Stephen E. Humphrey and James M. Lebreton
Stephen E. Humphrey and James M. Lebreton
Stephen E. Humphrey and James M. Lebreton
INTRODUCTION
Stephen E. Humphrey and James M. LeBreton
The domain of social science is that of inter- Scholars often like to focus solely on one level
actions between people—how people think, feel, to simplify theoretical and empirical challenges. In
and behave, and how the collectives to which they contrast, this handbook begins with the premise
belong do the same. In different disciplines, there that social science is fundamentally multilevel
is often a particular focus on one aspect of this in nature—the specific level or levels that are
(e.g., psychology generally begins with a focus on the focus on a particular scholar (or normatively
the individual, sociology on the system), but resident studied within disciplines) may vary. However,
in all of these disciplines is the idea that there are the thesis of this handbook is that ignoring that
multiple levels within which the focal units reside. multiple levels exist (and concentrating on only one
For example, within the organizational sciences level) is no longer a productive option for social
(broadly and liberally defined), there has been a scientists. For example, studying organizations
growing awareness of the importance of unpacking without considering their industry is a fatal flaw
multilevel relationships to explain social phenomena within strategic management research; studying
in work contexts. Employees are frequently students without considering their classroom, their
members of a team (or multiple teams); they exist school, or their area is a nonstarter in educational
within a social network of relationships, working psychology.
under a supervisor (or multiple supervisors), within Even when scholars explicitly acknowledge the
or spanning organizations (Klein & Kozlowski, multilevel nature of the field (e.g., Chan, 1998;
2000). Teams are composed of individuals who are Klein & Kozlowski, 2000; Morgeson & Hofmann,
embedded in multiple dyadic relationships; the 1999), the guidance they have produced has
teams reside within or span departments and are been isolated to a small subsection of the broader
nested within or across organizations (Humphrey organizational domain. Surprisingly, micro research
& Aime, 2014). Organizations are composed of (i.e., research focused on intra- and interindividual
many individuals, residing in multiple departments phenomena) has had little cross-pollination with
or functions, and the organizations themselves may macro research (i.e., research focused on intra-
be embedded within multiple joint ventures or and interorganizational phenomena). This lack of
alliances (Lepak, Smith, & Taylor, 2007). Moreover, cross-disciplinary connections has occurred despite
all of these subjects exist across time. These levels the fact that both micro and macro researchers are
define organizational research, bounding theories inherently interested in the multilevel nature of
and shaping discourse. (organizationally relevant) phenomena. Essentially,
http://dx.doi.org/10.1037/0000115-001
The Handbook of Multilevel Theory, Measurement, and Analysis, S. E. Humphrey and J. M. LeBreton (Editors-in-Chief)
Copyright © 2019 by the American Psychological Association. All rights reserved.
3
Humphrey and LeBreton Copyright American Psychological Association
what the literature has been sorely lacking is within the social and organizational sciences,
an inclusive guide to multilevel research that and failure to consider context severely limits the
recapitulates and extends the current state of the application and interpretation of theoretical models.
science. After reviewing where the study of context has
We thus felt that it was time for a systematic been, she introduces a structure for interpreting,
and inclusive treatment of multilevel research, one assessing, and analyzing context. This chapter is an
that reviews and melds the three core “silos” of important addition to the growing discussion of the
multilevel research (multilevel theory, multilevel importance of context in organizational research,
measurement, and multilevel analysis) and does so providing a practical and user-friendly guide to
from a perspective that considers micro, meso, and integrating context into theory and empirical
macro frameworks. The purpose of this handbook research.
is therefore to provide guidance for scholars Next, Meyer, England, Kelly, Helbling, Li, and
working in the social and behavioral sciences who Outten (Chapter 3) extend the previous chapter by
wish to consider the implications that multilevel presenting three metatheoretical frameworks for
research (i.e., theory, measurement, and analysis) understanding context. After reviewing the three
may have for their research programs. Although frameworks, the authors review several empirical
the majority of contributors to this handbook have conceptualizations of context, interpreting these
backgrounds in the organizational sciences, the papers through the lenses of the metatheoretical
chapters have been largely written in a manner frameworks. They conclude by providing guidance,
that should be accessible to researchers from derived from this exercise, for conceptualizing
a wide array of research disciplines including context in one’s own work.
(but not limited to) communication, education, Shifting the focus from context to dynamics,
sociology, psychology (clinical, developmental, Cronin and Vancouver (Chapter 4) provide an
industrial, social), management (strategy, in-depth examination of “dynamics” in the context
human resources, organizational behavior), of multilevel theory. Recognizing that behavior is
and nursing. a series of ongoing events and happenings (rather
than a singular, static perspective of a stimulus
and response), Cronin and Vancouver discuss
PART I: MULTILEVEL THEORY
how to theorize in a way that incorporates time
In the first part of the handbook, scholars focus and change. After developing five principles of
on providing guidance on how to improve theory multilevel dynamics, they apply their model to
by integrating a multilevel perspective. Gully and a concrete example. This chapter is extremely
Phillips (Chapter 1) begin this part by positing the helpful for making sense of dynamics in social
most fundamental question for multilevel research: and organizational research. Despite the fact that
What is the appropriate level for your research? scholars intuitively recognize that behavior is
These authors define identifying the appropriate continuous, research has consistently neglected
level as determining which level has the most to theorize or test dynamic models. With this
explanatory power for the particular outcome of chapter, scholars are provided with a clear and
interest; they discuss the challenges scholars face concise set of instructions for creating dynamic
in determining the appropriate level before shifting models.
to a more grounded process of figuring out what In Chapter 5, Aiken, Hanges, and Chen shift
is appropriate to one’s own research. Their chapter to a discussion of complexity science. Similar
serves to orient readers as to how to begin thinking to the surrounding chapters, Aiken et al. are
as a multilevel scholar. interested in emergence—the idea that constructs
In the next chapter, Ostroff (Chapter 2) engages evolve and “emerge” over time. In contrast
readers with a discussion of how to “contextualize to the other chapters, however, they focus on
context.” As she documents, context is omnipresent emergence within the context of complex adaptive
4
Copyright American Psychological Association Introduction
systems. In such systems, emergence is expressed Paruchuri, Goossen, and Phelps (Chapter 9)
in patterns of relationships that are nonlinear expand on this introduction to social network
and nondeterministic. This chapter provides research in by presenting a guide for theorizing
a thoughtful summary of complexity theory, and specifying multilevel social networks. Whereas
particularly as it intersects with multilevel most social network research collapses multilevel
theory, giving readers a fantastic introduction data (e.g., individuals nested in dyads nested in
to how one might better integrate concepts departments) to a single level (e.g., individuals),
from complexity theory into their multilevel Paruchuri and colleagues argue that much may
research. be gained by scholars simultaneously examining
In the next chapter, Ployhart and Hendricks multiple levels. In their chapter, they present a
(Chapter 6) begin a conversation about bottom-up primer on social network research, moving from
theory and methods. Combining the micro and macro single-level to multilevel networks. Throughout
organizational literatures, they propose a framework the chapter, they couple an organizing framework
for conducting microfoundations research—that is, with illustrative examples, grounding the reader
research that connects the individual to organizational- in current science. Finally, they conclude by
or group-level outcomes. Their four-step process providing multiple avenues for future research,
connects theorizing, measurement, and analysis in laying out a broad research agenda for aspiring
this domain. This chapter is invaluable for those scholars.
scholars interested in looking “inside the box”
to understand how lower level antecedents may
PART II: MULTILEVEL MEASUREMENT
emerge to influence higher level (e.g., organizational)
AND DESIGN
outcomes.
Mathieu and Luciano (Chapter 7) integrate In this part of the handbook, we transition from
several themes in their chapter: construct clarifica- focusing largely on issues related to multilevel
tion, emergence, and temporal dynamics. The goal theory, to a discussion of issues related to multilevel
of this integration was to propose an authentic measurement and research design. These chapters
and dynamic way of considering emergence, one are particularly important for those readers who
that focused less on statistical processes and more have specified their theory and are now ready to set
on theoretical mechanisms. Within multilevel about collecting data to test it.
research, understanding emergence is fundamental Zhou, Song, Alterman, Liu, and Wang
because constructs at a higher level are dependent (Chapter 10) present a valuable and detailed
on, connected to, or otherwise affected by lower guide for collecting multilevel data. As they note,
level phenomena, yet without a clear understanding there are unique challenges associated with multi-
of how focal constructs are connected to others, level data collection—beyond what is required for
models are likely misspecified. single-level data collection—and understanding how
The next two chapters focus on social networks to avoid fatal flaws is critical for anyone pursuing
within organizational sciences. First, as Brass a multilevel research agenda. In addition to the
and Borgatti (Chapter 8) note, social networks in-depth guidance, they provide a pocket guide to
are inherently multilevel in nature. They suggest their recommendations, which is likely to be a
that taking a social network approach to research go-to handout for aspiring multilevel scholars.
will open both theoretical and empirical avenues, Jebb, Tay, Ng, and Woo (Chapter 11) offer a
allowing researchers the opportunity to better primer on multilevel construct validation. Construct
specify and analyze relationships. With this as validation is an important component of research,
their focus, they spend the majority of the chapter serving as the bedrock on which theory testing
orienting the reader to the complexities of social rests. In this chapter, Jebb and colleagues walk the
network research, particularly as they pertain to reader through the process of construct validation
multilevel research questions. for multilevel studies, giving readers guidance for
5
Humphrey and LeBreton Copyright American Psychological Association
several construct forms. Finally, they include syntax summaries, it is both logical and necessary for
for replicating their analyses, giving readers the scholars to understand how variance is partitioned
opportunity to easily apply Jebb et al.’s guidance to in multilevel research, as well as have the tools to
their own work. estimate explained variance in their own research.
Chapter 12, by Krasikova and LeBreton, has After discussing several ways to estimate explained
three goals: to provide readers with a summary of variance, the authors walk readers through the
the reliability and agreement indices that are most estimations, providing syntax for running the
commonly used in multilevel research, to present a estimations in their own models.
user-friendly guide to these indices, and to extend In Chapter 16, Grund, Lüdtke, and Robitzsch
the application of these indices to the study of address a common problem in multilevel research:
dyadic phenomena. In an effort to summarize this missing data. Although scholars may be tempted to
topic as succinctly as possible, the authors present simply drop missing data from their analyses (listwise
a clear and interpretable set of guidelines for deletion), there are two clear problems: potential
assessing agreement and reliability. Finally, within reductions in statistical power and the potential
their illustrative example, they present syntax for for biased estimates of population parameters.
replicating and extending their analyses. The authors discuss two solutions to this problem
Beal and Gabriel (Chapter 13) shift the focus to (multiple imputation and maximum likelihood
the unique challenges associated with measurement estimation), followed by an illustration that walks
within the contexts of within-person research the reader through the application of these methods.
designs. Given the benefit to theory stemming from Similar to several other chapters, the authors provide
examining intraindividual development and change, their syntax for implementing these solutions.
understanding how to conduct these types of studies
is imperative. They discuss several types of within-
PART III: MULTILEVEL ANALYSIS
person research designs and highlight the strengths
and weaknesses of each design. Finally, they connect The third part of the book deals with questions of
the within-person methods to the between-person analysis—now that you have a theory and have
level, expanding the number of levels that may collected the data, how do you go about testing your
be studied. hypotheses? The first chapter in this part, written by
Next, Scherbaum and Pesner (Chapter 14) Shiverdecker and LeBreton (Chapter 17), presents
deal with a crucial issue for multilevel research: a simple primer for conducting multilevel (random
determining whether one has sufficient power to coefficients) regression analyses. In an effort to be
adequately test his or her hypotheses. As the authors comprehensive, the authors walk through both
note, although power analysis is rather straight- the mathematical models that underlie multilevel
forward and well documented within single-level regression and the actual steps for running these
research, the issue is appreciably more complex (and models. Using an illustrative example and providing
less well documented) in multilevel research. After syntax, the authors offer even the most novice
reviewing the state of the literature and discussing multilevel researcher the tools for modeling
the central equations, they present several tools multilevel relationships.
for estimating multilevel power analysis. Finally, Next, Knight and Humphrey (Chapter 18)
they walk readers through the use of these tools, discuss how to analyze dyadic data. They begin by
providing a hands-on tutorial or guide to estimating providing a detailed introduction to dyadic data
power in multilevel research. analysis, describing the different levels that exist
LaHuis, Blackmore, and Bryant-Lees (Chapter 15) in dyadic data structures (using both theoretical
examine how to calculate, interpret, and present and empirical terminology). They then provide an
estimates of explained variance in multilevel empirical illustration of social relations modeling
research. Given the movement to better integrate (one specific dyadic data analysis technique),
information about effect sizes into research walking the reader through each step (from
6
Copyright American Psychological Association Introduction
structuring the data set to interpreting the results). In the final chapter of this part of the handbook,
After providing the syntax used in their chapter, Newman and Wang (Chapter 23) provide an alter-
they conclude by discussing alternative software native method for assessing emergence through the
packages for analyzing dyadic data. use of network analysis. They begin by deriving the
In Chapter 19, Vandenberg and Richardson equations for their model, which they use to develop
provide an introductory, “nuts and bolts” guide to a computational model of climate emergence.
multilevel structural equation modeling (MLSEM). They demonstrate the applicability of their model,
After discussing the benefits of MLSEM, the authors providing useful syntax for applying their model to
build a testable model, walking the reader through other contexts.
each step necessary for eventually running an Some appendices for Chapters 11, 12, 16, 17, 18,
MLSEM. They not only present syntax for running 21, 22, and 23 include R codes. These appendices
an MLSEM but also discuss each component of can also be found on the American Psychological
the syntax and the output, giving readers in-depth Association website at http://pubs.apa.org/books/
knowledge of an MLSEM. supp/humphrey/.
Going beyond the in-depth primer from the prior
chapter, Zyphur, Zhang, Preacher, and Bird dedicate PART IV: REFLECTIONS
Chapter 20 to discussing moderated mediation in ON MULTILEVEL RESEARCH
MLSEM. After discussing moderation, mediation,
and moderated mediation, they provide an illustrative The two concluding chapters provide perspective on
example of testing multilevel moderated mediation. the development of multilevel research. In Chapter 24,
In this example, they introduce a “Bayesian ‘plausible Yammarino and Gooty discuss the use of cross-
values’ approach to latent variable interactions” level models within the organizational sciences.
to avoid numerical integration (which ultimately They begin by providing a historical perspective
ensures convergence in their analyses). These on cross-level models, examining the similarities
authors also provide the syntax they used to run and differences in how the concept of “cross-level
the models in their chapter. models” has been applied by various groups of
In Chapter 21, Zachary, Moore, and Ballinger researchers. They then transition to the presentation
focus on how to analyze nonnormal data in of an integrative cross-level model, discussing its
multilevel models. They begin by discussing the applicability to theory building and testing within
challenges associated with analyzing nonnormal the organizational sciences.
data in multilevel analyses. They follow this by The final chapter of the handbook (Chapter 25)
describing the benefits of generalized linear mixed consists of a series of interviews (conducted by
modeling before providing an illustrative example. Michael Hoffman) with five distinguished multilevel
In walking the reader through the example, they scholars: David Chan, Gilad Chen, Fred Dansereau,
provide relevant syntax for replicating their Denise Rousseau, and Benjamin Schneider. In these
analyses. interviews, the scholars reflected on their experiences
Lang and Bliese (Chapter 22) offer a new working on multilevel research, discussing the
statistical framework for detecting a specific form challenges they faced and the seminal work that
of emergence—consensus among lower level units influenced their thinking and their work. They then
nested in higher level collectives—over time. They discuss where multilevel research is going, sharing
develop a three-level emergence model that both their individual visions for the future.
provides a formal test of consensus emergence and
allows researchers to include variables that might References
help to predict consensus. The authors conclude by Chan, D. (1998). Functional relations among constructs
in the same content domain at different levels of
demonstrating their model with two examples and
analysis: A typology of composition models. Journal
again offer syntax so that others may verify and extend of Applied Psychology, 83, 234–246. http://dx.doi.org/
the results from this chapter to their own work. 10.1037/0021-9010.83.2.234
7
Humphrey and LeBreton Copyright American Psychological Association
Humphrey, S. E., & Aime, F. (2014). Team microdynamics: Lepak, D. P., Smith, K. G., & Taylor, M. S. (2007). Value
Toward an organizing approach to teamwork. creation and value capture: A multilevel perspective.
The Academy of Management Annals, 8, 443–503. Academy of Management Review, 32, 180–194.
http://dx.doi.org/10.5465/19416520.2014.904140 Morgeson, F. P., & Hofmann, D. A. (1999). The
Klein, K. J., & Kozlowski, S. W. J. (2000). Multilevel structure and function of collective constructs:
theory, research, and methods in organizations: Implications for multilevel research and theory
Foundations, extensions, and new directions. development. Academy of Management Review,
San Francisco, CA: Jossey-Bass. 24, 249–265.