g., Wu et al., 2009). The SEM approach is also
preferred if you want to use latent dependent variables, and when you
want to look at relationships between trajectories for two different sets of
dependent variables.
Finally, there are a number of techniques that were not covered in this
overview chapter, but that might be interesting directions for further
learning. Wickrama et al. (2016) provide illustrations and guidance on a
wide variety of growth curve models. In addition, when the sample contains different groups with known membership for whom it is believed
the growth trajectories might differ, multiple group growth curve modeling using an SEM approach provides a fairly straightforward extension
of single group models. This type of technique would allow testing for
differences in leadership development trajectories for men versus women,
or for groups receiving different leadership interventions. When group
membership is not known in advance, but it is expected that there might
be heterogeneity in growth trajectories, latent growth curve mixture
modeling can be used to identify different patterns of change over time –
the chapter by Pastor and Gagné (2013) on mean and covariance structure mixture modeling provides a very readable illustration of a linear
growth mixture model. The Day and Sin (2011) article also provides an
­illustration of a similar approach, described in detail in Nagin’s (2005)
book.
In sum, GCM provides a flexible and useful tool for furthering our
understanding of dynamic processes and relationships in the domains of
leadership and followership. Finding an appropriate source of longitudinal data and learning to properly use GCM takes some investment in time
and effort to achieve, but the results are likely to advance our understanding of dynamic leadership and followership processes.
REFERENCES
Bentler, P.M. (2006). EQS 6 structural equations program manual. Encino, CA: Multivariate
Software.
Boswell, W.R., Shipp, A.J., Payne, S.C., & Culbertson, S.S. (2009). Changes in newcomer
job satisfaction over time: Examining the pattern of honeymoons and hangovers. Journal
of Applied Psychology, 94(4), 844–858.
Bryk, A.S., & Raudenbush, S.W. (1987). Application of hierarchical linear models to assessing change. Psychological Bulletin, 101(1), 147–158.
Bryk, A.S., & Raudenbush, S.W. (1992). Hierarchical linear models: Applications and data
analysis methods. Newbury Park, CA: Sage.
Cudeck, R., & Du Toit, S.H.C. (2002). A version of quadratic regression with interpretable
parameters. Multivariate Behavioral Research, 37(4), 501–519.
Curran, P.J. (2003). Have multilevel models been structural equation models all along?
Multivariate Behavioral Research, 38(4), 529–569.
SCHYNS_9781785367274_t.indd 344
10/11/2017 15:20
Leadership-related change with a growth curve approach ­345
Curran, P.J., & Bollen, K.A. (2001). The best of both worlds: Combining autoregressive and
latent curve models. In L.M. Collins and A.G. Sayer (Eds.), New methods for the analysis
of change. Washington, DC: American Psychological Association.
Curran, P.J., Obeidat, K., & Losardo, D. (2010). Twelve frequently asked questions about
growth curve modeling. Journal of Cognition and Development, 11(2), 121–136.
Day, D.V., & Dragoni, L. (2015). Leadership development: An outcome-oriented review
based on time and levels of analyses. Annual Review of Organizational Psychology and
Organizational Behavior, 2(1), 133–156.
Day, D.V., & Sin, H.-P. (2011). Longitudinal tests of an integrative model of leader development: Charting and understanding developmental trajectories. The Leadership Quarterly,
22(3), 545–560.
Day, D.V., Fleenor, J.W., Atwater, L.E., Sturm, R.E., & McKee, R.A. (2014). Advances
in leader and leadership development: A review of 25 years of research and theory. The
Leadership Quarterly, 25(1), 63–82.
DeRue, D.S. & Ashford, S.J. (2010). Who will lead and who will follow? A social process of
leadership identity construction in organizations. Academy of Management Review, 35(4),
627–647.
Duncan, T.E., Duncan, S.C., & Strycker, L.A. (2006). An introduction to latent variable
growth curve modeling: Concepts, issues, and applications (2nd ed.). Mahwah, NJ: Lawrence
Erlbaum Associates.
Ferrer, E., Hamagami, F., & McArdle, J.J. (2004). Modeling latent growth curves with
incomplete data using different types of structural equation modeling and multilevel software. Structural Equation Modeling, 11(3), 452–483.
Fichman, M., & Levinthal, D.A. (1991). Honeymoons and the liability of adolescence: A new
perspective on duration dependence in social and organizational relationships. Academy of
Management Review, 16(2), 442–468.
Goldstein, H. (1995). Multilevel statistical models (2nd ed.). London: Edward Arnold.
Goldstein, H., Rabash, J., Plewis, I., Draper, D., Browne, W., Yang, M.,. . .Healy, M.
(1988). A user’s guide to MLwiN, version 1.0. University of London, Institute of Education.
Ghisletta, P., & McArdle, J.J. (2001). Latent growth curve analyses of the development of
height. Structural Equation Modeling, 8(4), 531–555.
Graham, J.W. (2009). Missing data analysis: Making it work in the real world. Annual
Review of Psychology, 60, 549–576.
Grimm, K.J., Ram, N., & Hamagami, F. (2011). Nonlinear growth curves in developmental
research. Child Development, 82(5), 1357–1371.
Hannah, S.T., Avolio, B.J., Walumba, F.O., & Chan, A. (2012). Leader self and means
efficacy: A multi-component approach. Organizational Behavior and Human Decision
Processes, 118(2), 143–161.
Jokisaari, M., & Nurmi, J.-E. (2009). Change in newcomers’ supervisor support and
socialization outcomes after organizational entry. Academy of Management Journal, 52(3),
527–544.
Jöreskog, K.G., & Sörbom, D. (2015). LISREL 9.20 for Windows. Skokie, IL: Scientific
Software.
Kline, R.B. (2016). Principles and practice of structural equation modeling (4th ed.). New
York: Guilford Press.
Laird, N.M., & Ware, J.H. (1982). Random effects models for longitudinal data. Biometrics,
38(4), 963–974.
Lester, P.B., Hannah, S.T., Harms, P.D., Vogelgesang, G.R., & Avolio, B.J. (2001).
Mentoring impact on leader efficacy development: A field experiment. Academy of
Management Learning and Education, 10(3), 409–429.
Li, F. Duncan, T.E., Duncan, S.C., & Hops, H. (2001). Piecewise growth mixture modeling
of adolescent alcohol use data. Structural Equation Modeling, 8(2), 175–204.
Lindenberger, U., & Ghisletta, P. (2004). Modeling longitudinal changes in old age: From
covariance structures to dynamic systems. In R. Dixon, L. Backman, & L.-G. Nilsson
(Eds.), New frontiers in cognitive aging (pp. 199–216). New York: Oxford University Press.
SCHYNS_9781785367274_t.indd 345
10/11/2017 15:20
346 Handbook of methods in leadership research
Lord, R.G., & Hall, R.J. (2005). Identity, deep structure and the development of leadership
skill. The Leadership Quarterly, 16(4), 591–615.
Lord, R.G., Hall, R.J., & Halpin, S.M. (2010). Leadership skill development and divergence:
A model for the early effects of sex and race on leadership development. In S.E. Murphy
& R.J. Reichard (Eds.), Early development and leadership: Building the next generation of
leaders. New York: Psychology Press/Routledge.
McArdle, J.J. (1988). Dynamic but structural equation modeling of repeated measures data.
In J.R. Nesselroade & R.B. Cattell (Eds.), Handbook of multivariate experimental psychology (pp. 561–614). New York: Plenum.
McArdle, J.J., & Epstein, D. (1987). Latent growth curves within developmental structural
equation models. Child Development, 58(1), 110–133.
McArdle, J.J., & Nesselroade, J.R. (2014). Longitudinal data analysis using structural equation models. Washington, DC: American Psychological Association.
Meredith, W. & Tisak, J. (1990). Latent curve analysis. Psychometrika, 55(1), 107–122.
Muthén, L.K., & Muthén, B.O. (1998–2015). Mplus user’s guide (7th ed.). Los Angeles, CA:
Muthén & Muthén.
Nagin, D.S. (2005). Group-based modeling of development. Cambridge, MA: Harvard
University Press.
Pastor, D.A., & Gagné, P. (2013). Mean and covariance structure mixture models. In
G.R. Hancock and R O. Mueller (Eds.), Structural equation modeling: A second course.
Charlotte, NC: Information Age Publishing.
Rabash, J., Steele, F., Browne, W.J., & Goldstein, H. (2016). A user’s guide to MLwiN,
version 2.36. University of Bristol, Centre for Multilevel Modelling.
Rabe-Hesketh, S., & Skrondal, A. (2012). Multilevel and longitudinal modeling using Stata
(Vol. II, 3rd ed.). College Station, TX: Stata Press.
Rogosa, D.R., & Willett, J.B. (1985). Understanding correlates of change by modeling individual differences in growth. Psychometrika, 50(2), 203–228.
SAS. (n.d.). The mixed procedure. SAS/STAT(R) 9.2 user’s guide (2nd ed.). Retrieved from
https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm​
#mixed_toc.htm
Shafer, J.L., & Graham, J.W. (2002). Missing data: Our view of the state of the art.
Psychological Methods, 7(2), 147–177.
Singer, J.D., & Willett, J.B. (2003). Applied longitudinal data analysis: Modeling change and
event occurrence. Oxford: Oxford University Press.
StataCorp (1996–2016). Stata statistical software. College Station, TX: StataCorp LP.
Wickrama, K.A.S., Lee, T.K., Walker O’Neal, C., & Lorenz, F. (2016). Higher-order growth
curves and mixture modeling with Mplus: A practical guide. New York: Routledge.
Willet, J.B., & Sayer, A.G. (1994). Using covariance structure analysis to detect correlates
and predictors of individual change over time. Psychological Bulletin, 116(2), 363–381.
Wu, W., West, S.G., & Taylor, A.B. (2009). Evaluating model fit for growth curve models:
Integration of fit indices from SEM and MLM frameworks. Psychological Methods, 14,
183–201.
Zhang, Z., & Wang, L. (2009). Statistical power analysis for growth curve models using SAS.
Behavior Research Methods, 41(4), 1083–1094.
SCHYNS_9781785367274_t.indd 346
10/11/2017 15:20
PART IV
QUALITATIVE METHODS
AND ANALYTIC
APPROACHES
SCHYNS_9781785367274_t.indd 347
10/11/2017 15:20
SCHYNS_9781785367274_t.indd 348
10/11/2017 15:20
14. Qualitative content analysis in leadership
research: principles, process and
application
Jan Schilling
INTRODUCTION
While the stream of leadership studies using qualitative research methods
is still rather small, it has become much more common in recent years to
use them. Different authors have called for more qualitative research in the
leadership area (e.g., Bryman, 2004; Conger, 1998; Gordon & Yukl, 2004;
Insch, Moore, & Murphy, 1997), not simply as an alternative but as a complement to quantitative methods (Stentz, Plano Clark, & Matkin, 2012).
Rather than contempt for and rejection of qualitative methods by leadership researchers, the reasons for the small number of studies using qualitative methods may lie more in uncertainty about their exact application.
While it is quite easy to find excellent introductions to theory, methodology, and qualitative data collection in general (e.g., Merriam & Tisdell,
2015; Neimeyer & Gemignani, 2003) and their application in leadership
research specifically (e.g., Klenke, 2008), it is still sometimes difficult to
gain access to “pragmatic knowledge,” especially when it comes to the
process of qualitative data analysis and interpretation (Schilling, 2006).
The present chapter does not strive to develop a process model for all qualitative research traditions, but to provide insight into the procedures and
practices of one special approach: qualitative content analysis (Mostyn,
1985). As Parry, Mumford, Bower, and Watts (2014) point out in their
summary of qualitative and historiometric methods in leadership research
published in The Leadership Quarterly, content analysis is certainly (one
of) the most widely used qualitative analytical methods in this area.
Content analysis in general and qualitative content analysis in particular are valuable tools in leadership research, as the analysis of oral and
written communications of leaders permits a deep-level examination of
contextually rich leader and/or follower communication, which is highly
relevant in the search for factors related to leader effectiveness (Insch et
al., 1997). The aim of this chapter is to give guidance on how to design the
process of qualitative content analysis, not the “one right way” (Tesch,
1990), but as stimulation for the qualitative researcher.
349
SCHYNS_9781785367274_t.indd 349
10/11/2017 15:20
350 Handbook of methods in leadership research
PRINCIPLES OF QUALITATIVE CONTENT
ANALYSIS
While the general term content analysis is associated with a wide range
of theoretical frameworks, methods, and techniques (Denzin & Lincoln,
2000), it aims at studying the contents, structures, and language of
written (e.g., books, articles, newspaper headlines, historical documents)
and transcribed texts (e.g., interviews, written observations, but also
audio-visual material like TV segments or photography) (cf. Klenke,
2008; Krippendorff, 2004). As Mayring (2000) points out, quantitative
and qualitative content analysis should not be regarded as opposites,
but rather that the latter developed on the foundations of the first.
Quantitative content analysis can be defined as a technique of systematically and objectively identifying specified characteristics (e.g., keywords,
combinations of terms) of communication in order to make inferences
about the frequency of certain variables (Holsti, 1969). Quantitative
content analysis may include word frequencies, place (e.g., number of
times a certain theme is presented on the front page of a newspaper)
and space measurements (e.g., amount of space for articles on a certain
theme), or time counts (e.g., amount of time a certain person gets on
television). The strengths of quantitative content analysis (fitting the
material into a model of communication, clear rules and steps of analysis,
categories at the heart of the analysis, applying criteria of reliability and
validity) were preserved in the development of qualitative content analysis (Mayring, 2000). Objections against quantitative content analysis were
raised with regard to the risk of superficial and possibly distorting word
counts without acknowledging latent contents and contexts. Based on
this criticism, qualitative content analysis was developed to address this
potential weakness.
Qualitative content analysis in particular is defined as “an approach
of empirical, methodological controlled analysis of texts within their
context of communication, following content analytic rules and step by
step models, without rash quantification” (Mayring, 2000, p. 5). It tries to
overcome the shortcomings of purely quantitative content analysis by providing clear guidelines for deriving and interpreting categories (Klenke,
2008). Based on this definition, four basic principles can be derived.
First, qualitative content analysis is methodologically controlled in the
sense of applying systematic procedures. Conger (1998) criticized that
qualitative data analysis procedures (“the dirty work”; p. 114) are often
missed out or described only vaguely in qualitative studies. Qualitative
content analysis addresses this critique by defining clear-cut quality checks
in the process of data analysis. This is necessary to counter the common
SCHYNS_9781785367274_t.indd 350
10/11/2017 15:20
Qualitative content analysis in leadership research ­
351
criticism that qualitative research is “not scientific,” “arbitrary” or purely
“subjective” (cf. Silverman, 2000).
Second, qualitative content analysis is contextual in the sense that it
acknowledges the importance of the specific place and time of data collection, the background as well as relationships of the research subjects
(in the case of interviews or observations) or objects (in the case of documents) for the interpretation of the results (cf. Lord in Chapter 16 of this
volume). For example, in the investigation of the content of written corporate leadership principles, it is necessary to acknowledge that these documents (rather popular in many German companies) are often meant to be
marketing instruments rather than real guidelines for corporate leaders.
This background is crucial to understand that they mainly do not contain
statements concerning negative or destructive leadership and very often
refer to democratic leadership styles (e.g., Schilling, 2005) in terms of a
social desirability effect.
Third, the method is process-oriented as it tries to define distinguishable
steps of data treatment, analysis and interpretation, which build on each
other and thereby form a coherent sequence of decision and action. This
feature of qualitative content analysis is not self-evident as some researchers reject the idea that qualitative research could be conducted in the form
of a linear process of stages (e.g., Miles & Huberman, 1994). Even though
qualitative research requires flexibility (Maxwell, 1998) in the sense that
its process is often recursive and dynamic (Merriam & Tisdell, 2015),
this does not mean that it is impossible to describe a course of action and
concrete rules for handling verbal and textual data (Bachiochi & Weiner,
2002; Creswell, 1998).
Fourth, even if it may seem like a contradiction in terms, qualitative
content analysis is not only focused on qualitative data analysis, but could
better be described as integrative as it often combines qualitative and
quantitative analyses (cf. Insch et al., 1997; Parry et al., 2014; Stentz et al.,
2012). As will be described in more detail below, defining the scope of and
interpreting categories can be supported by different measures of category
frequency. Rejecting rash quantification implies a cautious approach of
unraveling the meaning of data and not prematurely comparing apples
and oranges. This needs to be illustrated with an example. In an interview
study on the meaning of leadership (Schilling, 2001), leaders often named
delegation as an important facet of good leadership. Interestingly, when
explaining their understanding of delegation, only some of the subjects
stressed the importance of empowering, assigning responsibility and
granting latitude to their followers. Many of the interviewees understood
delegation as the process of passing certain tasks (“not suited for a supervisor”) to subordinates. Hence, it would have been a misapprehension of
SCHYNS_9781785367274_t.indd 351
10/11/2017 15:20
352 Handbook of methods in leadership research
the true meaning of the subjects’ conceptions of leadership to just simply
count the number of times the word “delegation” is mentioned.
Based on the discussion above, one may question whether qualitative
content analysis is actually an adequate term to describe the method, as
it may seem too qualitative for quantitative researchers and too quantitative for qualitative researchers (e.g., Bryman, 2004). In fact, mixedmethods content analysis (e.g., Creamer & Ghoston, 2013; Schram, 2014)
or combined content analysis (e.g., Hamad, Savundranayagam, Holmes,
Kinsella, & Johnson, 2016) might be better terms, but are (at least until
now) not commonly used, especially not in leadership research.
However, Klenke (2008) warns that it would be overly simplistic to limit
qualitative research to non-quantitative (or non-statistical) modes of data
collection and analysis: “The qualitative paradigm embraces a diverse
array of methodologies that can be mapped on a continuum ranging from
purely qualitative to highly quantitative” (p. 6). Qualitative content analysis represents the more quantitative end of this continuum. Klenke (2008)
offers five key features of qualitative research that can be used to illustrate
this point. Qualitative (in contrast to quantitative) research is focused
on subjective meanings and interpretations, is inductive (data-driven),
uses purposive/theoretical sampling, derives data from the participants’
perspective, and implies flexible designs. While approaches like grounded
theory (Glaser & Strauss, 1967) or phenomenology (Giorgi, 1997) are
rather purist with regard to these characteristics, qualitative content
analysis deviates from them in different ways. Following systematic procedures in a process of distinguishable steps of analysis naturally restricts
the subjectivity and flexibility of the method (without completely eradicating them). Qualitative content analysis is open to both an inductive,
data-driven approach of building categories and a deductive data analysis
strategy with pre-defined category systems (see below). Likewise, purposive/theoretical sampling (i.e., intentionally selecting participants who
can contribute an in-depth, information-rich understanding of the topic;
Klenke, 2008) is very common, but not necessarily applied in qualitative
content analysis (e.g., analysing the content of complete data bases like
Dissertation Abstracts International in the study by Meindl, Ehrlich, and
Dukerich, 1985). However, the basic feature of taking the participants’
perspective characterizes qualitative content analysis as a method basically rooted in the area of qualitative research.
In summary, qualitative content analysis is a rule-based sequence of
intertwined steps including both data treatment and quality checks that is
sensitive to the context in which the data have been produced and offers
opportunities for generating meaningful qualitative and quantitative
results.
SCHYNS_9781785367274_t.indd 352
10/11/2017 15:20
Qualitative content analysis in leadership research ­
353
AREAS FOR QUALITATIVE CONTENT ANALYSIS IN
LEADERSHIP RESEARCH
As in all empirical studies, qualitative content analysis of course should
be based on a literature review that induces the development of research
questions (Insch et al., 1997). While the method is generally suitable for
a wide variety of topics and research interests in leadership (Bryman,
2004), some areas may be especially interesting. Conger (1998) stresses the
importance of qualitative research as the concept of leadership involves
“multiple levels of phenomena, possesses a dynamic character, and has a
symbolic component” (p. 109). From my point of view, qualitative content
analysis is particularly helpful in investigating this symbolic component of
leadership as the method allows for uncovering meanings and connotations of behavior, concepts, and relationships.
Follower perceptions and sense-making of leader behavior are increasingly acknowledged as being of pivotal importance for the process of
leadership (Lord & Dinh, 2014). Particularly approaches such as ethical
leadership and abusive supervision imply a value-loaded and therefore
necessarily subjective perspective on leadership. Interesting research questions in this area might include the comparison of different groups (e.g.,
leaders vs followers, lower- vs higher-level leaders) concerning their
interpretation of behaviors in certain situations (e.g., under what circumstances is a certain behavior regarded as ethical, when may it be evaluated
as unethical or abusive). As Martinko, Harvey, Brees, and Mackey (2013)
point out, different subordinates perceive a supervisor as inspirational
whereas others view the same supervisor as abusive (e.g., Steve Jobs,
Lyndon B. Johnson). It would be very insightful to understand how different individuals interpret the same behavior (e.g., by showing video scenes
or vignettes). Qualitative content analysis could be a useful tool with
which to analyse the answers to open-ended questions on the perception
and interpretation of certain behaviors in audio-visual scenes or written
scenarios.
A second area of research is concerned with the comparison between
scientific and everyday understanding of leadership-related concepts.
Schilling (2009) investigates the meaning of negative leadership by analysing the content of qualitative interviews with corporate leaders. He is
able to show that leadership practitioners have quite elaborate views on
different facets of destructive and ineffective leadership, their antecedents
and consequences. Similar investigations on the meaning of charismatic,
ethical, or authentic leadership (all of which are terms used in everyday
language) could be stimulating for future research. Similarly, while the
majority of research on implicit leadership theories (ILTs; e.g., Cronshaw
SCHYNS_9781785367274_t.indd 353
10/11/2017 15:20
354 Handbook of methods in leadership research
& Lord, 1987; Offermann, Kennedy, & Wirtz, 1994) has been quantitative
in nature, qualitative content analysis can complement existing evidence
as it investigates what kind of characteristics people have in mind when
they talk about typical or ideal leaders (cf. Schyns & Schilling, 2011). For
instance, it would be interesting to analyse archival data like newspapers,
business periodicals, Internet sources or corporate leadership principles
(cf. Meindl et al., 1985; Schilling, 2005) to find out which characteristics
and skills leaders are expected to possess.
Finally, research on the relationship between leaders and followers (e.g.,
leader–member exchange; Liden, Sparrowe, & Wayne, 1997; Schyns &
Day, 2010) could benefit from qualitative content analysis. It could, for
example, be used to systematically analyse textual diary data from leaders
and/or followers on the development of their relationship: which kind of
evaluations, terms and emotions are used at different points of time to
describe the other person and one’s relation with him or her. Contrasting
the views of the leader and his or her respective followers on the quality
and development of their relationship would be particularly interesting
to get a deeper understanding of the dynamics and subjectivity of leader–
member exchanges.
These are, of course, just some examples of possible research questions
to illustrate different applications of this method that can be a valuable
aid to better understand the complexities and subtleties of perception and
sense-making in the area of leadership.
PROCESS AND APPLICATION OF QUALITATIVE
CONTENT ANALYSIS
One criticism might be that it is insufficient just to look at the data analysis
process (Schilling, 2006). The choice of method should always depend on
the research question(s) (Silverman, 2000). While it is important to discuss
each step in the data analysis process with recourse to the chosen conceptual framework (e.g., implicit leadership theories; Schyns & Schilling,
2011), it will be shown that there are concerns and problems in the course
of qualitative content analysis that transcend the boundaries of different
theoretical perspectives.
The following description aims to give an overview on the decisions
a researcher has to take at different stages of the process of qualitative
content analysis. As its focus lies on the analysis of themes, it may seem
strange to start an overview on content analysis with data collection, but,
as Merriam and Tisdell (2015) have pointed out, qualitative research
emphasizes the importance of beginning analysis early. As will be illus-
SCHYNS_9781785367274_t.indd 354
10/11/2017 15:20
Qualitative content analysis in leadership research ­
355
trated below, this is important as decisions (often implicitly) made in this
phase may restrict possible data analyses at later stages.
Collecting the Data
It is important to distinguish between two fundamental forms of data collection as the basis for qualitative content analysis. Researchers may use
qualitative content analysis with regard to generated (actively creating
material, e.g., interviews; cf. Bresnen, 1995; Schyns & Schilling, 2011; Sims
& Lorenzi, 1992; Waldman et al., 1998) or collected texts (searching for
already existing material, e.g., articles in newspapers and business periodicals, Meindl et al., 1985; corporate leadership principles, Schilling, 2005;
or presidential speeches, Bligh, Kohles, & Meindl, 2004).
With regard to generated texts, most qualitative researchers (e.g.,
Silverman, 2000) recommend tape-recording interviews to make sure
that their content is exactly retained. It should be noted though that tape
recording may deter some potential interviewees from participating in the
study. Also, it may be important to clearly document the context of data
collection (e.g., in the form of a contact or document summary sheet; Miles
& Huberman, 1994), as it may influence the data and should be considered
when interpreting results. Basic questions should include (Schilling, 2006):
●
Who are the interviewees (e.g., position in the firm, team/department)?
●
What is their relationship with the interviewer (e.g., personal con-
tacts, unknown volunteers)?
●
When (e.g., during work or leisure time, which year and month) and
under which circumstances (e.g., during a process of downsizing in
their company, during a crisis) were the interviews done?
●
Where were the interviews done (e.g., in the office of the interviewees,
at their homes, in a conference room)?
●
In which context were the interviews done (e.g., as part of a management learning project, as a research project unconnected to organizational developments)?
●
Were there any disturbances or outstanding reactions from (some of)
the interviewees (e.g., comments after the “official” interview ended,
mails before or after the interviews)?
Likewise, if the study involves collected texts, the researcher should carefully document the following aspects:
●
●
the source (what is the origin of the document?);
aim (what is the intention and target group of the document?);
SCHYNS_9781785367274_t.indd 355
10/11/2017 15:20
356 Handbook of methods in leadership research
context (under which circumstances was the document created?);
retrieval (how, where and when was it retrieved?); and
●
validity (source validity: to what extent is the set of texts to be analysed actually representative of the group of interest?; Insch et al.,
1997).
●
●
For example, in the case of analysing corporate documents (e.g., corporate
guidelines, company reports), it may be important whether the texts focus
on an internal and/or external audience (i.e., in the latter case, positive
impression management may be of particular importance) or whether it
was created at a time of crisis or prosperity (cf. Meindl et al., 1985).
Finally, many authors (e.g., Bogdan & Knopp Biklen, 2006; Miles &
Huberman, 1994) stress the importance of writing down comments and
memos on what the researcher is learning in the course of the interviews
or document collection (e.g., ideas for categories, interpretations). For
instance, while conducting a series of interviews on implicit leadership
theories (cf. Schilling, 2001, 2009), I got the impression from the statements and descriptions of the interviewees that certain leadership beha­
viors often occurred together. Based on this observation, I decided to code
and later analyse the co-occurrence of subjective leadership concepts by
the means of multidimensional scaling (see below). Thus, picking up one’s
impressions and ideas during data collection is often very helpful in later
stages of qualitative content analysis (Klenke, 2008).
Preparing the Data
Before verbal data can be analysed for their content, data preparation has
to take place, which involves three major steps: data transcription, directing data analysis, and condensing data.
At the beginning, the protocols of the interviews (written and/or taperecorded) have to be transferred into text files (data transcription). As
Wiedemann (2013) states, different software packages like Ethnograph,
MAXQDA, NVivo or ATLAS.ti (for a review see Banner & Albarran,
2009; Alexa & Zuell, 2000; Klenke, 2008) have been developed since the
1980s to specifically support manual tasks of qualitative data analysis
(e.g., data preparation, condensing, and coding). Creswell (1998) states
that these programs are especially useful for studies with large or diverse
databases. As they require some effort to learn their handling, these programs are mainly used by researchers who plan to apply them in repeated
studies. For single studies, existing spreadsheet software (e.g., Microsoft
Excel, Apache OpenOffice Calc) may often be sufficient to edit and
analyse the data.
SCHYNS_9781785367274_t.indd 356
10/11/2017 15:20
Qualitative content analysis in leadership research ­
357
While transferring documents (collected texts) typically involves a
rather simple text transfer, the transcription of interviews necessitates
a definition of explicit rules (Schilling, 2006). Before determining these
rules, it is helpful to review the material (i.e., listen to some or all of the
tapes and/or protocols) to obtain an idea of the overall data (Creswell,
1998; Tesch, 1990). Besides rather simple formal aspects of the transcript
(program, font, size, margins), the researcher has to decide how to treat
the following:
●
Dialect or slips of the tongue. Should they be preserved, ignored
or respectively corrected (content-focused)? As most researchers
are only interested in the content of the interview, these aspects are
often ignored/corrected. If dialects are to be transcribed, it should be
defined (e.g., by some examples) how the terms should be spelled in
order to provide an intelligible transcript.
●
Observations during the interview like sounds (like “uhs” or “ers”) as
well as audible behavior (like coughing or drumming of the fingers).
From a pragmatic point of view, it can be recommended to drop
these aspects as long as they do not shape or alter the meaning of the
content (speech focused).
●
The specific questions of the interviewer (besides the main questions
from the interview guideline). For example, a specific question–
answer sequence (Q: “What exactly do you mean by ‘delegating’?” A:
“For me, it implies passing on tasks to followers that I do not have
the time to take care of myself”) could be transcribed in the form of
an answer (“For me, delegating implies passing on tasks to followers
that I do not have the time to take care of myself”).
Schilling (2006) emphasizes the danger of only transcribing the main
guiding questions. For instance, it will not be easy to control if the interviewer broke the defined rules (e.g., by posing leading questions). A
careful researcher may control for these concerns by listening to a random
sample from the tapes and critically searching for such incidents.
If necessary, the texts are made anonymous by replacing names of people
and institutions with descriptive terms (e.g., “our CEO” instead of “Mr.
Smith”). A special coding scheme may be needed and applied here if the
researcher is interested in comparing, for instance, the opinions of different
interviewees towards a certain person or institution (Schilling, 2006).
As the extent of possible analytic units may differ within the data (from
single word to more than a sentence), the researcher has to direct the analysis. The literature is often rather vague in this respect. Locke (2002) states
that the researcher has to use some judgment to decide what a meaningful
SCHYNS_9781785367274_t.indd 357
10/11/2017 15:20
358 Handbook of methods in leadership research
unit of analysis is. Meaningful unit in this sense would mean a “segment
of text that is comprehensible by itself and contains one idea, episode, or
piece of information” (Tesch, 1990, p. 116). It is important to define at
least the boundaries of unitizing. Following Mayring (1994), three kinds
of units can be differentiated:
1. The smallest text component that is to be categorized has to be defined
(coding unit: single word, half-sentence, full sentence, paragraph or
complete text). Of course, this choice depends on the aim and topic
of the study, but normally, single words or half-sentences form the
basis for the smallest coding unit (Insch et al., 1997). For instance, in
the case of analysing leaders’ implicit leadership theories (cf. Schilling,
2001), words like “motivating” or half-sentences like “supporting followers in case of questions” would be enough to code them as single
meaningful statements.
2.The next decision concerns the biggest text component to be categorized in the study (context unit: single word, half-sentence, full
sentence, paragraph or complete text). Again, while depending on
the specific research interest, it can be stated that paragraphs typically form the context unit. For example, while coding corporate
leadership principles (Schilling, 2005), the paragraph “Followers are
supported by their leaders in their personal development. This is done
by the means of daily exchange as well as regular team meetings” was
­categorized as one connected statement.
3.Finally, the order of the analysis has to be determined (sequencing
unit: cross-paragraph or cross-text). The cross-paragraph strategy
(i.e., text after text) should be chosen when the guiding questions (in
case of interviews) or the paragraphs (in case of collected texts) are
highly overlapping (e.g., answers to one guiding question are likely
to occur in the course of another question) and aim at the same topic
from different directions (e.g., the first question in an interview is
focused at the characteristics of leaders in general, the second at those
of good leaders; cf. Schyns & Schilling, 2011). By that, the researcher
gets an idea of the full complexity of each interview or text. If the
questions are rather distinct from each other and/or focus on different
topics (e.g., the first question in the interview is concerned with good
leadership, the second with good followership), the cross-text procedure is helpful in giving an impression of the complexity of possible
answers from different interviewees towards a distinctive topic.
After these initial steps, the process of condensing content analysis can
begin. The next step is to reduce the material to its basic content (called
SCHYNS_9781785367274_t.indd 358
10/11/2017 15:20
Qualitative content analysis in leadership research ­
359
Table 14.1
Example of paraphrasing texts
Original
Paraphrases
“A leader has to make decisions, that
is, you know, leadership. Leaders
decide. But, you certainly know this,
you should not, not do that alone, if
the decision is important, without first
talking to. . . people, I mean, people
that you lead. You should consult
them, discuss the options with them.
And then, of course, communicate
the decision to your followers and
communicate the reasons for it. This is
especially necessary, if you don’t follow
their advice. They will be motivated
by this. This not common, not every
leader does this, unfortunately, in this
firm.”
Making decisions
Talking to one’s followers before
taking important decisions
Consulting followers and discussing
options with them
Communicating the decision and the
reasons for the decision to the followers
especially if the leader does not follow
their advice
Followers will be motivated
Not common practice in the firm
“paraphrasing”) by deleting all the words that are not necessary to understand the statement, and transforming the sentences into a short form (see
Table 14.1 for an example). As paraphrasing can be very time-consuming,
researchers often skip this step, especially if they use content-analysis software and directly categorize meaningful text segments. However, while
often cumbersome, paraphrasing allows the researcher to break down the
often very extensive material to a well-arranged dataset, which makes it
easier to find fitting codes and compare the similarities and differences
within and between the texts.
If the researcher specifically aims at analysing the logical structure of
the texts, remaining statements should be generalized and reduced (see
Table 14.2 for an example). First of all, especially with regard to possible
quantitative analyses later, it is important to make a decision on how to
deal with conjunctions (e.g., “and,” “or,” “but,” “by,” “after,” “because”).
Schilling (2006) recommends dissolving these relationships in order to get
a realistic picture of the complexity of statements in the text. The general
rule applied here should be to divide only statements, if each of them has
a discrete meaning. An example in Table 14.2 (e.g., “Communicating the
decision to followers” and “Communicating the reasons of the decision to
followers”) underlines the difficult task for the researcher. “The complexity in the use of human language makes it unlikely that a researcher will be
able to reach and apply a ‘perfect’ system of rules” (Schilling, 2006, p. 31).
SCHYNS_9781785367274_t.indd 359
10/11/2017 15:20
360 Handbook of methods in leadership research
Table 14.2
Example of generalizing and reducing the paraphrases
Paraphrases
Generalization and reduction
Making decisions
Making decisions
Talking to one’s followers before taking An important decision has to be taken
Talking to followers
important decisions
Consulting followers and discussing
options with them
Consulting followers
Discussing options with followers
Communicating the decision and the
reasons for the decision to the followers,
especially if the leader does not follow
their advice
Leader does not follow the advice of
the followers
Communicating the decision to
followers
Communicating the reasons of the
decision to followers
Followers will be motivated
Followers will be motivated
Not common practice in the firm
Based on theoretical considerations, all statements that are not related
to one’s specific research interests should be deleted at this stage. For
example, Schyns and Schilling (2011) were interested in attributed traits of
leaders in general and chose to delete all statements the subjects made that
included leader behavior.
To ensure data quality, the full material (material-related validity)
should be checked against the original texts, if any relevant statement
has been falsely excluded (based on theoretical considerations) by the
researcher and – if possible – a second person who should be trained in
regard to the process, but not involved in the previous steps. Also, if conjunctions in the text have been split up, it has to be controlled if the defined
rules were kept accurately.
Structuring the Data
As Mostyn (1985) states, the development and application of a category
system (coding) lies at the heart of qualitative content analysis. While it is
very difficult to develop an approach applicable to the multitude of different research questions and aims, two main steps can be distinguished in
the endeavor to code the text.
First, the material may be submitted to a structuring content analysis
SCHYNS_9781785367274_t.indd 360
10/11/2017 15:20
Qualitative content analysis in leadership research ­
361
(Schilling, 2006), which means that the statements are separated in different basic dimensions. For example, in the study on implicit leadership
theories (Schilling, 2001), three fundamental dimensions of implicit leadership theories (“perceived leadership behavior,” “attributed antecedents
of leadership,” “attributed consequences of leadership”) were defined a
priori and the text material was structured accordingly. The researcher’s
basic question at this point of the analysis is if the material includes
clearly separable aspects that do not easily fit into one category system.
However, if the study focuses on only one specific aspect (e.g., attributed
traits of leaders; Schyns & Schilling, 2011), this step may be skipped. Like
before, this step needs to be controlled in order to secure the accuracy of
the structuring. As structuring the material is not particularly error prone
(due to the rather clearly distinguishable nature of the dimensions), a team
check of structure accuracy is normally sufficient. In this process, a pair or
team of researchers discusses samples of the material together to develop a
common understanding of the dimensions. Afterwards, they structure the
material independently from each other, compare their results, document
and resolve all cases of doubt.
As Insch et al. (1997) stress, the true value added by content analysis is
the classification of units into categories. For each basic dimension, a category system is now developed and applied. There are two basic approaches
in the development of a category system (inductive and deductive) that
should not be regarded as mutually exclusive, but rather as complementary (Boyatzis, 1998; Tesch, 1990). Based on the ideas of Glaser & Strauss
(1967) and Strauss & Corbin (1990), the process of qualitative analysis is
sometimes associated with a situation where the researcher will generate
dimensions and categories purely from the data (inferred category scheme;
Insch et al., 1997). Mainly, researchers have at least preliminary models
(assumed category scheme; Insch et al., 1997) guiding their “data-driven”
approach (e.g., charismatic leadership theory as the basis for categorizing
charismatic behaviors; House, Spangler, & Woycke, 1991). This model
can be applied, elaborated on or changed within the course of the analysis.
In some cases, the developed system of dimensions and categories influences theory building and can be used in future studies (e.g., the influential
typology of counterproductive work behavior by Robinson & Bennett,
1995).
Within the framework of qualitative approaches, it is often of central
interest to build the categories as near to the material as possible. However,
category codes should add information to the text and not simply reproduce “through the process of interpretation that simultaneously breaks
down the text into meaningful chunks or segments” (Klenke, 2008, p. 92).
According to Holsti (1969), there are certain criteria to keep in mind when
SCHYNS_9781785367274_t.indd 361
10/11/2017 15:20
362 Handbook of methods in leadership research
Table 14.3
Example of a good code (cf. Klenke, 2008)
Elements
Example
Label: What am I going to call it?
Definition: How am I going to define it?
Destructive leadership
Leaders who are acting in a hostile or
obstructive way towards their followers
Description: How am I going to recognize When respondents explicitly say they have
it?
destructive leaders who are perceived as
undermining the work, performance and
well-being of their followers
Exclusion: What do I want to exclude?
Destructiveness implies active behavior.
Not leading (in the sense of laissez-faire:
e.g., not deciding, not communicating)
does not qualify as destructive leadership
Example: What is an example?
“My supervisor criticizes me and my work
in a humiliating way in front of others”
building or selecting categories. They should reflect the purpose of the
research, be exhaustive and mutually exclusive (avoid multiple classification; cf. Insch et al., 1997; Weber, 1990). Boyatzis (1998) offers a helpful
overview on the features of a good code, illustrated in Table 14.3.
Inductive category development therefore includes the following steps
(cf. Boyatzis, 1998; Mayring, 2000; Schilling, 2006):
1.
Defining preliminary codes. Based on theory, criteria of selection
(i.e., which aspects should be taken into account and which not), and
levels of abstraction (i.e., how far away from the text material) for the
categories have to be established. To accompany the codes, a list of
acronyms for the various categories is established (Conger, 1998).
2.
Step-by-step formulation of inductive categories. If necessary, old
categories should be subsumed or new categories formulated in this
process (Conger, 1998). A rule of thumb in building categories is to
write a formal definition (see Table 14.3) of the code label when a
category contains more than six and fewer than 12 data fragments
(Locke, 2002).
3.
Revising the categories (formative check of reliability). After 10–50
percent of the material has been coded, the agreement of different
raters should be checked, cases of doubt as well as problems with
scope and overlapping of the categories should be discussed within the
research team (Bachiochi & Weiner, 2002).
4.
Checking for idiosyncrasies. After two-thirds of the material has
already been categorized, the coding should be checked again to
SCHYNS_9781785367274_t.indd 362
10/11/2017 15:20
Qualitative content analysis in leadership research ­
363
prevent drifting into an idiosyncratic sense of what the codes mean
(Miles & Huberman, 1994). If possible, the researcher may test the
applicability of the coding instructions at this point with a fresh set of
independent raters (Krippendorff, 1980).
While the exact procedure of deductive category application is often
poorly described (Mayring, 2000), it greatly resembles inductive category
development. Also starting with a theoretical discussion and explanation of the system, the researcher has to define main and – if necessary –
subcategories as well as formulate anchor examples (prototypes) and
­
coding rules (see Table 14.3). An example of my own research (Schilling,
2001) illustrates this process of delimiting categories. If a statement includes
the term “delegating” only with “task” (i.e., delegating a task to a follower),
it should be categorized as part of the category “Planning and Organizing”.
This rule is derived from the definition of the category “Delegating”:
“Allowing subordinates to have substantial responsibility and discretion
in carrying out work activities, handling problems, and making important
decisions.” Therefore, the essence of the delegation is not to give a task, but
to grant freedom for decision and action. Hence, when it comes to coding
the statements, it is of great importance to be careful with superficial resemblance (Schilling, 2006). The described steps of revising and checking the
category system (see above) can be applied to deductive coding accordingly.
But coding not only means assigning each statement to one content category (i.e., thematic coding; Boyatzis, 1998). This would mean disregarding other information within the text material. For the purpose of further
analyses, the statements may also be coded with regard to their context.
Such contextual coding can be defined as a process of categorizing textual
data on the basis of its specific position (e.g., at the beginning or end of a
text or paragraph) and thematic context (e.g., together with other statements of the same or other thematic categories). Possible applications for
contextual coding may include:
●
●
●
sequence analyses – for example, coding the guiding question to
which a statement was made (necessary for analyses of typical
interview courses and of the usefulness of interview questions; cf.
Schilling, 2001);
position analyses – for example, which aspects of leader attributes
are the first to be named as an indicator of high salience in subject’s
implicit leadership theories (Schyns & Schilling, 2011);
analyses of argument structures – for example, the co-occurrence of
statements of different thematic categories in one paragraph or text
as an indicator for certain cognitive connections (cf. Schilling, 2009).
SCHYNS_9781785367274_t.indd 363
10/11/2017 15:20
364 Handbook of methods in leadership research
Finally, some checks have to be made to secure the quality of the data,
especially with regard to thematic coding. In a formative check of reliability, a pair or team of researchers discusses the material together in order to
develop (a common understanding of) the categories (Insch et al., 1997).
They document and resolve cases of doubt – that is, all statements that
cannot be categorized easily (ensuring semantic validity; Weber, 1990).
For a summative check of coding reliability, independent raters should
be used to check interrater reliability (reproducibility reliability; Weber,
1990) of the structuring (either all the material or randomly selected
samples; Conger, 1998). Klenke (2008) provides an overview on different
indices to measure intercoder agreement. Alternatively, the statements
could be coded and then coded again by the same person (intrarater or
stability reliability; Insch et al., 1997) after some time (Erdener & Dunn,
1990). Finally, an important step is to closely examine all statements that
could not be thematically categorized (“misfit analysis”; Schilling, 2006).
The residual statements should be analysed carefully for their frequency
(as an indicator for the inclusiveness of the category system), content
(what aspects are possibly missing in the category system) and if there
are any systematic regularities (e.g., that all statements came from one or
a special group of interviewees). In the example of the study on implicit
leadership theories (Schilling, 2001), the misfit analysis showed that only
33 of all statements (0.6 percent) could not be categorized. Concerning
the content, the majority of these statements were formulated in a very
abstract way (e.g., “leadership situation” as an antecedent). But there was
also a very small number of statements concerning “leadership of the own
person” (self-leadership), a category that proved useful in a later study
on implicit leadership theories in corporate mission statements (Schilling,
2005).
Analysing the Data and Displaying Qualitative and Quantitative Results
As the major focus of this chapter is on qualitative analyses, only some
general statements about the role of quantification and statistical analyses
of the content data will be presented here. Descriptive numerical analyses
in the context of content analysis are sometimes devalued as “simple” and
uninspiring words or themes counting (Mostyn, 1985). From experience of
my own study on implicit leadership theories (Schilling, 2001), three basic
quantitative indicators can be distinguished:
● Absolute
topic frequency (i.e., total number of times a topic is
addressed in all texts). This measure can be used as an indicator of
the salience (i.e., prominence and accessibility) of a theme.
SCHYNS_9781785367274_t.indd 364
10/11/2017 15:20
Qualitative content analysis in leadership research ­
365
●
Relative topic frequency (i.e., average percentage of a topic by text).
Especially in the case of large amounts of text material, relative frequencies may be better suited to summarize the salience of a theme
and also enable simple comparisons within (i.e., in case of different
category systems; see above) and between studies.
●
Person/text frequency (i.e., total number of texts which address a
certain topic at least once). This index offers some insight into the
pervasiveness of a theme within the group of texts or subjects. It
complements the topic frequency measures, which may be biased if
single texts address a certain category very often.
Quantitative analyses in particular may help the researcher avoid weighing single comments too heavily and generalizing findings too quickly.
Although it is tempting to include the most vivid or surprising quotes
(Bachiochi & Weiner, 2002), the described frequency analyses can help to
critically evaluate how representative these statements are for the whole
sample of texts. Finally, beyond the commonly used frequency analyses,
there is a great variety of more complex methods for the analysis of categorical data (Agresti, 2013) that can prove helpful in answering certain
questions the researcher may have (e.g., comparison of subgroups, analysis of latent classes, search for types or configurations).
A fruitful example for further quantitative analyses is non-metric multidimensional scaling (n-MDS). It can be used to analyse the relationships
between the different categories to reveal underlying topics or dimensions
(Schilling, 2009). The data basis for this analysis may be the co-occurrence
of different main categories in one sentence, paragraph or text. Schilling
(2009) gives an example of this: an interviewee stated that negative leadership to him meant “taking decisions without involving the employees and
then not correcting these decisions for reasons of face-saving.” Two different aspects are mentioned in this sentence (i.e., deciding without involving
the employees, not correcting decisions to save one’s face). In order to get an
idea of the implicit structure of the views on negative leadership, Schilling
(2009) counted all cases in which statements of different categories were
mentioned together. This co-occurrence can be interpreted in the way that
these categories are obviously related to each other in the experience of an
interviewee. A frequency matrix of these category coincidences formed the
basis for n-MDS. To explain the idea behind this procedure, Robinson and
Bennett (1995, p. 560) use the analogy of having a computer draw maps of
various dimensions by relying on information about the distances between
cities: “The resulting maps could be one-dimensional in that they would be
placed along a single line; two-dimensional like a typical road map; threedimensional like a globe with bas-relief reflecting elevation; and so forth.”
SCHYNS_9781785367274_t.indd 365
10/11/2017 15:20
366 Handbook of methods in leadership research
Promoting team identity
Being open
for questions
Analysing problems
in the team
Supporting followers
in case of problems
Analysing options
for decisions
Consulting followers
Taking decisions
Managing conflicts
Organizing task
accomplishment
Developing working
conditions
Supporting
Deciding
Staffing
Planning & Organizing
Constructive
Leadership
Networking
Delegating
Motivating
Granting latitude
Being an example for others
Assigning challenging tasks
Figure 14.1
Praising followers
Building personal relations
in the company
Building personal relations
with customers
Example of a concept map (cf. Schilling, 2006)
The goal of the concluding qualitative content analyses is to fracture and
rearrange the data in a way that facilitates the comparison of objects
within and between categories (Maxwell, 1998) in order to draw and verify
conclusions. For that purpose, it is necessary to find adequate forms to
display the results in the sense of using a “compressed assembly of information that permits conclusion drawing and action” (Miles & Huberman,
1994, p. 11). While quantitative results can be displayed parsimoniously
in the form of tables, distributions, and statistical values, the display of
qualitative results is often difficult (cf. Flick, 1991). In the past, the most
frequent, but cumbersome form for qualitative data has been extended
text (in the form of transcripts; Miles & Huberman, 1994), while the
typical, but unsatisfying solution for the display of results was to justify
own conclusions by reporting “typical” quotations (selective justification;
Flick, 1991). Both forms of display are obviously prone to different kinds
of biases. The concept map (Schilling, 2001) in Figure 14.1 may serve as
a helpful example of how to give an overview on the major themes given
by one, a group or – in this case – all the interviewees. (An illustrative and
informed overview on other possible forms of qualitative displays can be
found in Miles & Huberman, 1994.)
Based on the coded statements of all interviewees, dominant (recurrent)
themes were identified within each category by sorting the statements
for each category based on their semantic similarity. By this means, the
SCHYNS_9781785367274_t.indd 366
10/11/2017 15:20
Qualitative content analysis in leadership research ­
367
exact content of the categories is extracted and graphically displayed.
The sequence in the presentation of the themes can be used to express
the rank order in the salience or importance of the different themes (e.g.,
how often a certain theme is addressed in total or by how many subjects).
Schilling (2006) outlines that researchers may discuss patterns (e.g., are
there certain similarities concerning the views on “good” leadership transcending the boundaries of the categories?), cluster objects (e.g., which
categories have many, and which categories have only one or two main
themes?), make contrasts (e.g., how do categories with many main themes
differ from those with only a few?) and comparisons (e.g., if used for an
individual or a group: comparing different views on a topic) and also look
for the unsaid (Mostyn, 1985). Not only the presence, but also the absence
of expected statements can yield interesting results.
Discussing the Validity of the Results
While ensuring reliability (see above) has extensively been discussed in the
literature, Klenke (2008) points to the importance of validity measures
in content analysis. Following Krippendorff (2004), she distinguishes
between different aspects of validity:
●
Construct validity relates to the question if the constructs, theories
and models used in the study have been successfully tested and
applied in previous research. It is an important question for the
validity of a content-analytic study if its results generally correspond
with past research and to discuss reasons for discrepancies (see
Schyns and Schilling, 2011, for an example in the area of implicit
leadership theories).
●
Semantic validity means the appropriateness of the category definitions, the key examples, and the rules for coders (see above).
Researchers should discuss potential problems of semantic validity
on the basis of their results (e.g., which categories produced the
highest numbers of cases of doubt; how often was it necessary to use
the “miscellaneous category”; which examples were rather untypical
of the statements actually coded into a category; which rules proved
difficult to be applied based on the feedback of the coders?).
●
Sampling validity addresses the representativeness of the analysed
material with regard to the population it is supposed to map. As
Klenke (2008) states, in many practical situations texts become available only by their source’s choice. It should be discussed, if and how
the often-used convenience samples may have biased the results (cf.
Schilling, 2009).
SCHYNS_9781785367274_t.indd 367
10/11/2017 15:20
368 Handbook of methods in leadership research
●
Correlative validity is focused on “the degree to which the findings
obtained by one method correlate with findings obtained by other
methods such as observation” (Klenke, 2008, p. 103). While the combination of different methods is of importance to minimize singlemethod biases, it is rather seldom used due to the high effort involved
in conducting such studies. A still intriguing example can be found
in a paper on the romance of leadership by Meindl et al. (1985),
who combined archival and experimental studies to triangulate their
findings.
● Predictive validity implies the degree to which the results of a content
analysis accurately anticipate behavior, events, properties, or states
of affairs. While Klenke (2008) underlines that content-analytic
studies often do not aim at making predictions, but rather describe
the meaning of verbal and textual material, Wofford, Goodwin,
and Whittington (1998) offer an interesting example for predictive validity. They investigated leaders’ schemas and scripts using
open-ended questions and analysed them for statements representing transformational and transactional leadership (Bass & Riggio,
2006). At the same time, their leadership behavior was measured by
Multifactor Leadership Questionnaire (MLQ) ratings of the leaders’
subordinates. The results show a clear relationship between leaders’
schemas and followers’ ratings concerning transformational and
transactional leadership respectively (Schilling, 2008).
CONCLUSION
Many researchers point out that qualitative data analysis does not come
after data gathering (e.g., Insch et al., 1997; Maxwell, 1998; Silverman,
2000; Tesch, 1990), but should be intertwined activities in the research
process. In this sense, the researcher will progressively try to develop his
or her focus of inquiry and test emerging conclusions. Hence, qualitative
content analysis may not always be such a straightforward process as
presented here, but rather resemble a complex, circular process in which
the researcher develops and changes his or her proceedings, generates and
discards his or her ideas (Schilling, 2006). Silverman (2000, p. 121) stresses
that “in most qualitative research, sticking to your original research design
can be a sign of inadequate data analysis rather than demonstrating a
welcome consistency.”
This chapter aimed at describing specific steps and problems in the
course of qualitative content analysis. The intention was not to formulate
rigid, inflexible rules, but rather to give recommendations for professional
SCHYNS_9781785367274_t.indd 368
10/11/2017 15:20
Qualitative content analysis in leadership research ­
369
behavior (De Bruyn, 2003). Possible applications of content analytical
procedures within the field of leadership are manifold. While Klenke
(2008) has documented different examples, it can generally be stated that
the method is especially suited to uncovering the complexity and subjectivity of meanings when people think, write, and talk about leadership.
The combination of qualitative and quantitative analyses is the specific
strength of the method, making it particularly useful in leadership research
to complement and expand traditional survey designs.
REFERENCES
Agresti, A. (2013). Categorical data analysis. New York: Wiley.
Alexa, M., & Zuell, C. (2000). Commonalities, differences, and limitations of text analysis
software: The results of a review. Quality and Quantity, 34(3), 299–321.
Bachiochi, P.D., & Weiner, S.P. (2002). Qualitative data collection and analysis. In
S.G. Rogelberg (Ed.), Handbook of research methods in industrial and organizational psychology (pp. 161–183). Oxford: Blackwell Publishers.
Banner, D.J., & Albarran, J.W. (2009). Computer-assisted qualitative data analysis software:
A review. Canadian Journal of Cardiovascular Nursing, 19(3), 24–27.
Bass, B.M., & Riggio, R.E. (2006). Transformational leadership. Mahwah, NJ: Lawrence
Erlbaum Associates, Inc.
Bligh, M., Kohles, J., & Meindl, J. (2004). Charting the language of leadership: A methodological investigation of President Bush and the crisis of 9/11. Journal of Applied
Psychology, 89(3), 562–574.
Bogdan, R., & Knopp Biklen, S. (2006). Qualitative Research for Education: An Introduction
to Theories and Methods. Boston, MA: Allyn & Bacon.
Boyatzis, R.E. (1998). Transforming qualitative information: Thematic analysis and code
development. Thousand Oaks, CA: Sage.
Bresnen, M. (1995). All things to all people? Perceptions, attributions, and constructions of
leadership. The Leadership Quarterly, 6(4), 495–513.
Bryman, A. (2004). Qualitative research on leadership: A critical but appreciative review. The
Leadership Quarterly, 15(6), 729–769.
Conger, J.A. (1998). Qualitative research as the cornerstone methodology for understanding
leadership. The Leadership Quarterly, 9(1), 107–121.
Creamer, E.G., & Ghoston, M. (2013). Using a mixed methods content analysis to analyze
mission statements from colleges of engineering. Journal of Mixed Methods Research,
7(2), 110–120.
Creswell, J.W. (1998). Qualitative inquiry and research design: Choosing among five traditions.
Thousand Oaks, CA: Sage.
Cronshaw, S.F., & Lord, R.G. (1987). Effects of categorization, attribution, and encoding
processes on leadership perceptions. Journal of Applied Psychology, 72(1), 97–106.
De Bruyn, E.E.J. (2003). Assessment process. In R. Fernández-Ballesteros (Ed.), Encyclopedia
of psychological assessment (pp. 93–97). London: Sage.
Denzin, N.K., & Lincoln, Y.S. (Eds.) (2000). Introduction: The discipline and practice of
qualitative research. Handbook of qualitative research (pp. 1–29). Thousand Oaks, CA:
Sage.
Erdener, C.B., & Dunn, C.P. (1990). Content analysis. In A.S. Huff (Ed.), Mapping strategic
thought (pp. 291–300). Chichester, UK: Wiley.
Flick, U. (1991). Stationen des qualitativen Forschungsprozesses [Steps of the qualitative
research process]. In U. Flick, E. von Kardorff, H. Keupp, L. von Rosenstiel & S. Wolff
SCHYNS_9781785367274_t.indd 369
10/11/2017 15:20
370 Handbook of methods in leadership research
(Eds.), Handbuch Qualitative Sozialforschung – Grundlagen, Konzepte, Methoden und
Anwendungen (pp. 147–179). München: PVU.
Giorgi, A. (1997). The theory, practice, and evaluation of the phenomenological method as
a qualitative research procedure. Journal of Phenomenological Psychology, 28(2), 235–260.
Glaser, B., & Strauss, A. (1967). The discovery of grounded theory. Chicago, IL: Aldine.
Gordon, A., & Yukl, G.A. (2004). The future of leadership research: Challenges and opportunities. Zeitschrift für Personalforschung, 3, 359–365.
Hamad, E.O., Savundranayagam, M.Y., Holmes, J.D., Kinsella, E.A., & Johnson, A.M.
(2016). Toward a mixed-methods research approach to content analysis in the digital age:
The combined content-analysis model and its applications to health care Twitter feeds.
Journal of Medical Internet Research, 18(3), e60. doi: 10.2196/jmir.5391
Holsti, O.R. (1969). Content analysis for the social sciences and the humanities. Reading, MA:
Addison-Wesley.
House, R.J., Spangler, W.D., & Woycke, J. (1991). Personality and charisma in the
U.S. presidency: A psychological theory of leader effectiveness. Administrative Science
Quarterly, 36(3), 364–396.
Insch, G.S., Moore, J.E., & Murphy, L.D. (1997). Content analysis in leadership research:
Examples, procedures, and suggestions for future use. The Leadership Quarterly, 8(1),
1–25.
Klenke, K. (2008). Qualitative research in the study of leadership. Bingley, UK: Emerald.
Krippendorff, K. (1980). Content analysis: An introduction to its methodology. Beverly Hills,
CA: Sage.
Krippendorff, K. (2004). Content analysis: An introduction to its methodology (2nd ed.).
Beverly Hills, CA: Sage.
Liden, R.C., Sparrowe, R.T., & Wayne, S.J. (1997). Leader–member exchange theory:
The past and potential for the future. Research in Personnel and Human Resources
Management, 15, 47–119.
Locke, K. (2002). The grounded theory approach to qualitative research. In F. Drasgow &
N. Schmitt (Eds.), Measuring and analyzing behavior in organizations – Advances in measurement and data analysis (pp. 17–43). San Francisco, CA: Jossey-Bass.
Lord, R.G., & Dinh, J. (2014). What have we learned that is critical in understanding
leadership perceptions and leader performance relations? Industrial and Organizational
Psychology, 7(2), 158–177.
Martinko, M.J., Harvey, P., Brees, J., & Mackey, J. (2013). Abusive supervision: A review
and alternative perspective. Journal of Organizational Behavior, 34(S1), S120–S137.
Maxwell, J.A. (1998). Designing a qualitative study. In L. Bickman & D.J. Rog (Eds.),
Handbook of applied social research methods (pp. 69–100). Thousand Oaks, CA: Sage.
Mayring, P. (1994). Qualitative Inhaltsanalyse: Grundlagen und Techniken [Qualitative
content analysis: Fundamentals and techniques]. Weinheim: Deutscher Studien Verlag.
Mayring, P. (2000). Qualitative content analysis. Forum: Qualitative Social Research, 1(2),
1–28. Retrieved from http://www.qualitative-research.net/index.php/fqs/article/view/1089
Meindl, J.R., Ehrlich, S.B., & Dukerich, J.M. (1985). The romance of leadership.
Administrative Science Quarterly, 30(1), 78–102.
Merriam, S.B., & Tisdell, E.J. (2015). Qualitative Research: A Guide to Design and
Implementation. San Francisco, CA: Jossey-Bass.
Miles, M.B., & Huberman, A.M. (1994). Qualitative data analysis: An expanded sourcebook.
London: Sage.
Mostyn, B. (1985). The content analysis of qualitative research data: A dynamic approach.
In M. Brenner, J. Brown & D. Cauter (Eds.), The research interview (pp. 115–145).
London: Academic Press.
Neimeyer, G.J., & Gemignani, M. (2003). Qualitative methods. In R. Fernández-Ballesteros
(Ed.), Encyclopedia of psychological assessment (pp. 785–800). London: Sage.
Offermann, L.R., Kennedy, J.K., & Wirtz, P.W. (1994). Implicit leadership theories:
Content, structure, and generalizability, Leadership Quarterly, 5(1), 43–58.
Parry, K., Mumford, M.D., Bower, I., & Watts, L.L. (2014). Qualitative and historiometric
SCHYNS_9781785367274_t.indd 370
10/11/2017 15:20
Qualitative content analysis in leadership research ­
371
methods in leadership research: A review of the first 25 years of The Leadership Quarterly.
The Leadership Quarterly, 25(1), 132–151.
Robinson, S.L., & Bennett, R.J. (1995). A typology of deviant workplace behaviors: A multidimensional scaling study. Academy of Management Journal, 38(2), 555–572.
Schilling, J. (2001). Wovon sprechen Führungskräfte, wenn sie über Führung sprechen? Eine
Analyse subjektiver Führungstheorien [What do leaders talk about when they are talking
about leadership? An analysis of implicit leadership theories]. Hamburg: Verlag Dr.
Kovač.
Schilling, J. (2005). Führungsgrundsätze auf dem Prüfstand – Was Unternehmen unter
Führung verstehen? [Corporate leadership principles: What does leadership mean for
companies?]. Zeitschrift für Personalpsychologie, 4, 123–131.
Schilling, J. (2006). On the pragmatics of qualitative assessment: Designing the process for
content analysis. European Journal of Psychological Assessment, 22(1), 28–37.
Schilling, J. (2008). Implicit leadership theories: Theory, research, and application. In
K. Heinitz (Ed.), Psychology in organizations – Issues from an applied area (p. 47–62).
Frankfurt: Lang.
Schilling, J. (2009). From ineffectiveness to destruction – A qualitative study on the meaning
of negative leadership. Leadership, 5(1), 102–128.
Schram, A.B. (2014). A mixed methods content analysis of the research literature in science
education. International Journal of Science Education, 36(15), 2619–2638.
Schyns, B. & Day, D. (2010). Critique and review of leader–member exchange theory: Issues
of agreement, consensus, and excellence. European Journal of Work and Organizational
Psychology, 19(1), 1–29.
Schyns, B., & Schilling, J. (2011). Implicit leadership theories: Think leader, think effective?
Journal of Management Inquiry, 20(2), 141–150.
Silverman, D. (2000). Doing qualitative research – A practical handbook. London: Sage.
Sims, Jr. H.P., & Lorenzi, P. (1992). The new leadership paradigm: Social learning and cognitions. Thousand Oaks, CA: Sage.
Stentz, J.E., Plano Clark, V.L., & Matkin, G.S. (2012). Applying mixed methods to leadership research: A review of current practices. The Leadership Quarterly, 23(6), 1173–1183.
Strauss, A., & Corbin, J. (1990). Basics of qualitative research: Techniques and procedures for
producing grounded theory. London: Sage.
Tesch, R. (1990). Qualitative research: Analysis types and software tools. New York: Farmer
Press.
Waldman, D.A., Lituchy, T., Gopalakrishnan, M., Laframboise, K., Galperin, B., &
Kaltsounakis, Z. (1998). A qualitative analysis of leadership and quality improvement.
The Leadership Quarterly, 9(2), 177–201.
Weber, R. (1990). Basis of content analysis. Thousand Oaks, CA: Sage.
Wiedemann, G. (2013). Opening up to big data: Computer-assisted analysis of textual data
in social sciences. Forum Qualitative Sozialforschung, 14(2), 1–16.
Wofford, J.C., Goodwin, V.L., & Whittington, J.L. (1998). A field study of a cognitive
approach to understanding transformational and transactional leadership. The Quarterly
Leadership, 9(1), 55–84.
SCHYNS_9781785367274_t.indd 371
10/11/2017 15:20
15. Biographical methods in leadership
research
Miguel Pina e Cunha, Marianne Lewis,
Arménio Rego and Wendy K. Smith
How useful are life histories to students of leadership?
To us the answer is clear: Very.
(Kellerman & Webster, 2001, p. 490)
Research portrays leaders as superhuman creatures with extraordinary
powers to change entire organizations or even nations. Literature emphasizes such grandeur in depicting leadership as romantic (Meindl, Ehrlich,
& Dukerich, 1985) or heroic (Allison & Goethals, 2013; Grint, 2010), and
leaders as saviors (Khurana, 2004). Recent methodological approaches
to leadership studies, which highlight surveys and quantitative data,
emphasize these depictions. Such approaches often focus on a narrow set
of variables, highlighting a small handful of factors (superpowers) that
distinguish great leaders from their mediocre counterparts.
While this heroic view of leadership offers much optimism, it fails to
convey a full picture of how leaders develop, grow and change over time,
and the broad array of contextual factors that influence this development. By inflating the power and influence of leaders with one moment of
time, these extant studies neglect the view of leadership as a process not
a position (Barker, 2001; Day, Fleenor, Atwater, Sturm, & McKee, 2014;
Uhl-Bien, 2006). Yet leaders emerge over time. Much of their growth and
development comes from confronting their own foibles and weaknesses
and learning from their mistakes, while constantly facing and addressing
contradictory demands (Bennis & Thomas, 2002). As Day et al. (2014,
p. 79) argued:
[. . .]researchers need to focus on conceptualizing process theories related to the
development of leaders and leadership over time and testing these models using
relevant methodologies. Leadership as a field has perhaps been preoccupied
with proposing and testing static models, even those that hypothesize mediation (i.e., causal) effects. . . But if leadership is a process and not a position, and
leadership development is a longitudinal process involving possibly the entire
lifespan, then we need to put forward comprehensive process models and test
them appropriately.
372
SCHYNS_9781785367274_t.indd 372
10/11/2017 15:20
Biographical methods in leadership research ­373
To offer more processual, holistic insights that capture a more nuanced,
dynamic, and realistic view of how leaders develop and change, we need
alternative research methods (Bryman, Bresnen, Beardsworth, & Keil,
1988). In this chapter, we discuss biographical methods as one such possibility. Biographical methods refer to a variety of approaches including
self-narratives, autobiographies, and historical biographies that explore
an individual’s life story to elucidate nuanced dynamics over time. The
goal of biographical methods “is to inquire into lived experience and to
re-present that experience in a narrative form that provides rich detail and
context about the life (or lives) in question” (Gough, 2008, p. 484). These
methods have a long tradition across social sciences, from Freudian psychology to the Chicago School of Sociology, and offer great opportunity
for leadership studies.
Specifically, biographical methods (1) surface deep-level and holistic
insights into the behaviors, relationships, thoughts, emotions of leaders,
and (2) make sense of how such behaviors, relationships, thoughts, and
emotions develop dynamically over time and in interaction with the
context (Mathias & Smith, 2016). Doing so provides a rich combination
of the breadth of dimensions across a leader’s entire life and the depth
of intimate details about the leader’s life and circumstances over time
(ibid.). This breadth and depth allows for greater insight into factors
regularly excluded from leadership studies. For example, emotions significantly influence leaders’ actions and can be inferred through diaries and
memories, but are often left out of more traditional, quantitative methods
(Voronov & Vince, 2012). Biographical data offers insight into how the
intimate-self influences expressions of the public-self, the context in which
leaders operate, and how the interaction between leaders, their followers
and the context unfold over time. Similarly, self-reflections and sensemaking inform how leaders understand a previous situation and approach
new challenges (Weick, 1993), yet this too is hard to view through a more
traditional lens. Finally, biographical studies often surface tensions and
inconsistencies, evident through different accounts between autobiographies and biographies, opposing insights between different biographies, or
even contrasting views across the life of a leader. Rather than being seen
as a challenge as to which account offers the “true” insight, such inconsistencies highlight the paradoxical tensions inherent in leadership processes
(Lewis, Andriopoulos, & Smith, 2014).
In this chapter we start by defining the nature of biographical methods
and their respective assumptions. We elaborate the main characteristics
of biographical approaches to leadership, followed by a review of the
varieties of biographical methods available to leadership researchers. We
continue by discussing strategies for using biographical methods (mainly
SCHYNS_9781785367274_t.indd 373
10/11/2017 15:20
374 Handbook of methods in leadership research
in terms of data collection, sampling, consistency, reliability, and validity
issues), and limitations associated with their use. We conclude with a note
on the use of biographical methods for developmental and pedagogical
purposes, followed by an illustrative case.
DEFINITIONS AND ASSUMPTIONS
Five clarifying assumptions inform our argument. First, we define
“biographical methods” as a conceptual umbrella to involve multiple
approaches, including self-narratives, autobiographies, historical biographies, and so on. Second, we refer to a “life story” as one outcome from
the data coming from biographical research, rather than a specific method
involving self-narratives (Smith, 1998). Often these life stories emerge
from a leader’s self-narrative (Shamir & Eilam, 2005), their personalized
“account of the relationships among self-relevant events across time”
(Gergen & Gergen, 1986, p, 19).
Third, although biographical methods predominantly capture
­individual-level data, they can also reflect data at other levels (Mathias &
Smith, 2016). At the individual level, biographical methods include data
about leaders’ traits, motivations, strengths, behaviors, attitudes, roles,
and emotions, and about how those characteristics unfold over time. At
the meso-level, biographical methods depict relations between the leader
and the team/organization/context. For example, these data can demonstrate how the environment, key events, or critical people can influence
one’s leadership style and approach. Organizational and economic contexts informed the leaderships of people such as Jeff Bezos, Steve Jobs
and Elon Musk, as we will discuss below. At the meso-level, biographical
studies can further depict how the personalities of leaders impact other
key leaders and followers in the organization. At the organizational level,
biographical methods may be used to characterize leader influence over
organizational cultures, and identities as well as the influence of culture
and identity on leaders. At the macro-level, biographical methods may be
used to make sense of how visionary leadership processes influence collective phenomena such as the emergence of a new industry, a new market or
a new political movement (Mathias & Smith, 2016).
Fourth, the data emerging from biographical methods not only helps
researchers to make sense of the complexities of the leadership phenomenon, they also influence the leader and the leadership process. The
process of building a self-narrative, a life story, may be a source of selfdevelopment (Shamir & Eilam, 2005) with consequences for how leaders
see themselves and how they lead. When a leader engages with perspective
SCHYNS_9781785367274_t.indd 374
10/11/2017 15:20
Biographical methods in leadership research ­375
taking, seeing themselves through the eyes of others (Boland & Tenkasi,
1995; Roberts, Dutton, Spreitzer, Heaphy, & Quinn, 2005), the leader may
make use of such a portrait to carry out developmental actions and change
behaviors in order to realize his or her potential. As Shamir, DayanHoresh, and Adler (2005, p. 16) pointed out: “leaders’ biographies are an
important missing link in leadership research because biographies produce
leaders, and leaders, being at least partially aware of that, produce biographies, and both processes are important to the development of a leadership relationship.”
Fifth, life stories are most valuable when generated from multiple
methods using input from varying sources (Fillis, 2006; Mathias & Smith,
2016; Shamir et al., 2005). Each biographical approach integrates distinct biases. Individuals selectively recount events when retrospectively
making sense of their own lives and the lives of others. Combining multiple approaches that integrate varied viewpoints across different time
periods triangulates evidence and enables greater data reliability and
validity. While the monophonic insight from one biography can foster
a ­
“romantic” view of leadership, a polyphonic approach integrating
multiple lenses advances richer, more complex insights (Bryman et al.,
1988; Kornberger, Clegg, & Carter, 2006; Wilner, Christopoulos, Alves,
& Guimarães, 2014). These varied accounts can provide confirmatory
evidence but also surface inconsistencies (Mathias & Smith, 2016) that
demonstrate dilemmas in a leader’s self-image over time or between self
and others’ understanding of the leader.
CHARACTERISTICS OF BIOGRAPHICAL METHODS
In this section we explore eight main features of the biographical methods:
(1) narrative, (2) holistic, (3) constructivist, (4) context-sensitive, (5)
dynamic and temporally situated, (6) relational, (7) self-reflexive, and (8)
contradiction-sensitive. We highlight how these distinct features work
together to construct a more in-depth view of leadership.
Narrative
Biographical methods draw upon stories about individuals’ lives to help
elucidate their thoughts and actions. These stories can be more historical,
focusing on events, actions and their impact, or more psycho-historical,
emphasizing emotions and cognitions (Shamir et al., 2005). Constructing
a narrative that explores an individual’s story over time can give sense to
one’s lives, decisions, motivations and sensemaking (Fenton & Langley,
SCHYNS_9781785367274_t.indd 375
10/11/2017 15:20
376 Handbook of methods in leadership research
2011). These stories also help offer insight across time, depicting how one
event or experience might inform and define another. As Witherell and
Noddings (in Dhunpath, 2000, p. 547) described:
Stories are powerful research tools. They provide us with a picture of real
people in real situations, struggling with real problems. They banish the indifference often generated by samples, treatments and faceless subjects. They
invite us to speculate on what might be changed and with what effect. And, of
course, they remind us of our persistent fallibility. Most important, they invite
us to remember that we are in the business of teaching, learning and researching
to improve the human condition.
Holistic
Biographical methods enable a more holistic view of individuals by integrating data from multiple informants holding varied lenses across differential points in time. This holistic approach can offer novel insights
about leadership. At the individual level, these methods integrate a
leader’s motives, traits, behaviors, and emotions. Such methods further
demonstrate how multiple elements evolve over time, and how they both
influenced and were influenced by the personal, family, economic, technological, and political context. At the more macro-levels, biographical
methods can offer deep insights into the interactions across multiple
leaders. For example, David Owen (2008), a medic and politician, provided a rich, holist perspective of the impact of dozens of political leaders’
illnesses upon their behaviors. Through data collected in biographies,
historical and medical records, and on the basis of his own first-hand
experience as a politician who met some of these leaders, Owen provided
detailed insights into how the leaders’ illnesses informed their thoughts
and decisions.
Constructivist
Biographical methods adopt a constructivist approach, emphasizing the
meaning individuals create from their experiences, rather than searching
for reliability in accounts (Bruner, 1991). As Dhunpath (2000) noted,
“the focus is not on the factual accuracy of the story constructed, but
on the meaning it has for the respondent. . .the story is a composition
of construed meanings and self-representations” (p. 545). Rather than
emphasize a “courtroom style” of interviewing that seeks to minimize
retrospective biases and strives for accuracy and consistency (Eisenhardt,
1989; Eisenhardt & Graebner, 2007), biographical methods also invite
individuals to offer their own reflections and interpretations of key facts
SCHYNS_9781785367274_t.indd 376
10/11/2017 15:20
Biographical methods in leadership research ­377
and events. These sensemaking processes particularly help leaders reflect
on paradoxical tensions of leadership (Smith, 2014; Zhang, Walkman,
Han, & Li, 2015) and leadership development (Bennis, 1989), as well as
their emotions and cognitions (Lewis, 2000; Vince & Broussine, 1996). A
constructivist approach, however, can be limited when trying to assess a
“true” story about one’s leadership, as individuals are motivated to present
themselves in a positive light (Goffman, 1959; Pfeffer, 2015). As a result,
“self-serving judgments, in which the self is viewed more favorably than
other people, are ubiquitous” (Roese & Olson, 2007, p. 124). One way to
reduce these self-serving biases is through complementing self-narratives
with third-person narratives, the consequence being that the generated
output – that is, the told life story, is more reliable when discrepancies are
included in the overall portrait.
Context-sensitive
Biographical methods surface rich contextual data, situating individuals in
their life circumstances. For example, the socio-technological changes of
the 1970s and 1980s significantly defined leaders such as Jeff Bezos, Steve
Jobs, Elon Musk, Larry Ellison, and Bill Gates. A common way of connecting leader and context consists in introducing timelines that present, in
parallel, the major life events of leaders and the most representative events
of their lifetimes. The timeline and the events narrated in two autobiographies separated by more than a decade allows a better understanding of
how Howard Schultz (Schultz, 2011; Schultz & Yang, 1997) reconfigured
Starbucks’s mission and strategy: “with changes in the economic environment and the passing of time, Howard Schultz made sense of Starbucks’s
pursuit of growth quite differently – moving from an aggressive and
relentless approach to growth to a more restrained and conscientious
approach” (Mathias & Smith, 2016, p. 13).
Dynamic and Temporally Situated
Biographical methods explore the dynamics of individual lives and how
those lives relate to the context. They also aim to capture how leaders
develop over time as a consequence of how they have responded to different life experiences. Over time, people change their motives, feelings,
goals and behaviors. They may gain confidence as leaders and as managers (see, e.g., Catmull, 2014), question their inner purpose, or reinvent
themselves as leaders. Biographical approaches provide useful templates
to approach these types of dynamics. For example, one would not have an
accurate portrait of Steve Jobs as an effective leader of “Apple II” without
SCHYNS_9781785367274_t.indd 377
10/11/2017 15:20
378 Handbook of methods in leadership research
considering his failure as leader of “Apple I.” One of the rewards of the
biographic method is that it may help leaders gain a better understanding
of the “inner theatres” (Kets de Vries, 1995) underpinning their actions.
The case of Steve Jobs is illustrative of how he made sense of the traumatic
experience at Apple I. Several years after his dismissal, in the famous
speech at Stanford University1 (June 12, 2005) he observed that:
[. . .]it turned out that getting fired from Apple was the best thing that could
have ever happened to me. The heaviness of being successful was replaced by
the lightness of being a beginner again, less sure about everything. It freed me
to enter one of the most creative periods of my life.
Relational
Understanding someone’s life necessarily involves taking central human
relationships into account. Biographical methods depict the relational
environment of individuals, the human connections that define one’s life.
Leaders accomplish their goals through relationships, and these fulfill
several functions, from affiliative to instrumental (e.g., Colbert, Bono, &
Purvanova, 2016). Biographical accounts should involve an element of
polyphony (Kornberger et al., 2006), not to counter the person’s narrative but to enrich it with different points of view. As Jones (1983) notes,
biographical methods can represent leaders as part of a cultural milieu
that takes into account collective meaning systems such as assumptions,
social rules, conventions and the like, and frames individuals’ actions as
articulated within a social context. To compose a more accurate picture of
the complex social-relational world of leaders such as Elon Musk (Vance,
2015) or Jeff Bezos (Stone, 2013a), one needs to consider what they tell
about themselves and what others (e.g., former and current employees,
significant others) tell about them. Different employees told very different
stories about Jeff Bezos’s temperament and sociability. For the beholders,
Bezos is all those stories simultaneously, and gaining access to the paradoxes and contradictions of leaders, their multiple stories, is difficult if not
impossible via more canonical methodological procedures such as surveys.
Self-reflexive
Bennis (2009) argued that the essential task for leaders is to reflect
upon their inner self, their dilemmas and uncertainties in order to better
inform their own leadership. The life story can be used to learn about
how leaders engage in this identity work, as well as being a vehicle for
ongoing self-reflection (Sonenshein, Dutton, Grant, Spreitzer, & Sutcliffe,
2013; Sturm, Taylor, Atwater, & Braddy, 2013; Tawadros, 2015). Self-
SCHYNS_9781785367274_t.indd 378
10/11/2017 15:20
Biographical methods in leadership research ­379
reflection, however, can be problematic, particularly for leaders with low
self-awareness competencies. On the other hand, comparing self-reflection
accounts with hetero-accounts (i.e., accounts provided by others) may
be crucial to understanding how the (dis)connection between the leader’s
inner and the outer theaters explain his or her behaviors, decisions, relations and performance (Sturm et al., 2013).
Contradiction-sensitive
The biographical approach also offers insights into how individuals understand and respond to paradoxes and contradictions. As Musson notes:
[A]llowing people to explain for themselves the experience of contradictions
and confusions, moments of indecision and turning points, can illustrate
graphically. . .how individuals and organizations function, more than methods
that reduce experience to abstracted definitions and moribund descriptions. In
this case the focus of the research would be specific lives, as they have construed
and developed, within the organization. (Musson, 1998, pp. 13–14)
This self-reflection becomes particularly relevant in leadership studies,
given their extensive contradictory yet interrelated tensions (Lewis et al.,
2014). For example, the “sanctified” Nelson Mandela (Roberts, 2008),
was “a man of contradictions” (Pfeffer, 2015, pp. 97–98). Queen Victoria
“was among the most fascinating and self-contradictory of all British
monarchs” (Wilson, 2014, p. 10). Consider also the following illustration
about Elon Musk:
When Musk sets unrealistic goals, verbally abuses employees, and works them to
the bone, it’s understood to be – on some level – part of the Mars agenda. Some
employees love him for this. Others loathe him but remain oddly loyal out of
respect for his drive and mission. . . Employees fear Musk. They adore Musk.
They give up their lives for Musk, and they usually do all of this simultaneously.
Musk’s demanding management style can only flourish because of the otherworldly – in a literal sense – aspirations of the company. (Vance, 2015, pp. 17, 218)
Isaacson’s (2011) biography of Steve Jobs portrays the psychological complexity of the man in a way that, for example, the same author’s
work on the leadership lessons (Isaacson, 2012) does not. The difference
between these two texts is instructive. In the biographical account, the
reader meets a person with contradictory facets. The complexity of the
subject is notorious. The reader learns about Jobs’s demanding approach
to employees, his division of the world into binary categories of excellence
and mediocrity, his unique way of expressing tough love. This makes the
character thicker, more nuanced, more interesting, and more human. The
SCHYNS_9781785367274_t.indd 379
10/11/2017 15:20
380 Handbook of methods in leadership research
Steve Jobs portrayed by Isaacson is interesting not because of his product
genius, but because he was able to lead people in an inspirational way as
a function of and in spite of his flaws as a manager and as a person. The
outcome of this biography is clear: one can derive a number of leadership
lessons from Steve Jobs but should not necessarily manage à la Steve Jobs,
because he was so singular as a person that there is no point in imitating
him, as he himself pointed out. This is a critical lesson, as one of the problems with leadership research comes from the assumption that by emulating great leaders one can become a great leader. This risk is mitigated in
the complex portrait of leaders composed through biographical research.
In comparison, the leadership lessons advanced by Isaacson (2012) do
not uncover any of the contradictions of the person. They are managerial
prescriptions that could in principle be applied by competent managers
in whatever circumstance. This contrast exemplifies why biographical
methods must be designed to collect polyphonic accounts.
VARIETIES OF BIOGRAPHICAL SOURCES
Life stories draw from a variety of sources, each with distinct benefits and
detriments. Table 15.1 summarizes the main features of several biographical sources, classified into two main categories: self-narratives and thirdperson narratives.
Self-narratives surface a leader’s inner reflections, the public identity
the leader wishes to portray, or the rationalization efforts the leader uses
to make sense of his or her achievement and failures. Self-narratives also
illuminate a leader’s development, beginning with early influences from
family relationships, especially parental influences. For example, Steve
Jobs’s self-narrative offered insight into how his family background
influenced his entrepreneurial accomplishments. In his famous Stanford
speech, he started by sharing his life story, explaining that “[t]he first story
is about connecting the dots.” He “connected the dots” by relating apparently disconnected events: (1) the adoption by a family that promised to
his biological mother that he would go to college; (2) going to college; (3)
dropping out from college and taking a calligraphy class; and (4) designing
the Mac: “Of course it was impossible to connect the dots looking forward
when I was in college. But it was very, very clear looking backward 10
years later. Again, you can’t connect the dots looking forward; you can
only connect them looking backward.”
Third-person narratives, those on the right side of Table 15.1, are especially relevant to understand the leader’s outer theater. With the exception of biographies authorized or commissioned by leaders themselves,
SCHYNS_9781785367274_t.indd 380
10/11/2017 15:20
381
SCHYNS_9781785367274_t.indd 381
10/11/2017 15:20
Interviewing
Methods
Biographical;
self-analysis,
introspection (the
data may be used by
the leader themselves
to support his or her
autobiography, or
by a researcher who
portraits the leader
based on the written
material)
More or
Holistic
less holistic,
depending on the
research question
Reach
A “document created
by an individual who
has maintained a
regular, personal and
contemporaneous
record” (Alaszewski,
2006, p. 1)
The researcher
follows a
protocol and
invites the focal
individual to
focus on his or
her whole life, or
on a specific part
of his or her life
Individual diaries
(own-initiative)a
Description
Self-narratives
shared through
life interviews
Self-narratives
Table 15.1 Varieties of biographical methods
Biographical
works
accepted and
aligned with
the person’s
vision
Authorized or
commissioned
biographies
Biographical;
self-analysis,
introspection,
personal diaries
Interviewing,
documents
Holistic (i.e., entire Holistic
life story)
Autobiographical
works. Leaders
defend their
personal
perspectives in the
first person
Autobiographies
and memoirsb
Interviewing,
document analysis
Holistic
Biographical works
non-endorsed nor
necessarily aligned
with the subject’s
perspective
Non-authorized
biographies
Biographical
leadership
studies
Historical
methods, in
general
Holistic
As mixing the
other methods in
this row
Sectional (i.e.,
leadership
behaviors only)
Biographies
Biographical
focused on past approaches
leaders
focused on the
leadership roles
rather than in
the entire life of
leaders
Historical
biographies
Third-person narratives
382
SCHYNS_9781785367274_t.indd 382
10/11/2017 15:20
Present
understanding of
one life
Allowing
researcher (a)
to probe deeper
into areas of
interests and (b)
observing nonverbal behaviors
(Mathias &
Smith, 2016)
Limited selfreflection,
defensiveness,
justification
Preventing a
detailed and
thorough account
of the life story.
Missing relevant
life events not
Advantages
and benefits
Disadvantages,
risks, and
drawbacks
Self-narratives
shared through
life interviews
(continued)
Temporal
focus
Table 15.1
Unavailability to the
public. Informally
written and often
haphazardly kept
Trivial daily
occurrences
disturbing
identifying
important events
(Mathias & Smith,
2016)
Minimizing recall
bias. Allowing
access to sensitive
information that the
targeted individual
might not share
through other
means. “Unfiltered”
account of events
(Mathias & Smith,
2016)
Present
understanding of
one life
Individual diaries
(own-initiative)a
Self-narratives
Hagiographic
propensity,
defensiveness,
justification.
Self-serving and
sensational (to
promote oneself or
sell books)
Recall bias
Focusing on
Presenting one
leader’s life with
no filters
Providing depth
through revealing
a rare glimpse into
the thoughts and
relationships of
the targeted leader
(Mathias & Smith,
2016)
Present
understanding of
one life
Autobiographies
and memoirsb
Parts of the
biography may
be omitted.
Biased
view as a
consequence of
a fascination
for the focal
individual
(Mathias &
Smith, 2016)
Presenting a
leader’s life
with other
voices. More
polyphonic
Present
understanding
of one life
Authorized or
commissioned
biographies
A sensationalist
approach may
be tempting.
Biased view as a
consequence of
a fascination (or
disdain) for the
focal individual
(Mathias & Smith,
2016)
Balanced
presentation and
multiple viewpoints
of the leader’s life.
More polyphonic.
Critical, less
biased, perspective
(Mathias & Smith,
2016)
Present
understanding of
one life
Non-authorized
biographies
Biographical
leadership
studies
Disadvantages
of time
Benefits from
the advantages
of time; the
case can be
used in context
In comparison
with other
biographical
accounts fails to
put the leader’s
life in context
Conceptually
efficient way of
exploring a life
cycle
Explaining past Understanding
lives of leaders the life cycle of
leadership
Historical
biographies
Third-person narratives
383
SCHYNS_9781785367274_t.indd 383
10/11/2017 15:20
Shamir et al.
(2005)
Wilson (2014) based
the biography of
Queen Victoria on
her diaries (“she
kept journals
from infancy to
old age” and “her
diaries were those
of a compulsive
recorder,” p. 8)
Branson (2007);
Uribe (2012);
Schultz (2011)
internal rather
than external
success factors
(attribution bias)
(Mathias & Smith,
2016)
Isaacson
(2011); Vance
(2015)
Schlender & Tetzeli
(2015); Stone
(2013a)
Kets de Vries
(2004a, 2004b)
Harari (2014);
Zúquete (2011)
Notes:
a. Diary as a method of collecting, over a limited time period, specific data at the researcher request is not included (Sonnentag, Dormann, &
Demerouti, 2010; Zacker & Wilden, 2014).
b. The terms “memoirs” and “autobiographies” are often used interchangeably, although some have argued that while the latter encompass one’s
entire life, the former focus more narrowly on a specific theme or event (see Mathias & Smith, 2016).
Examples
included in
the researcher
agenda (Mathias
& Smith, 2016)
384 Handbook of methods in leadership research
third-person narratives produce a more distant picture of the leader and
embrace more diverse, varied viewpoints. Organizational theorists have
used third-person narratives to explain the leadership of major historical
figures such as Alexander the Great (Kets de Vries, 2004a), Shaka Zulu
(Kets de Vries, 2004b), Cleopatra (Riad, 2011), and many others. They
also have been considered to depict modern corporate leaders such as
Steve Jobs (Schlender & Tetzeli, 2015), Jeff Bezos (Stone, 2013a), Elon
Musk (Vance, 2015), and Larry Ellison (Wilson, 2002). In the case of
historical leaders, research is less personal and more historically situated
(e.g., Kabalo, 2017) with biographical methods used to study the recurs