ArticlePDF Available

Leadership behaviors and follower performance: Deductive and inductive examination of theoretical rationales and underlying mechanisms

Authors:

Abstract and Figures

There are competing theoretical rationales and mechanisms used to explain the relation between leadership behaviors (e.g., consideration, initiating structure, contingent rewards, and transformational leadership) and follower performance (e.g., task performance and organizational citizenship behaviors). We conducted two studies to critically examine and clarify the leadership behaviors–follower performance relation by pitting the various theoretical rationales and mechanisms against each other. We first engaged in deductive (Study 1) and then inductive (Study 2) theorizing and relied upon 35 meta-analyses involving 3327 primary-level studies and 930 349 observations as input for meta-analytic structural equation modeling. Results of our dual deductive–inductive approach revealed an unexpected yet surprisingly consistent explanation for why leadership behaviors affect follower performance. Specifically, leader–member exchange is a mediating mechanism that was empirically determined to be involved in the largest indirect relations between the four major leadership behaviors and follower performance. This result represents a departure from current conceptualizations and points to a common underlying mechanism that parsimoniously explains how leadership behaviors relate to follower performance. Also, results lead to a shift in terms of recommendations for what leaders should focus on to bring about improved follower performance.
Content may be subject to copyright.
Leadership behaviors and follower performance:
Deductive and inductive examination of theoretical
rationales and underlying mechanisms
RYAN K. GOTTFREDSON
1
*AND HERMAN AGUINIS
2
1
Department of Management, Mihaylo College of Business and Economics, California State University, Fullerton, &
Gallup, Inc., California, U.S.A.
2
Department of Management, School of Business, George Washington University, Washington, District of Columbia, U.S.A.
Summary There are competing theoretical rationales and mechanisms used to explain the relation between leadership
behaviors (e.g., consideration, initiating structure, contingent rewards, and transformational leadership) and
follower performance (e.g., task performance and organizational citizenship behaviors). We conducted two
studies to critically examine and clarify the leadership behaviorsfollower performance relation by pitting
the various theoretical rationales and mechanisms against each other. We rst engaged in deductive
(Study 1) and then inductive (Study 2) theorizing and relied upon 35 meta-analyses involving 3327
primary-level studies and 930 349 observations as input for meta-analytic structural equation modeling.
Results of our dual deductiveinductive approach revealed an unexpected yet surprisingly consistent
explanation for why leadership behaviors affect follower performance. Specically, leadermember
exchange is a mediating mechanism that was empirically determined to be involved in the largest indirect
relations between the four major leadership behaviors and follower performance. This result represents a
departure from current conceptualizations and points to a common underlying mechanism that parsimoni-
ously explains how leadership behaviors relate to follower performance. Also, results lead to a shift in
terms of recommendations for what leaders should focus on to bring about improved follower perfor-
mance. Copyright © 2016 John Wiley & Sons, Ltd.
Keywords: consideration; initiating structure; contingent rewards; transformational leadership; leader
member exchange; employee performance
The relation between leadership behaviors and follower performance is one of the oldest and most widely
researched topics in organizational behavior (e.g., Stogdill, 1950; Yukl, 2012). In fact, as of the writing of
our manuscript, there are 19 published meta-analyses on leadership behaviorsfollower performance relations
(e.g., DeRue, Nahrgang, Wellman, & Humphrey, 2011; Dulebohn, Bommer, Liden, Brouer, & Ferris, 2012;
Judge, Piccolo, & Ilies, 2004). Historically, the focus has been on whether leader behaviors enhance follower
performance and the degree to which they do so across different types of leadership behaviors and follower
performance. However, a key question in terms of advancing our understanding of this relation is the following:
Copyright © 2016 John Wiley & Sons, Ltd.
Received 31 July 2015
Revised 1 June 2016, Accepted 31 August 2016
*Correspondence to: Ryan Gottfredson, Department of Management, California State University, Fullerton, and Gallup, Inc. E-mail:
ryangottfredson1@gmail.com
Portions of this manuscript are based on Ryan K. Gottfredsons doctoral dissertation, which was conducted at the Kelley School of Business,
Indiana University, under the supervision of Herman Aguinis. We thank committee members Timothy T. Baldwin, Jeffery S. McMullen, and
Edward R. Hirt for constructive feedback on previous drafts. Additionally, we thank Journal of Organizational Behavior Associate Editor, Mark
Martinko and three anonymous reviewers for their helpful suggestions for improvement. A previous version of this manuscript was presented at
the meetings of the Academy of Management, Philadelphia, PA, August 2014.
A Video Abstract to accompany this article is available at https://youtu.be/FlUQy0dBZG8.
Journal of Organizational Behavior, J. Organiz. Behav. 38, 558591 (2017)
Published online 5 October 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/job.2152
Research Article
Why do positive leadership behaviors improve various types of follower performance? This question is critically
important for theoretical progression in the leadership domain because if we do not understand why these
specic relations occur, we do not have a solid theory (Bacharach, 1989; Dubin, 1978; Sutton & Staw, 1995;
Whetten, 1989). Additionally, if we do not clearly understand why leadership behaviorsfollower performance
relations occur, we will be limited in our ability to provide accurate and actionable recommendations for leaders
that will result in the most favorable performance outcomes.
While the question above is critically important for theoretical and practical reasons, results to date have led
to multiple answers across the various leadership behaviorsfollower performance relations. For example,
consider one of the most well-researched leadership behaviorfollower performance relation: transformational
leadership and task performance. An examination of the literature reveals that there are at least eight empirically
supported mediators explaining this relation, including self-congruence, empowerment, positive effect, trust,
person/job t, core job characteristics, leadermember exchange (LMX), and work engagement (Aryee,
Walumbwa, Zhou, & Hartnell, 2012; Chi & Pan, 2012; Piccolo & Colquitt, 2006; van Knippenberg & Sitkin,
2013; Wang, Law, Hackett, Wang, & Chen, 2005). Similar observations can be made for other often-researched
leadership behaviors. For consideration, initiating structure, contingent rewards, and transformational leadership
(i.e., the four most frequently studied leadership behaviors), the presence of multiple mediating mechanisms
oftentimes more than 10 serving as explanations for their relations with follower performance suggests that
a critical challenge in terms of advancing this domain is not a lack of theory, but the existence of too many
theories. Thus, there is an opportunity to clarify and advance theory in this domain by seeking to identify
one or several mediating mechanisms that may outperform others in explaining specic leadership behavior
follower performance relations.
The purpose of our research is to identify the mechanisms, and their respective theories, that provide the best
explanations for leadership behaviorsfollower performance relations across four different types of leadership
behaviors (consideration, initiating structure, contingent rewards, and transformational leadership) and two types
of follower performance (task performance and organizational citizenship behaviors). To achieve our goal, we
know that a single study or even several primary studies would be inadequate. Accordingly, we used meta-
analytic structural equation modeling (MASEM; Bergh et al., 2016), and relied on a total of 35 meta-analyses
(i.e., 26 already published and an additional nine that we conducted for this study), comprising a total of
3327 primary-level studies and 930 349 observations. MASEM allowed us to move beyond the traditional
meta-analytic approach focusing on bivariate relations and investigate a more complete representation of leader-
ship behaviorsfollower performance phenomena by including as many potential mediating mechanisms in the
leadership behaviorsfollower performance relations as possible (Bergh et al., 2016). Specically, in Study 1,
we adopted a deductive approach and in Study 2 we adopted an inductive approach that was informed by results
from Study 1. Together, these studies allowed us to examine the vast majority of data collected thus far in the
leadership domain to prune and rene the theoretical rationales and mechanisms. Our two-study research
program allowed us to identify a single best explanation for why leadership behaviors lead to follower perfor-
mance across four different leadership behaviors and two types of follower performance. This concise and
parsimonious result across the leadership behaviorsfollower performance relations not only enhances theoretical
precision but also provides leaders with a clearer idea of what they should focus on in order to ensure their
leadership behaviors bring about high levels of follower performance.
While we believe our study enhances both theoretical and practical precision, we recognize that it is a step in
the right direction, but not likely a nalanswer on this topic. Although we identied a single best explanation,
it is based upon a comparison of mediating mechanisms that have been studied frequently enough to be examined
meta-analytically. This means that while theory guided our inclusion of constructs in our models, what we could
actually include in the models was limited to what could be tested meta-analytically. Despite this constraint, our
results help advance our knowledge from the current situation involving multiple and competing theoretical
explanations to the identication of a single theoretical rationale that is the best explanation based on an
examination of the empirical evidence accumulated to date.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 559
Study 1
Theory and hypotheses
There are four types of leadership behaviors that have historically dominated leadership research, as evidenced by
meta-analyses associated with these behaviors. They include consideration, initiating structure, contingent rewards,
and transformational leadership. Because these leadership behaviors are considered to be conceptually distinct and
developed from different theoretical perspectives, researchers have relied upon different theoretical mechanisms
when explaining the relation between each behavior and follower performance. Because the theoretical mediating
mechanisms vary across leadership behaviors, we investigate each leadership behavior separately.
In an effort to prune and rene the theoretical rationales and mechanisms involved in the leadership behaviors
follower performance relations, we rst identied the theories and theoretical mechanisms that are most prevalent.
As noted earlier, there are a few mediating mechanisms and theories that we did not include in our MASEM
because of the lack of sufcient empirical research. While this does leave out potential mediating mechanisms
and respective theories, the lack of empirical research on those mechanisms is an initial indication that perhaps they
are not as theoretically strong or interesting as those mediating mechanisms and theories that we were able to
include in our study. Further, if a particular mediating mechanism or theory not heavily researched and conse-
quently not included in our study is believed to be a valid competing explanation for why leadership behaviors lead
to follower performance, our results are informative because they serve as a benchmark and baseline against which
to compare those alternative explanations in the future.
Consideration and initiating structure
In the 1950s and 1960s a series of studies identied two factors related to leader effectiveness (i.e., Ohio State lead-
ership behaviors): consideration and initiating structure (see Stogdill, 1950). Consideration is the degree to which a
leader emphasizes relationships by showing concern and respect for followers, looking out for their welfare, and
expressing appreciation and support (Burke, Stagl, Klein, Goodwin, Salas, & Halpin, 2006; Judge et al., 2004).
Initiating structure is the degree to which a leader organizes his role and the roles of his followers, is oriented toward
the accomplishment of task objectives, and establishes well dened patterns and channels of communication (Burke
et al., 2006; Judge et al., 2004).
Consideration and initiating structure were the focus of much of the research in the leadership domain from the
time of their creation through the early 1970s. A signicant portion of this research used these two leadership
behaviors to examine path-goal theory (Judge et al., 2004; Wofford & Liska, 1993), which identied followers
satisfaction and motivation as important theoretical mechanisms involved in the relation between leadership and
follower performance (Wofford & Liska, 1993). More specically, it has been common for leadership researchers
to suggest that relationship-oriented leadership (which historically included consideration) creates positive affect,
usually measured in the form of follower job satisfaction and/or satisfaction with the leader, which in turn leads
to increased performance (House, 1971). Additionally, in explaining why task-oriented leadership (which
historically included initiating structure) was related to follower performance, researchers have suggested that
task-oriented leadership increases motivation, commonly measured as a reduction in role ambiguity and role
conict, which in turn leads to increased performance (House, 1971)
Additionally, strong relations between (i) these leadership behaviors and commitment (Luthans, Baack, & Taylor,
1987; Mathieu & Zajac, 1990; Wofford & Liska, 1993) and (ii) commitment and follower performance (Meyer,
Stanley, Herscovitch, & Topolnytsky, 2002) identies commitment and commitment theory as an alternative
explanation for why the two Ohio State leadership behaviors relate to follower performance. This rationale suggests
that consideration and initiating structure should enhance followersdesire to remain members of the organization,
which in turn should lead to improved performance.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
560 R. K. GOTTFREDSON AND H. AGUINIS
In sum, there are ve theoretical mechanisms representing two different theories that have received signicant
theoretical and empirical attention in the explanation of the relation between the two Ohio State leadership behaviors
and follower performance: job satisfaction, satisfaction with leader, role ambiguity, role conict, and commitment.
Although it is likely that each of these theoretical mechanisms plays at least a small role in the pathway from lead-
ership to follower performance, the most prevalent theory (path-goal theory) associated with these leadership behav-
iors suggests that (i) follower job satisfaction and satisfaction with leader should be involved in the largest indirect
relations between consideration and follower performance; and (ii) role ambiguity and role conict should be
involved in the largest indirect relations between initiating structure and follower performance. Thus, we offer the
following hypotheses:
Hypothesis 1: In a model that includes (i) consideration as the antecedent variable; (ii) task performance and OCB
as the outcome variables (i.e., follower performance); and (iii) follower job satisfaction, satisfaction with leader,
role ambiguity, role conict, and commitment as mediators; the largest relations between consideration and
follower performance will be the indirect relations associated with follower job satisfaction and satisfaction with
leader.
Hypothesis 2: In a model that includes (i) initiating structure as the antecedent variable; (ii) task performance and
OCB as outcome variables (i.e., follower performance); and (iii) follower job satisfaction, satisfaction with leader,
role ambiguity, role conict, and commitment as mediators; the largest relations between initiating structure and
follower performance will be the indirect relations associated with role ambiguity and role conict.
Contingent rewards
Contingent rewards involve the use of rewards (e.g., recognition) provided by the leader based upon performance
and represent an exchange of give and takebetween the leader and the follower to establish a fair and trusting
relationship between the two parties and motivate higher performance (Judge & Piccolo, 2004; Podsakoff, Bommer,
Podsakoff, & MacKenzie, 2006). Contingent rewards have been studied primarily in two streams of research: (i)
leader reward and punishment behaviors and (ii) transactional leadership (Podsakoff et al., 2006). Although each
of these two research streams also considers other types of leadership behaviors (e.g., contingent punishment,
noncontingent rewards, noncontingent punishment for leader reward and punishment behaviors, and active and
passive management-by-exception for transactional leadership) the emphasis of the empirical research within
these streams of literature is on contingent rewards, and it is thus emphasized here, similar to other leadership
studies (e.g., DeRue et al., 2011).
The two main mechanisms that have been used to explain the pathway from contingent rewards to follower
performance are justice and motivation (Podsakoff et al., 2006). Justice theories have been relied upon because
contingent rewards emphasize the notion of contingency upon performance, which causes followers to assess the
degree of fairness and equity in the allocation of rewards. Studies that have investigated justice as a mediating
mechanism between contingent rewards and follower performance have most commonly relied upon the constructs
of distributive and procedural justice. Additionally, a variety of motivational theories have also been used, most of
which stem from expectancy theory and are related to the motivational components of instrumentality and valence
(e.g., reinforcement theory). The most common constructs measured to explain the role of motivation in the relation
between contingent rewards and follower performance have been role ambiguity and role conict because they serve
to weaken employeesperceptions of instrumentality.
In addition to fairness and motivation theories, we identied three other theories that have been used to explain
why contingent rewards is related to follower performance: job satisfaction theory, social exchange theory, and
relational leadership theory. These theories include the constructs job satisfaction, trust, affective commitment,
and LMX.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 561
Follower job satisfaction and a general job satisfaction theory have been heavily studied in the leadership litera-
ture, and when examined with contingent rewards, they suggest that the presence of contingent rewards improves the
positive affect of the follower, which then translates into improved performance (Judge & Piccolo, 2004; Judge,
Thoreson, Bono, & Patton, 2001). Conversely, the absence of contingent rewards (not receiving rewards when they
are merited) is likely to cause decreased job satisfaction, translating into lower levels of performance (Judge &
Piccolo, 2004; Podsakoff et al., 2006). In summary, contingent reward behaviors lead to improved performance
under the assumption that a happy employee is a productive employee.
Social exchange theory suggests that when leaders provide social benets, followers are likely to reciprocate with
enhanced performance (Cropanzano & Mitchell, 2005). While this theory has been used less often than the previ-
ously identied theories associated with the contingent reward-follower performance relation, it has been suggested
that proper reward allocation serves to indicate that the leader trusts the follower and to emotionally attach the
employee to the organization which, in turn, increases the followersdesires to exchange the leaders trust and good
will with improved performance (Rubin, Bommer, & Bachrach, 2010). Two commonly studied proxy mechanisms
used in studies adopting a social exchange conceptual framework are trust (i.e., willingness to be vulnerable to
another) and affective commitment (i.e., desire to remain a member of an organization because of a feeling of
emotional attachment) (Konovsky & Pugh, 1994; Rubin et al., 2010; Wayne, Shore, & Liden, 1997). In addition
to initial theoretical and empirical support for social exchange theory in this context, meta-analyses have identied
trust and affective commitment as consequences of contingent rewards (Podsakoff et al., 2006) as well as anteced-
ents of task performance and OCB (Colquitt, Scott, & LePine, 2007; Meyer et al., 2002).
An additional mediating mechanism proposed as an explanation for why contingent rewards is related to follower
performance is LMX. However, there are two challenges associated with viewing LMX as a mediating mechanism:
how should it be interpreted and what theory does it best represent based upon its interpretation? Traditionally, LMX
was developed as a construct to assess the quality of dyadic relationships between a leader and each of the leaders
followers, suggesting that the leader develops relationships of different qualities across followers (e.g., an in-group
and an out-group follower), and these differences in quality relationships matter (Boies & Howell, 2006; Dansereau,
Cashman, & Graen, 1973). LMX was originally meant to be assessed from the perspective of both the leader and the
follower (Gerstner & Day, 1997), and it was originally founded in role theory, but later deemed to be more strongly
related to social exchange theory (Gerstner & Day, 1997).
But, over time, LMX has been measured and interpreted in ways that differ from its traditional denition and
recommended measurement, particularly when assessed along with leadership behaviors (commonly as a conse-
quence of those behaviors) or in meta-analyses. In such instances, researchers are most interested in the followers
perceptions of their relationship with their leader, as opposed to the leadersperceptions of their relationships with
their followers (e.g., Dulebohn et al., 2012; Martin, Guillaume, Thomas, Lee, & Epitropaki, 2016). Thus, in such
research, LMX is interpreted as followersperceptions of the degree to which they have a positive relationship with
their leader. This seems to be an appropriate interpretation considering the items of the measure of LMX that is most
commonly used and included in leadership meta-analyses (LMX-7): my supervisor understands my problems and
needs,”“my supervisor recognizes my potential,”“I usually know where I stand with my supervisor,and how
would you describe your working relationship with your supervisor(Scandura & Graen, 1984).
Given that our research involves leadership behaviors and relies upon meta-analytic data based upon measures of
LMX from the followers perspective, we interpret LMX as the followers perception of the quality of relationship
with their leader. Considering both this interpretation, along with its measures, it is important to note that LMX is
not a leadership behavior (although it has been considered as such in past research); rather, it is the followers
perceptions of the quality of a specic relationship, with measures that differ from those of leadership behaviors
(perceptions of a relationship versus perceptions of behaviors). We address this and related measurement challenges
in more detail in the Discussion section.
The aforementioned interpretation, although somewhat related to social exchange theory, is actually most closely
related to relational leadership theory (Uhl-Bien, 2006). Relational leadership theory focuses on the relationship
between two parties and reects the inuence process that occurs through the strength of a relationship. This theory
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
562 R. K. GOTTFREDSON AND H. AGUINIS
ts well with the explanation for why contingent rewards is related to follower performance from an LMX perspec-
tive. Namely, the administration of rewards serves as an indication of a quality relationship (e.g., Only those with
good relationships with their leaders get rewards), and a better relationship between follower and leader will pro-
vide a social and emotional structure allowing the follower to focus and engage in high levels of performance. In
other words, followers perceive rewards to be an indication of the quality of leaderfollower relationship, and this
relationship is critical to follower performance for reasons that likely include having the condence and support
of the leader and feelings of safety.
Although we have identied eight prevalent mechanisms involved in the relation between contingent rewards and
follower performance, spanning ve theories, the most recent meta-analysis on contingent rewards relied solely on
fairness and motivation theories to explain its relation with follower performance, suggesting that distributive jus-
tice, procedural justice, role ambiguity, and role conict should be involved in the largest indirect relations between
contingent rewards and follower performance (Podsakoff et al., 2006). In sum, we offer the following hypotheses:
Hypothesis 3: In a model that includes (i) contingent rewards as the antecedent variable; (ii) task performance and
OCB as outcome variables (i.e., follower performance); and (iii) distributive justice, procedural justice, role
ambiguity, role conict, follower job satisfaction, trust, affective commitment, and LMX as mediators; the largest
relations between contingent rewards and follower performance will be the indirect relations associated with
distributive justice, procedural justice, role ambiguity, and role conict.
Transformational leadership
Transformational leadership is a construct used to describe how leaders inuence and inspire followers to make self-
sacrices, commit to difcult objectives, and perform beyond previous levels (Piccolo, Bono, Heinitz, Rowold,
Duehr, & Judge, 2012). Currently, it seems to be the most dominant leadership construct in organizational behavior
research (Yukl, 2012).
Because transformational leadership is expected to lead followers to perform beyond previous levels, researchers
have posited that transformational leadership should have a strong relation with OCBs, which are commonly viewed
as extra-role behaviors. Theoretically, the OCB literature has relied heavily upon social exchange theory
(Cropanzano & Mitchell, 2005). In turn, this has led transformational leadership researchers to also rely heavily
upon social exchange theory. The theoretical rationale is that behaviors associated with transformational leadership
should serve to (i) indicate to the follower that the leader is trustworthy and (ii) emotionally attach the employee to
the organization and, in turn, increase the followersdesires to exchange the leaders trustworthiness and good will
with improved performance (Konovsky & Pugh, 1994; Podsakoff, MacKenzie, Moorman, & Fetter, 1990). Once
again, the most commonly used mechanisms for social exchange are trust and affective commitment.
Because LMX has been associated with social exchange theory in the past, LMX has also been proposed and
tested as a mediator in the relation between transformational leadership and follower performance (Dulebohn
et al., 2012). But, we present it here as a mechanism better represented by relational leadership theory, which sug-
gests that transformational leadership behaviors are relational inuence tactics that enhance followersperceptions
of the quality of relationship with their leader. In turn, these relationship quality perceptions likely create a sense
of support and safety that allow the follower to focus on the tasks at hand, the success of those around them, and
excel in terms of performance.
Another mediating mechanism involved in the study of the pathway from transformational leadership to follower
performance is job satisfaction. The theoretical rationale is similar to the rationale discussed previously regarding
contingent rewards, in that the behaviors associated with transformational leadership (e.g., idealized inuence or
charisma, inspirational motivation, intellectual stimulation, and individualized consideration) are likely to engender
positive outcomes for the followers, leading to higher levels of satisfaction (Judge & Piccolo, 2004). In turn, job sat-
isfaction has been found to lead to enhanced task performance and OCB (Judge et al., 2001; Podsakoff, MacKenzie,
Paine, & Bachrach, 2000).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 563
Although job satisfaction and LMX are studied frequently in the transformational leadership literature, the
primary theory relied upon in the relation between transformational leadership and follower performance is social
exchange theory (Pillai, Schriesheim, & Williams, 1999). For that reason, we posit that trust and affective commit-
ment will be involved in the largest indirect relations between transformational leadership and follower performance.
In short, we offer the following hypotheses:
Hypothesis 4: In a model that includes (i) transformational leadership as the antecedent variable; (ii) task perfor-
mance and OCB as outcome variables (i.e., follower performance); and (iii) trust, affective commitment, LMX,
and follower job satisfaction as mediators; the largest relations between transformational leadership and the
two follower performance constructs will be the indirect relations associated with trust and affective commitment.
In sum, for each of the four major leadership behaviors, we identied mediating mechanisms and theories that
have been used to explain why each leadership behavior is related to follower performance. While they do not
include all possible mechanisms and theories that have been proposed or empirically studied, they have all been
given enough empirical attention to warrant meta-analytic assessments. This suggests that the mechanisms included
in our hypotheses are both theoretically and empirically important.
Method
Overview
We used MASEM, which consists of combining meta-analysis and structural equation modeling. As described in
detail by Bergh et al. (2016), MASEM is an ideally suited method to prune and rene theory because it (i) allows
us to integrate different theoretical mechanisms into more comprehensive models, and (ii) has a number of statistical
advantages that help increase the accuracy of and condence in the estimation of the relations under consideration.
These advantages include maximizing external validity by including all the available data for a particular relation,
maximizing internal validity by allowing for the testing of competing models with different underlying causal struc-
tures, and having a statistical power advantage over primary-level studies because of the large sample sizes associ-
ated with meta-analyses (Bergh et al., 2016; Cheung & Chan, 2005). Additionally, MASEM allows us to offer a
more complete representation of the leadership behaviorsfollower performance phenomena by including and
simultaneously considering many potentially important constructs and relations based upon meta-analytic research.
We implemented all procedures following best-practice recommendations as described by Bergh et al. (2016) and
Landis (2013). As the rst step, we created meta-analytically derived correlation matrices involving all variables in
the models. As will be further described, these matrices include 86 unique meta-analytically derived correlations
based on a total of 35 meta-analyses: 26 published meta-analyses, seven new meta-analyses that we conducted
for Study 1, and two new meta-analyses we conducted for Study 2. These meta-analytically derived correlations
were based on a total of 3327 primary-l evel studies and 930 349 observations. Then, we used structural equation
modeling to test each of our hypotheses which pits underlying mechanisms (i.e., mediating effects) against each
other. Specically, in each model, we formally compared the largest direct or indirect relation with all of the other
relations to determine if that largest relation is statistically larger than all other relations in each of the models.
Existing meta-analyses
We rst conducted a search of meta-analyses associated with each leadership construct, each mediator, and each type
of follower performance (i.e., task performance and OCB). This involved searching ABI Inform, Google Scholar,
PsychINFO, and Web of Science for combinations of the word metaand each of the leadership construct titles, each
of the plausible mediator titles, and search terms associated with task performance (i.e., task performanceand job
performance) and organizational citizenship behaviors (i.e., organizational citizenship behavior,and organiza-
tional citizenship behaviors). In the case of OCBs versus other similar constructs such as contextual behavior,
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
564 R. K. GOTTFREDSON AND H. AGUINIS
contextual performance, and extra-role behaviors, we found that each of these constructs is most frequently utilized in
different areas of research, and each area denes them slightly differently. Because we are examining leadership, we
wanted to be consistent with the domain, which primarily focuses on OCBs (compared to contextual performance,
contextual behavior, and extra-role behaviors). This search resulted in 38 meta-analyses that reported at least one
correlation between any of the constructs involved in our hypotheses (the list of these 38 meta-analyses is available
online as supporting information). Of these 38 meta-analyses, 17 were from meta-analyses on leadership behaviors
and the other 21 primarily focused on the plausible mediators or follower performance constructs.
In some cases, we identied more than one correlation from different meta-analyses representing a single relation.
To determine which of the meta-analytically derived correlations to include in the meta-analytic correlation matrices,
we relied upon multiple criteria as follows (Bergh et al., 2016): (i) correlations that involved constructs whose
operationalizations are consistent with a priori denitions; (ii) correlations that used appropriate pre-specied
meta-analytic techniques (e.g., measurement error corrections given that our interest is in construct-level relations;
Le, Schmidt, & Putka, 2009); and (iii) correlations reported by meta-analyses based on the largest sample size. Thus,
we eventually used meta-analytically derived correlations from 26 of the 38 initially identied meta-analyses for our
correlation matrices (we refer to each of these 26 meta-analyses in tables provided later in our manuscript).
The 26 meta-analyses allowed us to identify most but not all of the correlations required to assess our hypotheses.
Specically, for the consideration models, we identied 26 of the 28 needed correlations (all except follower job
satisfaction and satisfaction with leader, and satisfaction with leader and OCB); for the initiating structure models,
we identied 25 of the 28 needed correlations (all except follower job satisfaction and satisfaction with leader,
satisfaction with leader and OCB, and initiating structure and OCB); for the contingent rewards models, we identi-
ed 51 of the 55 needed correlations (all except role ambiguity and trust, role conict and trust, role ambiguity and
procedural justice, and role conict and procedural justice); and for the transformational leadership models, we
identied all of the 21 needed correlations. Together, this meant that our study required that we conduct seven
additional and original meta-analyses.
Original meta-analyses
We followed best-practice recommendations as offered by Aguinis, Pierce, Bosco, Dalton, and Dalton (2011),
Aytug, Rothstein, Zhou, and Kern (2012), and Kepes, McDaniel, Brannick, and Banks (2013). For each meta-
analysis, we relied upon ve different databasesABI Inform, Google Scholar, ProQuest Dissertations and Theses,
PsychINFO, and Web of Scienceto identify primary-level studies. Initial exclusion criteria were non-English
articles; clinical, prisoner, or church studies; and samples that involved children. Initial inclusion criteria included,
rst, identifying published articles, book chapters, dissertations, accessible conference abstracts, company reports,
and unpublished studies that included a correlation matrix that reported a correlation between the constructs
involved in the meta-analysis. For some of the meta-analyses, constructs were labeled differently than its most
recognized label (e.g., role clarity, which is a reverse-coded measure of role ambiguity), and in such cases we
initially included the article. Second, studies that met this previous criterion were examined to determine the source
of the measured constructs. For plausible mediators (e.g., initiating structure, job satisfaction, satisfaction with
leader, trust, procedural justice), it was necessary that the construct be measured from the perspective of the subor-
dinate. For OCB, we allowed the construct to be measured from any perspective (e.g., self, peer, manager). Occa-
sionally, a primary study reported relations between multiple dimensions of one construct and another construct
(e.g., initiating structure and OCB-individual as well as initiating structure and OCB-organization). In such
instances, we followed Hunter and Schmidts (2004) recommendation and created composite constructs to ensure
that only independent effect sizes were included in each meta-analysis. Because our interest was in construct-level
relations, we implemented the Hunter and Schmidt (2004) psychometric meta-analytic approach. Also, we
conducted publication bias (Egger, Davey-Smith, Schneider, & Minder, 1997) and outlier (Aguinis, Gottfredson,
& Joo, 2013) analyses. Only the meta-analysis addressing the relation between trust and role ambiguity resulted
in a signicant Egger et al. (1997) test value. We used the trim and ll method to assess the effect that such potential
bias might have on the meta-analytic correlation (Dalton, Aguinis, Dalton, Bosco, & Pierce, 2012). Because the
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 565
difference between the two corrected correlations was negligible (.456 compared to .450), we used the original
meta-analytic corrected correlation (.456). Outlier analysis based on the sample-adjusted meta-analytic deviancy
(SAMD) statistic showed no differences in results with and without outliers identied (Aguinis et al., 2013).
Accordingly, we did not omit any outliers from the analyses.
Meta-analytic structural equation modeling
Using our meta-analytic correlation matrices, MASEM involved testing a series of nested models pertaining to each
of the four leadership behaviors. First, we assessed a baseline model involving the leadership construct as the focal
antecedent (exogenous variable), mediators (endogenous variables), and task performance and OCB as the outcome
variables (endogenous variables). But, when engaging in structural equation modeling, it is necessary to identify
alternative plausible models (Vandenberg & Grelle, 2009; Williams, Vandenberg, & Edwards, 2009). Thus, we also
tested additional models for each of the leadership behaviors based on previous research. Specically, for consider-
ation and initiating structure we tested three additional alternative models: (i) commitment as a consequence of the
satisfaction and role constructs (for rationale, see Gaertner, 1999; Mathieu & Zajac, 1990; Tett & Meyer, 1993); (ii)
satisfaction constructs as a consequence of the role constructs (for rationale see Fried, Shirom, Gilboa, & Cooper,
2008); and (iii) covariation of disturbance terms of similar constructs (follower job satisfaction and satisfaction with
leadership, role ambiguity and role conict, and task performance and OCB) to account for unmeasured common
causes. For contingent rewards, we assessed the following four alternative models: (i) trust and affective commit-
ment as consequences of the other plausible mediators (for rationale see Cohen-Charash & Spector, 2001; Colquitt
et al., 2013; Dirks & Ferrin, 2002; Meyer et al., 2002); (ii) follower job satisfaction as a consequence of the justice
constructs, the role constructs, and LMX (for rationale see Cohen-Charash & Spector, 2001; Colquitt et al., 2013;
Fried et al., 2008); (iii) the justice and role constructs as consequences of LMX (for rationale see Colquitt et al.,
2013; Dulebohn et al., 2012); and (iv) covariation of disturbance terms of similar constructs (procedural justice
and distributive justice, role ambiguity and role conict, and task performance and OCB) to account for unmeasured
common causes. Finally, for transformational leadership, we assessed the following three alternative models: (i) trust
and affective commitment as consequences of follower job satisfaction and LMX (for rationale see Dirks & Ferrin,
2002; Dulebohn et al., 2012; Meyer et al., 2002); (ii) follower job satisfaction as a consequence of LMX (for ratio-
nale see Dulebohn et al., 2012); and (iii) covariation of disturbance terms of similar constructs (task performance and
OCB) to account for unmeasured common causes.
While some primary-level studies may not include the same ordering of constructs as in these alternative models,
we relied upon the orderings that are most commonly theorized and studied in the leadership eld and/or are what
are suggested in published meta-analyses, indicating common consensus. Thus, the ordering of relations in our
models are those that are most commonly agreed upon.
First, our alternative model specication procedure allowed us to identify one best tting model for each of the
four leadership behaviors. Second, we used the best tting models to test each of our hypotheses by rst identifying
the size of each of the direct and indirect relations in the models. Third, we used the post-hoc comparison procedure
outlined by MacKinnon (2000) and Preacher and Hayes (2008), to empirically compare the relative size of direct and
indirect relations. Specically, we compared the largest direct or indirect relation between the leadership construct
and task performance and OCB with all other direct and indirect relations involved in that particular leadership
behaviorfollower performance relation. If the largest relation was a hypothesized indirect relation, then the post-
hoc comparisons served as a rigorous and formal test of the hypothesis. But, if the largest relation was not a hypoth-
esized indirect relation, this immediately suggested that the hypothesis was not supported. Even if the largest relation
was not a hypothesized relation, we were still interested in determining if it was larger than all other relations,
providing us the opportunity to test whether a non-hypothesized mechanism outperformed a hypothesized mecha-
nism in explaining the relation between the leadership behavior and follower performance. Thus, we still engaged
in post-hoc comparisons in an effort to accomplish our goal of pruning and rening the theoretical rationales and
mechanisms involved in the relation between leadership behaviors and follower performance. The results of these
comparisons led us to engage in some inductive analyses, which we conducted in our follow-up Study 2.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
566 R. K. GOTTFREDSON AND H. AGUINIS
Results and discussion
A summary of results based on the seven original meta-analyses is included in Table 1 (the list of studies used for each
of the meta-analysis is available online as supporting information). Combining the 26 existing meta-analyses with the
seven new meta-analyses allowed us to complete each of the four meta-analytic correlation matrices needed to assess
our hypotheses. These matrices (see Tables 25), containing a total of 86 unique meta-analytically derived correla-
tions, served as input for MASEM, and can also be used by other researchers to replicate and/or extend our analyses.
We conducted MASEM using AMOS 21. Using the harmonic mean for assessing the signicance of path coef-
cients, the Ns ranged from 4816 to 5757. The best-tting models associated with each leadership construct are
shown in Figure 1. For ease of presentation, this gure only shows the largest relations in each model but 12
supplementary tables including detailed results regarding all post-hoc comparisons (i.e., coefcients, differences
between coefcients, standard errors for comparisons between coefcients, and condence intervals for differences
between coefcients) are available online as supporting information.
Consideration
There are two mechanisms tied for being the largest in the relation between consideration and task performance:
considerationjob satisfactiontask performance (.09) and considerationsatisfaction with leadertask performance
(.09). These results provide initial support for Hypothesis 1 regarding task performance. For the relation between
consideration and OCB, the largest relation was the direct relation (.34), with much smaller total indirect relations
(.02). These results did not provide support for Hypothesis 1 regarding the OCB outcome.
After identifying the largest relations, direct or indirect, we conducted post-hoc comparisons with all other direct
or indirect relations in the model (i.e., a total of 17). Results indicated that the considerationjob satisfactiontask
performance relation, which was .09, was not signicantly larger than any of the other direct and indirect relations
associated with task performance (relations ranged from .10 to .09). In other words, there is no indirect mechanism
in the relation between consideration and task performance that is statistically larger and more important than any of
the other relations, suggesting no theoretical mechanism is any more important than the others.
Because the largest relation between consideration and OCB was the direct relation (.34), we compared it to all of
the indirect relations (ranging from .07 to .02). Results indicated that the direct relation was signicantly larger
Table 1. Summary of results for original meta-analyses (relations 17 were used in study 1 and relations 89 were used in study 2).
80% CV 95% CI
Relation kN rr
c
SDrcQLower Upper Lower Upper
1. IS and OCB 12 2105 .24 .31 .07 71.46* .04 .44 .14 .34
2. JS and SL 179 75 114 .43 .50 .02 1288.49* .30 .55 .41 .44
3. SL and OCB 36 8415 .19 .23 .02 105.64* .08 .30 .15 .23
4. RA and T 34 10 843 .37 .46 .03 275.93* .54 .21 .42 .33
5. RC and T 17 5892 .28 .34 .02 63.19* .39 .18 .33 .24
6. RA and PJ 18 4281 .36 .41 .02 94.79* .51 .21 .42 .30
7. RC and PJ 8 4357 .34 .41 .02 72.45* .48 .20 .42 .26
8. C and LMX 23 6209 .64 .74 .09 1271.86* .31 .98 .54 .75
9. IS and LMX 22 5973 .56 .66 .06 636.81* .28 .84 .47 .65
Note:k= number of samples involved in primary-level studies included in the meta-analysis; N=total number of observations within samples;
r= uncorrected population correlation; r
c
= population correlation corrected for unreliability and range restriction; SDrc= percentage of variance
in r
c
explained by study artifacts; Q= chi-square test for homogeneity of effect sizes; 80% CV = 80% credibility interval based around r; 95%
CI = 95% condence interval based around r; C = consideration; IS = initiating structure; JS = follower job satisfaction; LMX = leadermember
exchange; OCB = organizational citizenship behaviors; PJ = procedural justice; RA = role ambiguity; RC = role conict; SL = satisfaction with
leader; T = trust.
*p<.05.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 567
Table 2. Meta-analytic correlation matrix for consideration (study 1).
Variables 1 2 3 4 5 6 7 8
1 Consideration
2 Role ambiguity (r,r
c
).30,.44
e
k,N25, 2854
CI ??:??
3 Role conict (r,r
c
).28,.42
e
.27, .42
e
k,N9, 1709 47, 10 217
CI ??:?? ??:??
4 Follower job satisfaction (r,r
c
) .40, .46
f
.36,.46
b
.34,.42
b
k,N76, 11 374 52, 11 287 54, 11 851
CI ??:?? ??:?? ??:??
5 Satisfaction with leader (r,r
c
) .68, .78
f
.36,.53
e
.36,.53
e
.43, .50
k,N49, 7871 17, 3619 14, 3440 179, 75 114
CI ??:?? ??:?? ??:?? .41:.44
6 General commitment (r,r
c
) .24, .34
i
.24,.32
c
.18,.23
c
.59, .70
m
.30, .41
i
k,N12, 2642 9, 7040 9, 7040 68, 35 282 23, 5236
CI ??:?? ??:?? ??:?? .57:.61 ??:??
7 Task performance (r,r
c
) .21, .13
o
.15,.21
n
.06,.07
n
.18, .30
g
.16, .19
d
.13, .18
l
k,N21, 3808 74, 11 698 54, 9910 312, 54 471 21, 3630 65, 15 511
CI ??:?? ??:?? ??:?? .27:.33 ??:?? .01:.35
8 Organizational citizenship behaviors (r,r
c
) .26, .32
j
.13,.15
a
.12,16
a
.20, .24
h
.19, .23 .17, .20
h
.39, .47
k
k,N8, 3062 24, 6458 22, 6257 22, 7100 36, 8415 17, 5133 24, 7947
CI .28:.36 .17:.08 .17:.08 .22:.26 .15:.23 .17:.24 .37:.40
Note.k= number of studies; N= combined sample size; r= mean sample-size weighted observed correlations; r
c
= mean sample-size-weighted corrected correlation; CI = 95% con-
dence interval; ?? = gure not reported in original meta-analysis.
Harmonic mean sample size = 5700.
Sources for correlations from previous meta-analyses:
a
= Eatough, Chang, Miloslavic, and Johnson (2011);
b
= Fried et al. (2008);
c
= Gaertner (1999);
d
= Iaffaldano and Muchinsky (1985);
e
= Jackson and Schuler (1985);
f
= Judge et al. (2004);
g
=Judge et al. (2001);
h
= LePine, Erez, and Johnson (2002);
i
= Mathieu and Zajac (1990);
j
= Organ and Ryan (1995);
k
= Podsakoff, Whiting, Podsakoff, and Blume (2009);
l
= Riketta (2002);
m
= Tett and Meyer (1993);
n
= Tubre and Collins (2000);
o
= Wofford and Liska (1993).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
568 R. K. GOTTFREDSON AND H. AGUINIS
Table 3. Meta-analytic correlation matrix for initiating structure (study 1).
Variables 1 2 3 4 5 6 7 8
1 Initiating structure
2 Role ambiguity (r,r
c
).28,.43
e
k,N31, 3705
CI ??:??
3 Role conict (r,r
c
).17,.27
e
.27, .42
e
k,N10, 1839 47, 10 217
CI ??:?? ??:??
4 Follower job satisfaction (r,r
c
) .19, .22
f
.36,.46
b
.34,.42
b
k,N72, 10 317 52, 11 287 54, 11 851
CI ??:?? ??:?? ??:??
5 Satisfaction with leader (r,r
c
) .27, .33
f
.36,.53
e
.36,.53
e
.43, .50
k,N49, 8070 17, 3619 14, 3440 179, 75114
CI ??:?? ??:?? ??:?? .41:.44
6 General commitment (r,r
c
) .21, .29
i
.24,.32
c
.18,.23
c
.59, .70
l
.30, .41
i
k,N14, 3019 9, 7040 9, 7040 68, 35 282 23, 5236
CI ??:?? ??:?? ??:?? .57:.61 ??:??
7 Task performance (r,r
c
) .15, .11
n
.15,.21
m
.06,.07
m
.28, .30
g
.16, .19
d
.13, .18
k
k,N25, 4219 74, 11 698 54, 9910 312, 54 471 21, 3630 65, 15 511
CI ??:?? ??:?? ??:?? .27:.33 ??:?? .01:.35
8 Organizational citizenship behaviors (r,r
c
) .24, .31 .13,.15
a
.12,16
a
.20, .24
h
.19, .23 .17, .20
h
.39, .47
j
k,N12, 2105 24, 6458 22, 6257 22, 7100 36, 8415 17, 5133 24, 7947
CI .14:.34 .17:.08 .17:.08 .22:.26 .15:.23 .17:.24 .37:.40
Note.k= number of studies; N= combined sample size; r= mean sample-size weighted observed correlations; r
c
= mean sample-size-weighted corrected correlation; CI = 95% con-
dence interval; ?? = gure not reported in original meta-analysis.
Harmonic mean sample size = 5747.
Sources for correlations from previous meta-analyses:
a
= Eatough et al. (2011);
b
= Fried et al. (2008);
c
= Gaertner (1999);
d
= Iaffaldano and Muchinsky (1985);
e
= Jackson and Schuler (1985);
f
= Judge et al. (2004);
g
=Judge et al. (2001);
h
= LePine et al. (2002);
i
= Mathieu and Zajac (1990);
j
= Podsakoff et al. (2009);
k
= Riketta (2002);
l
= Tett and Meyer (1993);
m
= Tubre and Collins (2000);
n
= Wofford and Liska (1993).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 569
Table 4. Meta-analytic correlation matrix for contingent rewards (study 1).
Variables 1 2 3 4 5 6 7 8 9 10 11
1 Contingent rewards
2 LMX (r,r
c
) .65, .73
e
k,N6, 1900
CI .58:.88
3 Role ambiguity (r,r
c
).36,
.42
n
.34,
.42
e
k,N25, 7940 18, 5813
CI .45:
.39
.47:
.36
4 Role conict (r,r
c
).26,
.30
o
.27,
.33
e
.27, .42
j
k,N12, 4881 14, 5480 47,
10 217
CI ??:?? .38:
.27
??:??
5 Distributive justice (r,r
c
) .42, .50
n
.38, .44
e
.14,
.18
h
.19,
24
h
k,N6, 1856 32, 6693 9, 7040 9, 7040
CI .47:.53 .36:.50 ??:?? ??:??
6 Procedural justice (r,r
c
) .48, .56
n
.48, .55
e
.36,
.41
.34,
.41
.51, .61
c
k,N6, 1856 30, 7211 18, 4281 8, 4357 184,
67 956
CI .49:.63 .48:.61 .42:
.30
.42:
.26
.49:.54
7 Follower job satisfaction (r,r
c
) .44, .52
n
.42, .49
e
.36,
.46
g
.34,
.42
g
.39, .47
a
.40,
.43
a
k,N43,
11 461
88,
22 520
52,
11 287
54,
11 851
23,
26 277
36,
29 028
CI .47:.57 .45:.52 ??:?? ??:?? .46:.48 .42:.44
8 Trust (r,r
c
) .59, .67
n
.69, .77
d
.37,
.46
.28,
.34
.50, .58
d
.61,
.68
d
.51, .65
d
k,N12, 4192 8, 1183 34,
10 843
17, 5892 15, 3562 30,
5972
34,
10 631
CI .59:.75 .66:.71 .42:
.33
.33:
.24
.48:.52 .59:.62 .50:.52
9 Affective commitment (r,r
c
) .39, .46
n
.36, .41
e
??,
.39
m
??,
.30
m
.37, .47
a
.43,
.50
a
??, .65
m
??,
.54
b
k,N3, 1297 21, 8118 12, 3774 9, 3225 27,
30 257
52,
27 437
69,
23 656
32,
7066
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
570 R. K. GOTTFREDSON AND H. AGUINIS
CI .39:.53 .34:.48 ??:?? ??:?? .46:.48 .49.51 ??:?? ??:??
10 Task performance (r,r
c
) .26, .28
n
.30, .34
e
.15,
.21
q
.06,
.07
q
.19, .26
c
.19,
.24
c
.18, .30
k
.26,
.33
b
??,
.16
m
k,N17, 6180 108,
25 322
74,
11 698
54, 9910 45,
11 336
57,
14 258
312,
54 471
27,
4882
25,
5938
CI .25:.31 .30:.37 ??:?? ??:?? .14:.24 .15:.23 .27:.33 .21:.31 ??:??
11 Organizational citizenship
behaviors (r,r
c
)
.19, .21
n
.32, .37
i
.13,
.15
f
.12,
16
f
.17, .21
c
.23,
.30
c
.20, .24
l
.22,
.27
b
??,
.32
m
.39,
.47
p
k,N3, 554 50, 9324 24, 6458 22, 6257 36,
10 100
71,
16 864
22, 7100 19,
4050
22,
6277
24,
7947
CI .06:.36 .33:.41 .17:
.08
.17:
.08
.14:.20 .20:.26 .22:.26 .19:.25 ??:?? .37:.40
Note.k= number of studies; N= combined sample size; r= mean sample-size weighted observed correlations; r
c
= mean sample-size-weighted corrected correlation; CI = 95% con-
dence interval; ?? = gure not reported in original meta-analysis.
Harmonic mean sample size = 4816.
Sources for correlations from previous meta-analyses:
a
= Cohen-Charash and Spector (2001);
b
= Colquitt et al. (2007);
c
= Colquitt et al. (2013);
d
= Dirks and Ferrin (2002);
e
= Dulebohn et al. (2012);
f
= Eatough et al. (2011);
g
= Fried et al. (2008);
h
= Gaertner (1999);
i
= Ilies, Nahrgang, and Morgeson (2007);
j
= Jackson and Schuler (1985);
k
=Judge et al. (2001);
l
= LePine et al. (2002);
m
= Meyer et al. (2002);
n
= Podsakoff et al. (2006);
o
= Podsakoff, MacKenzie, and Bommer (1996);
p
= Podsakoff et al. (2009);
q
= Tubre and Collins (2000).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 571
Table 5. Meta-analytic correlation matrix for transformational leadership (study 1).
Variables 1 2 3 4 5 6 7
1 Transformational leadership N/A
2 Leadermember exchange (r,r
c
) .66, .73
c
N/A
k,N20, 5451 N/A
CI .64:.81 N/A
3 Follower job satisfaction (r,r
c
) ??, .58
e
.42, .49
c
N/A
k,N18, 5279 88, 22 520 N/A
CI ??:?? .45:.52 N/A
4 Trust (r,r
c
) .72, .79
b
.69, .77
b
.51, .65
b
N/A
k,N13, 5657 8, 1183 34, 10 631 N/A
CI .71:.73 .66:.71 .50:.52 N/A
5 Affective commitment (r,r
c
) ??, .46
h
.36, .41
c
??, .65
h
??, .54
a
N/A
k,N4, 2361 21, 8118 69, 23 656 32, 7066 N/A
CI ??:?? .34:.48 ??:?? ??:?? N/A
6 Task performance (r,r
c
) .19, .21
j
.30, .34
c
.18, .30
f
.26, .33
a
??, .16
h
N/A
k,N31, 7016 108, 25 322 312, 54 471 27, 4882 25, 5938 N/A
CI .16:.26 .30:.37 .27:.33 .21:31 ??:?? N/A
7 Organizational citizenship behaviors (r,r
c
) .26, .30
j
.32, .37
d
.20, .24
g
.22, .27
a
??, .32
h
.39, .47
i
N/A
k,N28, 7970 50, 9324 22, 7100 19, 4050 22, 6277 24, 7947 N/A
CI .26:.34 .33:.41 .22:.26 .19:.25 ??:?? .37:.40 N/A
Note.k= number of studies; N= combined sample size; r= mean sample-size weighted observed correlations; r
c
= mean sample-size-weighted corrected correlation; CI = 95% con-
dence interval; ?? = gure not reported in original meta-analysis.
Harmonic mean sample size = 5610.
Sources for correlations from previous meta-analyses:
a
= Colquitt et al. (2007);
b
= Dirks and Ferrin (2002);
c
= Dulebohn et al. (2012);
d
= Ilies et al. (2007);
e
= Judge and Piccolo (2004);
f
= Judge et al. (2001);
g
= LePine et al. (2002);
h
= Meyer et al. (2002);
i
= Podsakoff et al. (2009);
j
= Wang et al. (2011).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
572 R. K. GOTTFREDSON AND H. AGUINIS
than all of the indirect relations associated with OCB. This result suggests that either there is no primary theoretical
mechanism that explains the relation between consideration and OCB or there is such a mechanism but it is not cur-
rently prevalent in the consideration literature, and thus not included in this model. Taken together, these results do
not provide support for Hypothesis 1.
Initiating structure
For the initiating structuretask performance relation, the largest mechanism was the indirect relation through role
ambiguity and follower job satisfaction (.05). For the relation between initiating structure and OCB, the largest mech-
anism is the direct relation (.28), with very little total indirect relations (.03), which is similar to the consideration model.
Similar to the analyses associated with consideration, there were 17 comparisons involving the relations between
initiating structure and the two types of follower performance, respectively. Results were also similar. We compared
Figure 1. Meta-analytic structural equation modeling results for underlying mechanisms linking leadership and follower perfor-
mance for consideration, initiating structure, contingent rewards, and transformational leadership (Study 1).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 573
the largest relation between initiating structure and task performance (initiating structurerole ambiguityjob
satisfactiontask performance; .05) to all other initiating structuretask performance relations (ranging from .04
to .04), and results indicated that it was not signicantly larger than any of the other relations. In other words, there
is no superior underlying mechanism in the relation between initiating structure and task performance out of all of
the mechanisms included in the model. We compared the largest relation between initiating structure and OCB (di-
rect relation; .28) to all indirect relations associated with OCB (ranging from .05 to .03) and it was signicantly
larger than all of those relations. These results suggest that either there is no primary theoretical mechanism that ex-
plains the relation between initiating structure and OCB or there is such a mechanism but it is not currently prevalent
in the initiating structure literature, and thus not included in this model. Taken together, these results do not provide
support for Hypothesis 2.
Contingent rewards
The largest mechanism for the contingent rewardtask performance and contingent rewardsOCB relations is the
indirect relations through LMX (.17 and .38, respectively), which was not one of the hypothesized mechanisms.
However, this is an intriguing result because LMX is a social mechanism, and although social mechanisms are
acknowledged to be important in the contingent reward literature, they are not the most heavily emphasized. We
compared the indirect relations of contingent rewardLMXtask performance and contingent performanceLMX
OCB relations to the other 59 direct and indirect relations associated with each type of follower performance, respec-
tively. For all of these comparisons, these relations were signicantly larger than all of the other relations associated
with the given type of follower performance (ranging from .22 to .08 across both models). These results conrmed
that LMX is indeed involved in the largest relation between contingent rewards and follower performance, and thus
it is the primary mediating mechanism in those relations. Not only does this result not support Hypothesis 3, but it
also suggests that the primary mechanism used to explain the relation between contingent rewards and follower
performance is different than what has been most heavily relied upon in the contingent rewards literature.
Accordingly, this result offers initial indication that the theory explaining the contingent rewardfollower performance
relations may need to be renedtoincludeandemphasizeLMX.
Transformational leadership
The largest mechanism for the transformational leadershiptask performance and transformational leadershipOCB
relations is also the indirect relations through LMX (.22 and .27, respectively). We compared the two largest mech-
anisms, both involving the indirect relation through LMX, to the other 11 direct and indirect relations associated with
each type of follower performance, respectively. For the comparisons associated with task performance, the largest
mechanism (.22) was larger than all other indirect relations (ranging from .26 to .12), but it was not signicantly
larger than the direct relation (.26). In short, LMX is the most primary mediating mechanism in the relation between
transformational leadership and task performance. For the comparisons associated with OCB, the largest relation
(.27) was found to be signicantly larger than all other direct and indirect relations (ranging from .08 to .09). This
suggests that LMX is the most primary mediating mechanism in the relation between transformational leadership and
OCB. Finally, for both of these sets of post-hoc analyses, indirect relations with trust and affective commitment were
small. Thus, overall, there was no support for Hypothesis 4.
Summary
Study 1 was an effort to prune and rene the theoretical rationale and mechanisms involved in the relations between
four leadership behaviors and two types of follower performance. In doing so, we relied upon the strongest theoret-
ical rationale associated with the four different leadership behaviors to identify and hypothesize which of the various
mediating mechanisms should be involved in the largest relation between the leadership behaviors and follower per-
formance. This top-down deductive theorizing and analysis provided results that, for the most part, did not conrm
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
574 R. K. GOTTFREDSON AND H. AGUINIS
the theoretical rationale most predominantly relied upon for each of the four leadership domains. Specically, we
found that (i) there is no dominant mediating mechanism explaining the pathway from consideration and initiating
structure to follower performance, and (ii) LMX is the dominant mediating mechanism explaining the pathway from
contingent rewards and transformational leadership to follower performance.
Across these results, we noticed a novel, consistent, and yet counterintuitive pattern, which is that when LMX is
included in the models, it is the largest and most predictive pathway (i.e., largest coefcient) explaining the leadership
behaviorfollower performance relation. Thus, there is initial empirically-based evidence that LMX may be a
common underlying mechanism that explains why leadership is related to follower performance across leadership
behaviors, perhaps a meta-theoretical principle. A meta-theoretical principle explains a variety of similar relations
across a domain, addressing phenomena at a higher level than specic theories (e.g., Pierce & Aguinis, 2013;
Richter, 1986).
In Study 1, we were not able to test whether LMX is the primary mediating mechanism in the consideration and
initiating structure models because LMX has not been theoretically or empirically emphasized in the leadership
literature as a consequence of those leadership behaviors. Accordingly, we conducted follow-up Study 2, in which
we meta-analyzed the relation between the two Ohio State leadership behaviors with LMX in order to inductively
test whether LMX is the primary mediating mechanism in the relations involving consideration and initiating
structure with follower performance. This effort is warranted because if LMX is identied as the most important
theoretical mechanism in the models associated with consideration and initiating structure, we would identify a
simple and parsimonious, yet counterintuitive and admittedly provocative rationale that explains the relation
between leadership behaviors and follower performance across a variety of leadership behaviors. This rationale
would be: leadership behaviors lead to follower performance because they increase followersperceptions of having
a good relationship with their leader.
Study 2
Theory and hypotheses
The theorizing process we have engaged in thus far is deductive in nature and the standard approach in organizational
behavior research (Aguinis & Vandenberg, 2014). It is a top-down approach that involves an identication of hypoth-
eses based upon theory, which is then followed by hypothesis testing. Inductive research, on the other hand, is a
bottom-up approach that begins with nding meaningfulness, tension, conict, or contradiction, which in turn leads
to hypothesis creation and testing based upon the data (Aguinis & Vandenberg, 2014; Shepherd & Sutcliffe, 2011).
Although underutilized, inductive theorizing and testing can lead to and is often necessary for theoretical progress
(Locke, 2007). Moreover, inductive approaches can improve our ability to prune and rene theories (Aguinis &
Vandenberg, 2014). Study 2 adopts an inductive approach in furthering a prediction derived from Study 1 that LMX
may be a common underlying mechanism explaining the relation between leadership and follower performance across
leadership behaviors. This effort is considered an inductive approach because relations between the Ohio State leader-
ship behaviors and LMX have not been theorized in the leadership literature; however, studies do exist that simulta-
neously examined these constructs (largely in the form of unpublished dissertations) making meta-analyses possible.
Falling in line with existing research between leadership behaviors and LMX (e.g., Dulebohn et al.,2012), it is rea-
sonable to believe that LMX mediates the relation between the two Ohio State leadership behaviors and follower per-
formance, but each for different reasons. Specically, when a leader engages in consideration by showing concern
and respect for the followers, followers are likely to perceive that they are important to and valued by the leader, en-
hancing their perceptions of a strong leaderfollower relationship. Additionally, when leaders engage in initiating
structure by organizing the roles of their followers, followers are likely to perceive that their leader wants them to
be successful, or even that only a leader who values the success of a follower would provide initiating structure, which
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 575
should also enhance perceptions of a strong leaderfollower relationship. In turn, these strong relationships should
provide a social and emotional context allowing followers to excel in terms of their performance. Thus, we provide
a conceptual representation (Shepherd & Sutcliffe, 2011) of these inductive relations by hypothesizing the following:
Hypothesis 5: In a model that includes (i) consideration as the antecedent variable; (ii) task performance and OCB
as outcome variables (i.e., follower performance); and (iii) LMX, follower job satisfaction, satisfaction with
leader, role ambiguity, role conict, and commitment as mediators; the largest relations between consideration
and the two follower performance constructs will be the indirect relations associated with LMX.
Hypothesis 6: In a model that includes (i) initiating structure as the antecedent variable; (ii) task performance and
OCB as outcome variables (i.e., follower performance); and (iii) LMX, follower job satisfaction, satisfaction with
leader, role ambiguity, role conict, and commitment as mediators; the largest relations between initiating struc-
ture and the two follower performance constructs will be the indirect relations associated with LMX.
Method
Original meta-analyses
We conducted two new meta-analyses of the relation between consideration and initiating structure and LMX. We
followed the same procedures as in the original meta-analyses in Study 1. For the meta-analysis between consid-
eration and LMX, we identied primary-level studies reporting a correlation between consideration and LMX by
searching the same ve databases for considerationand leadermembertogether in the same searches. These
searches resulted in 23 studies, each with only one unique sample (the list of sources is available online as
supporting information). For the meta-analysis between initiating structure and LMX, we used the same proce-
dures and the search term initiating structure.The searches resulted in 22 studies (the list of sources is available
online as supporting information). Most of these studies have been unpublished (e.g., dissertations); used consid-
eration and initiating structure as proxies for situational leadership (see Thompson & Vecchio, 2009); or studied
consideration, initiating structure, and LMX as separate leadership constructs. As such, they reported correlations
between the constructs, but did not specically theorize relations between them. Results summarizing these two
new meta-analyses are included in Table 1.
MASEM
As an outcome of the two new meta-analyses, we expanded the meta-analytically derived correlation matrices for
consideration and initiating structure to include their relations with LMX, thus allowing us to assess whether
LMX, role ambiguity, role conict, follower job satisfaction, satisfaction with leader, or general commitment is
the best explanation for the consideration/initiating structure-follower performance relations. These matrices are
included in Tables 6, 7 and the harmonic mean for the meta-analytically derived correlations was 6230 and 6267,
respectively. Study 2 involved a total of 44 unique meta-analytically derived correlations, which were based on
17 published and ve original meta-analyses (from either Study 1 or Study 2) and represented a total of 1969 studies
and 479 294 observations.
Similar to Study 1, we conducted MASEM which involved testing alternative models as follows. For consider-
ation, we tested a baseline model consisting of direct relations between consideration and each of the plausible me-
diators (follower job satisfaction, satisfaction with leader, role ambiguity, role conict, commitment, and LMX), and
direct relations between each of the plausible mediators and each of the follower performance constructs. In
addition, we tested the following four alternative models: (i) commitment as a consequence of the two satisfaction
constructs, the two role constructs, and LMX; (ii) satisfaction constructs as consequences of the role constructs and
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
576 R. K. GOTTFREDSON AND H. AGUINIS
Table 6. Meta-analytic correlation matrix for consideration (study 2).
Variable 1 2 3 4 5 6 7 8 9
1 Consideration
2 Leadermember exchange (r,r
c
) .64, .74
k,N23, 6209
CI .54:.75
3 Role ambiguity (r,r
c
).30,.44
g
.34,.42
a
k,N25, 2854 18, 5813
CI ??:?? .47:.37
4 Role conict (r,r
c
).28,.42
g
.27,.33
a
.27, .42
g
k,N9, 1709 14, 5480 47, 10 217
CI ??:?? .38:.27 ??:??
5 Follower job satisfaction (r,r
c
) .40, .46
h
.42, .49
a
.36,.46
c
.34,.42
c
k,N76, 11 374 88, 22 520 52, 11 287 54, 11 851
CI ??:?? .30:.37 ??:?? ??:??
6 Satisfaction with Leader (r,r
c
) .68, .78
h
.57, .68
a
.36,.53
g
.36,.53
g
.43, .50
k,N49, 7871 32, 11 195 17, 3619 14, 3440 179, 75 114
CI ??:?? .57:.76 ??:?? ??:?? .41:.44
7 General commitment (r,r
c
) .24, .34
k
.41, .47
a
.24,.32
d
.18,.23
d
.59, .70
o
.30, .41
k
k,N12, 2642 58, 14 208 9, 7040 9, 7040 68, 35 282 23, 5236
CI ??:?? .34:.48 ??:?? ??:?? .57:.61 ??:??
8 Task performance (r,r
c
) .21, .13
q
.30, .34
a
.15,.21
p
.06,.07
p
??, .30
i
.16, .19
e
.13, .18
n
k,N21, 3808 108, 25 322 74, 11 698 54, 9910 312, 54 471 21, 3630 65, 15 511
CI ??:?? .14:.24 ??:?? ??:?? .27:.33 ??:?? .01:.35
9 Organizational citizenship
behaviors (r,r
c
)
.26, .32
l
.32, .37
f
??,.15
b
??,16
b
.20, .24
j
.19, .23 .17, .20
j
.39, .47
m
k,N8, 3062 50, 9324 24, 6458 22, 6257 22, 7100 36, 8415 17, 5133 24, 7947
CI .14:.34 .14:.20 .17:.08 .17:.08 .22:.26 .15:.23 .17:.24 .37:.40
Note.k= number of studies; N= combined sample size; r= mean sample-size weighted observed correlations; r
c
= mean sample-size-weighted corrected correlation; CI = 95% con-
dence interval; ?? = gure not reported in original meta-analysis.
Harmonic mean sample size = 6230.
Sources for correlations from previous meta-analyses:
a
= Dulebohn et al. (2012);
b
= Eatough et al. (2011);
c
= Fried et al. (2008);
d
= Gaertner (1999);
e
= Iaffaldano and Muchinsky (1985);
f
= Ilies et al. (2007);
g
= Jackson and Schuler (1985);
h
= Judge et al. (2004);
i
=Judge et al. (2001);
j
= LePine et al. (2002);
k
= Mathieu and Zajac (1990);
l
= Organ and Ryan (1995);
m
= Podsakoff et al. (2009);
n
= Riketta (2002);
o
= Tett and Meyer (1993);
p
= Tubre and Collins (2000);
q
= Wofford and Liska (1993).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 577
Table 7. Meta-analytic correlation matrix for initiating structure (study 2).
Variable 1 2 3 4 5 6 7 8 9
1 Initiating structure
2 Leadermember exchange (r,r
c
) .56, .66
k,N22, 5973
CI .47:.65
3 Role ambiguity (r,r
c
).28,.43
g
.34,.42
a
k,N31, 3705 18, 5813
CI ??:?? .47:.37
4 Role conict (r,r
c
).17,.27
g
.27,.33
a
.27, .42
g
k,N10, 1839 14, 5480 47, 10 217
CI ??:?? .38:.27 ??:??
5 Follower job satisfaction (r,r
c
) .19, .22
h
.42, .49
a
.36,.46
c
.34,.42
c
k,N72, 10 317 88, 22 520 52, 11 287 54, 11 851
CI ??:?? .30:.37 ??:?? ??:??
6 Satisfaction with Leader (r,r
c
) .27, .33
h
.57, .68
a
.36,.53
g
.36,.53
g
.43, .50
k,N49, 8070 32, 11 195 17, 3619 14, 3440 179, 75 114
CI ??:?? .57:.76 ??:?? ??:?? .41:.44
7 General commitment (r,r
c
) .21, .29
k
.41, .47
a
.24,.32
d
.18,.23
d
.59, .70
o
.30, .41
k
k,N14, 3019 58, 14 208 9, 7040 9, 7040 68, 35 282 23, 5236
CI ??:?? .34:.48 ??:?? ??:?? .57:.61 ??:??
8 Task performance (r,r
c
) .15, .11
p
.30, .34
a
.15,.21
p
.06,.07
p
??, .30
i
.16, .19
e
.13, .18
n
k,N25, 4219 108, 25 322 74, 11698 54, 9910 312, 54 471 21, 3630 65, 15 511
CI ??:?? .14:.24 ??:?? ??:?? .27:.33 ??:?? .01:.35
9 Organizational citizenship behaviors (r,r
c
) .24, .31 .32, .37
f
??,.15
b
??,16
b
.20, .24
j
.19, .23 .17, .20
j
.39, .47
m
k,N12, 2105 50, 9324 24, 6458 22, 6257 22, 7100 36, 8415 17, 5133 24, 7947
CI .14:.34 .14:.20 .17:.08 .17:.08 .22:.26 .15:.23 .17:.24 .37:.40
Note.k= number of studies; N= combined sample size; r= mean sample-size weighted observed correlations; r
c
= mean sample-size-weighted corrected correlation; CI = 95% con-
dence interval; ?? = gure not reported in original meta-analysis.
Harmonic mean sample size = 6267.
Sources for correlations from previous meta-analyses:
a
= Dulebohn et al. (2012);
b
= Eatough et al. (2011);
c
= Fried et al. (2008);
d
= Gaertner (1999);
e
= Iaffaldano and Muchinsky (1985);
f
= Ilies et al. (2007);
g
= Jackson and Schuler (1985);
h
= Judge et al. (2004);
i
=Judge et al. (2001);
j
= LePine et al. (2002);
k
= Mathieu and Zajac (1990);
l
= Podsakoff et al. (2009);
m
= Riketta (2002);
n
= Tett and Meyer (1993);
o
= Tubre and Collins (2000);
p
= Wofford and Liska (1993).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
578 R. K. GOTTFREDSON AND H. AGUINIS
LMX; (iii) role constructs as consequences of LMX; and (iv) covariation of disturbance terms of similar constructs
(follower job satisfaction and satisfaction with leadership, role ambiguity and role conict, and task performance and
OCB) to account for unmeasured common causes. For initiating structure, we identied alternative models that
reected the consideration models above. The best tting models are included in Figure 2.
Figure 2. Meta-analytic structural equation modeling results for underlying mechanisms linking leadership and follower perfor-
mance for consideration and initiating structure (Study 2).
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 579
Results and discussion
Consideration
Results based on our inductive theorizing, which include LMX in the model, differed from those based on the de-
ductive approach in Study 1. Regarding task performance, a comparison of results in Figure 1 to those in Figure 2
shows that while the total relation between consideration and task performance stayed the same, the direct relation
became more negative (.10 to .36), and the total indirect relations became more positive (.23 to .49), with the
largest indirect relation involving LMX (.37). Additionally, when focusing on the relation associated with OCB,
the direct relation decreased from .34 to .18 and the total indirect relations increased from .02 to .14, with the larg-
est relation involving the indirect relation through LMX (.23).
Next, we compared the largest mechanisms (i.e., those going through LMX) with all other relations in the model.
For the comparisons between considerationLMXtask performance and all other direct and indirect relations, the
considerationLMXtask performance relation (.37) is signicantly larger than all other relations (ranging from
.36 to .07) except for the absolute value of the direct relation (.36), which is negative. Thus, LMX is the most
important mediating mechanism in the relation between consideration and task performance.
All comparisons between considerationLMXOCB (.23) and all other direct and indirect relations (ranging from
.09 to .18) were statistically signicant except for the direct relation (.18). Thus, these results suggest that in the
relation between consideration and OCB, LMX is also the largest mediating mechanism. Additionally, either the di-
rect relation is also important or there may be other mechanisms not included in the model that might help explain
this relation. In sum, these results support Hypothesis 5.
Initiating structure
Results regarding the inductively derived model involving initiating structure are displayed graphically in Figure 2.
Compared to results from Study 1, and regarding task performance as the outcome, the direct relation changed from
being small and positive to being larger and negative (.04 to .25), and the indirect relations became much stronger
(.07 to .36), with the largest relation being the indirect relation involving LMX (.37). Additionally, when focusing on
the relation associated with OCB, the direct relation decreased from .28 to .15 and the total indirect relations in-
creased from .02 to .16, with the largest relation being the indirect relation involving LMX (.17). Together, these
results are very similar to the results associated with the inductively derived consideration model.
Similar to previous models, we conducted comparisons involving the largest relations versus all others. All of the
comparisons involving task performance revealed that the initiating structureLMXtask performance relation (.37)
was signicantly larger than all other direct or indirect relations in the model (ranging from .11 to .07). This sug-
gests that LMX is indeed the largest mediating mechanism in the relation between initiating structure and task per-
formance. For the comparisons involving OCB, the initiating structureLMXOCB relation (.17) was signicantly
larger than all other relations (ranging from .04 to .14) except the direct relationship (.14). Thus, these results also
suggest that in the relation between initiating structure and OCB, LMX is the primary mediating mechanism. Addi-
tionally, our results indicate that the direct relation is also important. According to our model, this suggests that the
initiating structureOCB relation is partially mediated through LMX, but it may also suggest that there may be other
mechanisms not included in the model that might help explain this relation. Taken together, these results support
Hypothesis 6.
Summary
Results from Study 2 offered additional support that LMX is a superior explanation for the underlying mechanism
linking leadership behavior and follower performance. Although the role of LMX is not stated explicitly as a medi-
ator in theories associated with the relations involving consideration and initiating structure with follower
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
580 R. K. GOTTFREDSON AND H. AGUINIS
performance, we derived hypotheses inductively based on results from Study 1, and they were supported. Thus,
results suggest that the strongest pathway from all four leadership behaviors to followersperformance is through
LMX.
Two anonymous reviewers observed that there are high meta-analytically derived correlations between multiple
mediating mechanisms in our models (e.g., LMX, satisfaction with leader, trust). High correlations between
constructs suggest potential issues with discriminant validity (Joseph, Newman, & Sin, 2011), and raise concerns
regarding whether one construct, out of several that are interrelated, can be identied as being the primary
explanation between leadership behaviors and follower performance. While high correlations between constructs
may raise concerns when considering just their bivariate relations, these concerns are less relevant when consid-
ering models involving multiple constructs and various different relations. A particular strength of MASEM is
that it takes into consideration all relations in the model simultaneously, allowing us to assess models involving
complex relations between variables, even if the models include interrelated constructs (Bergh et al., 2016). As
indicated, LMX was still identied as being involved in the largest relations between leadership behaviors and
follower performance out of all of the mediating mechanisms in the models based on our post-hoc analyses
(detailed tables regarding these results, which we described earlier in summary form, are available online as
supporting information).
General Discussion
The purpose of our two-study research program was to bring clarity and precision to the theoretical rationales and
mechanisms involved in the relations between four leadership behaviors (consideration, initiating structure, contin-
gent rewards, and transformational leadership) and two types of follower performance (task performance and
OCBs). We sought to more clearly answer the question: Why do positive leadership behaviors improve various types
of follower performance? We engaged in a theory pruning and rening effort, using MASEM involving 35 meta-
analyses synthesizing a total of 3327 studies and 930 349 observations, which allowed us to combine multiple
theoretical mechanisms to empirically determine which were involved in the strongest indirect relations in the
leadership behaviorsfollower performance relations.
Our pruning and rening effort started with a deductive approach to hypothesis testing, where we hypothesized
that certain theoretical mechanisms, those that have been most strongly emphasized in the literature respective to
each major leadership behavior, would be involved in the largest indirect relations between the given leadership be-
havior and follower performance. To our surprise, our deductive analyses provided little support for the hypotheses,
suggesting that perhaps the primary theoretical rationales and mechanisms most strongly emphasized in the different
leadership domains require revisions. In fact, our results suggested that perhaps LMX is a common explanatory me-
diating mechanism across the various leadership behaviors. This nding led us to engage in a follow-up inductive
approach, in which we assessed whether LMX was the best explanatory mediating mechanism in the relations be-
tween consideration and initiating structure and follower performance. Results of these inductive analyses further
supported the deductive results, nding that LMX is a simple and parsimonious rationale that explains the relation
between leadership behaviors and follower performance, suggesting that relational leadership theory is perhaps the
best theoretical explanation, out of many currently in use, for why leadership behaviors lead to follower
performance.
To summarize our results from Study 1 and Study 2, Table 8 includes (i) a list of the most emphasized and fre-
quently studied mechanisms used to explain the relations between each leadership behavior and follower perfor-
mance (i.e., plausible mediating mechanisms); (ii) the few mechanisms associated with each leadership behavior
that have been suggested and/or studied as the best or most accurate explanations for why leadership behavior is re-
lated to follower performance prior to our study (i.e., our hypothesized mediating mechanisms); and (iii) our results
through our deductive (Study 1) and inductive (Study 2) analyses. Results are rather consistent on two accounts.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 581
First, the mediating mechanisms widely considered to be the best and most accurate explanations for why the lead-
ership behavior is related to follower performance were not empirically supported through a pruning and rening
process. Second, LMX, and along with it, relational leadership theory, was empirically determined to be the best ex-
planation for why leadership behaviors are related to follower performance across four of the most heavily studied
leadership behaviors. Our results suggest that perhaps we have discovered a meta-theoretical principle, explaining a
common phenomenon across various leadership domains.
Implications for theory
Our results indicating that LMX is the mediating mechanism involved in the largest indirect relations between
leadership behaviors and follower performance across the four most heavily studied leadership behaviors were
rather unexpected, but surprisingly consistent. Because results were not anticipated, our ndings have caused
us to reect upon why LMX plays such a critical role in the leadership behaviorsfollower performance relation.
One rationale for our consistent ndings is rooted in the idea that a positive leaderfollower relationship creates
a psychologically safe environment by which followers can (i) focus on the task at hand, as opposed to issues
related to an unsafe environment, resulting in higher levels of task performance; and (ii) divert their attention
away from themselves to their work group or organization as a whole, resulting in higher levels of
organizational citizenship behaviors (Kahn, 1990). A different, yet related rationale is rooted in relational lead-
ership theory, which states that the stronger the leaderfollower relationship, the greater the inuence the leader
possesses (Uhl-Bien, 2006). This ability to inuence provides the leader with the power to direct the follower to
greater levels of task performance and organizational citizenship behaviors (Aguinis, Nesler, Quigley, Lee, &
Tedeschi, 1996).
The meta-theoretical rationale uncovered by our studies meets the major criteria for what constitutes good theory
by Bacharach (1989): it is parsimonious, clear, and falsiable; it has utility; and has a broad scope. Specically, it is
parsimonious because it is a simple statement or relation that applies to a wide variety of leadership domains. It is
clear because it is easily understandable. It is falsiable in that empirical research can be conducted to prove it to be
Table 8. Summary of hypotheses and results from study 1 (deductive approach) and study 2 (inductive approach).
Focal leadership behavior leading
to task performance and
organizational citizenship
behaviors
Plausible
mediating
mechanisms
Hypothesized
mediating
mechanisms
(associated theory)
Study 1/deductive
analysis results
(associated theory)
Study 2/inductive
analyses results
(associated theory)
Consideration RA, RC, JS, SL,
and Comm
(inductive: add
LMX)
JS and SL (Path-goal
theory)
No primary
theoretical
mechanism
LMX (Relational
leadership theory)
Initiating structure RA, RC, JS, SL,
and Comm
(inductive: add
LMX)
RA and RC (Path-
goal theory)
No primary
theoretical
mechanism
LMX (Relational
leadership theory)
Contingent rewards LMX, RA, RC,
DJ, PJ, JS, T, and
AC
RA, RC, DJ, and PJ
(Motivation and
fairness theories)
LMX (Relational
leadership theory)
Transformational leadership LMX, JS, T, and
AC
T and AC (Social
exchange theory)
LMX (Relational
leadership theory)
AC = affective commitment; C = consideration; Comm = commitment; DJ = distributive justice; JS = follower job satisfaction; LMX = leader
member exchange; PJ = procedural justice; RA = role ambiguity; RC = role conict; SL = satisfaction with leader; T = trust.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
582 R. K. GOTTFREDSON AND H. AGUINIS
wrong (or a weaker mechanism compared to otherssomething we have done in our analyses given the available
data to date). It has utility in that it can be used by practitioners (as we discuss in the section on Implications for
Practice). Also, it has a broad scope in that it can be applied across several leadership constructs as well as types
of follower performance. In retrospect, our nding that a good leaderfollower relationship is the strongest pathway
from leadership behaviors to follower performance seems logical, simple, and perhaps even commonsensical. How-
ever, this conclusion seems obvious only nowafter we collected and analyzed a large dataset based on 35 meta-
analyses involving 3327 primary-level studies and 930 349 observations.
Our results also suggest a shift in theory across the four major leadership behavior domains. For initiating
structure and contingent rewards, not only have these two domains rarely relied upon LMX as an important
mediating mechanism, but there have been few researchers, if any, who believe that a social factor is the best
explanation for why these leadership behaviors are related to follower performance. Specically, consider the
relations between contingent rewards and follower performance. These relations have been theoretically rooted
in issues of fairness, equity, exchange, and motivation (Podsakoff et al., 2006), yet our results suggest that
contingent rewards operate as a currency indicative of the quality of leaderfollower relationship. Thus, for
initiating structure and contingent rewards, our results seem to suggest a shift in theory away from fairness
and motivation theories toward relationship theories. Additionally, while the consideration and transformational
leadership literatures have identied social factors as being important in their relations with follower
performance, neither has predominantly relied upon LMX or relational leadership theory. For example, Wang,
Oh, Courtright, and Colberts (2011) meta-analysis relied upon motivation and self-efcacy to explain why
transformational leadership leads to follower performance, which have little connection with LMX and relation-
ship theories.
Identifying LMX as a meta-theoretical principle may warrant a shift in how leadership researchers view the LMX
and relational leadership literature. Its absence from recently published reviews on leadership suggests that re-
searchers have been losing interest in LMX (Dinh, Lord, Gardner, Meuser, Liden, & Hu, 2014; Lord & Dinh,
2014; Yukl, 2012). Yet, our results suggest that such decreased interest should not be the case. Rather, our results
suggest that LMX plays a critical role in the success of leadership behaviors, and thus, should warrant increased at-
tention and interest.
While our results do suggest a shift from a broad understanding of why leadership behaviors lead to follower
performance, spanning various mechanisms and theories, to something more concise, we emphasize that our
results do not suggest that mechanisms and theories other than LMX and relational leadership theory are not
important in leadership behaviorsfollower performance relations. It would be inappropriate to negate mediating
mechanisms that have received empirical support in primary-level studies. For example, through primary-level
studies, trust has been identied as a mediating mechanism between both contingent rewards and transforma-
tional leadership and follower performance (Podsakoff et al., 1990; Rubin et al., 2010). So, a nding that
LMX is the strongest intervening mechanism between leadership behaviors and follower performance does
not suggest that leaders should not be concerned about trust, or any other previously identied consequences
of leadership behaviors (e.g., fairness, satisfaction, commitment). Rather, LMX may be an antecedent of these
other important consequences of leadership behaviors. For example, it is hard to imagine a situation where
there is high LMX and low trust. In fact, Martin et al.s (2016) MASEM found that trust is the best explana-
tion for the relation between LMX and follower performance, although they did not compare the strength of the
direct relation with the strength of the indirect relations. In light of this, we conducted additional analyses in-
cluding LMX as the primary antecedent; distributive justice, procedural justice, job satisfaction, trust, and affec-
tive commitment as plausible mediating mechanisms; and task performance and OCBs as the focal
consequences. Results of these analyses revealed that the direct relations between LMX and both types of fol-
lower performance were signicantly stronger than any of the indirect relations. So, while we do not want to
negate other obviously important outcomes of leadership, we do believe that our ndings clearly identify LMX
as the best explanation among all of those we considered in our study, which has important practical implica-
tions, as described next.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 583
Implications for practice
Until now, practical implications of leadership behavior research are generally: (i) to the degree that a manager en-
gages in leadership behaviors, he or she will see improvement in follower performance; and (ii) leaders need to be
concerned about a variety of follower perceptions, including satisfaction, fairness, motivation, commitment, and
trust, because they all underlie the relation between leadership behaviors and follower performance. Our results clar-
ify what leaders need to do and focus on in order to bring out improved follower performance. If leaders want to
serve as catalysts for high levels of follower performance, our results suggest that they need to focus on one partic-
ular follower perception: LMX, or the improve follower performance, namely: their followersperception of their
relationship with them. In short, our results simplify what leaders need to focus on in an effort to improve follower
performance, namely the followersperceptions of their relationship with their leader.
Additionally, while our results indicate that leader behaviors enhance the leaderfollower relationship, our results
may also lead to suppositions about the effectiveness of individual leadership behaviors (e.g., providing intellectual
stimulation, rewarding based upon performance). More specically, our results seem to suggest that the effective-
ness of any given leadership behavior is likely to be inuenced by the followersperceptions of their relationship
with their leader, such that followers with good relationships with their leader will respond more positively in terms
of performance to a given leadership behavior, compared to followers with poor relationships with their leader. To-
gether, this logic suggests that leaders should engage in positive leadership behaviors to improve their followers
relationship with them, which, as these perceptions improve, will increase the effectiveness of single leadership be-
haviors in terms of performance outcomes. Stated differently, to become a leader that guides followers to high levels
of performance, one must engage in leadership behaviors to develop positive relationships with followers, and the
degree to which this is done, the more effective leadership behaviors become at enhancing performance.
The aforementioned practical implications are further supported by considering the actual difference (i.e., effect
size) that the improvement of LMX can have on objective measures of performance across the four different lead-
ership models. Because all coefcients reported in our manuscript are based on SD metric, we needed three types of
information to calculate the practical impact of improving a leaders LMX score (Aguinis, Werner, Abbott, Angert,
Park, & Kohlhausen, 2010): (i) standard deviation for LMX; (ii) standard deviation for objective measures of per-
formance that have been studied as a consequence of LMX; and (iii) the coefcients between LMX and task perfor-
mance across our four models. The average standard deviation for LMX across ve different studies that used LMX-
7 is .85 (Graen, Novak, & Sommerkamp, 1982). A study by Siders, George, and Dharwadkar (2001) examined task
performance of sales representatives in ve different ways in the medical sales industry: sales volume (in millions),
annual sales growth (%), sales volume from new accounts (%), market share (%), and number of new products sold
in a given year. The standard deviations for each of these performance metrics were .22, 14.13, 3.77, 7.21, and 9.05,
respectively. Using the coefcient between LMX and task performance in each of our models (i.e., consideration:
.497, initiating structure: .557, contingent rewards: .238, and transformational leadership: .294), we were able to cal-
culate the consequence that raising the level of LMX by one standard deviation (.85) has on each of the ve different
types of objective task performance by multiplying the path coefcients with the standard deviation of each objec-
tive measure of task performance. Results are included in Table 9.
Table 9 shows that an increase in one standard deviation of LMX results in, on average and per sales repre-
sentative, an $85 000 increase in sales volume, 5.60 percent increase in annual sales growth, 1.50 percent increase
in sales volume from new accounts, 2.86 percent increase in market share, and 3.59 more new products sold, all
in a given calendar year. Clearly, these numbers represent substantial increases in actual performance per sales
representative per year. Additionally, consider the practical implications if we examine the effect across multiple
sales representatives (i.e., followers)not just one. That would represent meaningful increases not only for the
leader of those sales representatives, but also for the organization as a whole, all in a space of just one year. Thus,
our results suggest that there are tangible and meaningful implications when managers engage in leadership
behaviors in such a way that those behaviors increase the followersperceptions of the leaderfollower
relationship.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
584 R. K. GOTTFREDSON AND H. AGUINIS
In addition to practical implications regarding how leaders behave, our results also suggest that changes should
be made to how leaders are trained. Much of the training of managers as well as MBA students stemming from
leadership research places emphasis on engaging in various behaviors associated with each of the four leadership
behaviors. For example, common advice is for leaders to manage rewards in a contingent as opposed to non-
contingent manner (e.g., Aguinis, Joo, & Gottfredson, 2013), assuming that it is the nature of how rewards are
administered that affects follower performance. Our results suggest that leadership behaviors are effective primar-
ily because they improve the leaderfollower relationship. Thus, future executive, as well as business student
training, should focus more heavily on this underlying explanatory mechanism: LMX, or the followersrelation-
ship with their leader.
Limitations and future research
First, our results must be interpreted within the context of a methodological challenge regarding the measures used
to assess the mediators included in our study. Specically, measures of follower perceptions of the leader that serve
as mediators in our models often include items that may also measure leadership behaviors, which serve as anteced-
ents in our models. For example, an item from the most commonly used measure of LMX, LMX-7 (Scandura &
Graen, 1984), resembles a measure of a leaders behavior (e.g., My supervisor recognizes my potential). Similarly,
items from Liden and Maslyns (1998) LMX scale include My manager would defend me to others in the organi-
zation if I made an honest mistake,”“My manager would come to my defense if I were attackedby others,and
My manager defends (would defend) my work actions to a superior, even without complete knowledge of the issue
in question.The same potential confound involving measurement of LMX and its antecedents (i.e., leadership be-
haviors) holds true for other mediators included in our models. For example, the measure of trust by McAllister
(1995) includes the following item: This person approaches his/her job with professionalism and dedication.
The measure of role ambiguity by Rizzo, House, and Lirtzman (1970) includes the item explanation is clear of what
has to be done.The measure of role conict also by Rizzo et al. (1970) includes the item I receive an assignment
without adequate resources and materials to execute it.Finally, the measure of satisfaction with the leader by
Hackman and Oldham (1980) includes the item How satised are you with the amount of support and guidance
I receive from my supervisor.
One could make the case that all of these measures of potential mediators in the relation between leadership
behaviors and outcomes are to some extent psychometrically contaminated with leadership behaviors, which
may lead to inated correlations because of the overlap in item content across measures (Martinko, Harvey, &
Mackey, 2014). Specically, each of the mediators measures includes some aspect of leadership behaviors be-
cause leaders behave in such ways that they may recognize a subordinates potential (LMX measure), approach
subordinates with professionalism (trust), explain job-related tasks clearly (role ambiguity), provide subordinates
with sufcient resources (role conict), and provide subordinates with support and guidance (satisfaction with the
Table 9. Practical implications of increasing LMX by one standard deviation (.85) for various objective performance metrics.
Outcome Consideration
Initiating
structure
Contingent
rewards
Transformational
leadership Average
Sales volume (in millions) 0.11 0.12 0.05 0.06 0.09
Annual sales growth (%) 7.02 7.87 3.36 4.15 5.60
Sales volume from new
accounts (%)
1.87 2.10 0.90 1.11 1.50
Market share (%) 3.58 4.02 1.72 2.12 2.86
Number of new products sold 4.50 5.04 2.15 2.66 3.59
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 585
leader). This measurement confound seems to be endemic in leadership research and is not unique to studies
assessing LMX. Moreover, as pointed out by an anonymous reviewer, the items of other mediators may have less
overlap with the leadership constructs than does LMX which may be driving the importance of LMX as the key
mediator.
Second, a limitation of meta-analytic research is that it relies upon available data. In the case of our study, the data
were collected within the context of cross-sectional designs, which limits causal inferences. Clearly, these are
design-related challenges that limit our ability to draw strong inferences about causality. However, one way to
strengthen our condence regarding the nature and direction of causal relations is to eliminate alternative explana-
tions for our results (Aguinis & Vandenberg, 2014; Aguinis & Vandenberg, 2014). Specically, we engaged in test-
ing alternative plausible models (i.e., causal structures), as recommended by several sources (Bergh et al., 2016).
Comparing alternative explanations not only increases our condence regarding causal relations, but it also serves
to rene theoretical predictions (Aguinis & Vandenberg, 2014; Aguinis & Vandenberg, 2014). While such compar-
isons do not necessarily allow us to identify the one best explanation, they do allow us to identify the best explana-
tion after other plausible explanations have been ruled out, suggesting an enhancement to precision and renement
in theory.
Third, another limitation of relying upon meta-analytic data is that we were only able to include constructs that
have received a substantial amount of empirical attention, to the point that they have been meta-analyzed. Thus,
we were unable to include all possible mechanisms proposed to mediate the relation between a leadership behavior
and follower performance. Although we found substantial support for the importance of LMX across the different
leadership behaviors, we are not suggesting that the mechanisms that we were unable to include in our study are
unimportant. But, our results provide a benchmark by which to compare all other plausible mediating mechanisms.
If they cannot empirically perform as well or better than LMX, then they are not likely to be as important in the
leadership behaviorsfollower performance relation compared to what has been empirically established as being
the most important.
Fourth, our results lead to the conclusion that it is crucial for leaders to develop a positive relationship with fol-
lowers. Thus, we believe it is important for leadership researchers to place increased attention on further unlocking
how leaders can build high-quality relationships with followers. For example, future research could address the role
of mindfulness and emotional intelligence, which have received relatively little attention with relation to LMX. Spe-
cically, it has been suggested that LMX may mediate the relation between mindfulness (i.e., a present-moment
awareness with an observing, non-judging stance) and important employee outcomes (Reb, Narayanan, &
Chaturvedi, 2014). Additionally, there is some initial support that LMX mediates the relation between emotional
intelligence and important employee outcomes (e.g., turnover intention, job satisfaction), with calls for further
research in this area (Jordan & Troth, 2011).
Fifth, we also recognize that there are a number of theories in the leadership domain that suggest that contextual
moderators play an important role in the leadershipfollower performance relation. Such moderators or boundary
conditions could theoretically include tenure of leaderfollower relationship, culture of the organization, job of
the follower, and gender and/or race of leader and follower, to name a few. In most tests for moderators, product
terms are needed and MASEM is no exception. In order to test for moderators in the context of MASEM, it is nec-
essary to input meta-analytically derived correlations between product terms and other variables included in the
study. Because most primary-level studies do not report the correlations between product terms and other variables,
the data necessary to test for moderators are not available (Aguinis, Beaty, Boik, & Pierce, 2005). While such infor-
mation is currently not available, we have seen a movement toward increased transparency and better reporting prac-
tices in organizational behavior and related elds (e.g., Bettis, Ethiraj, Gambardella, Helfat, & Mitchell, 2016). We
hope that future work will be able to test some of these boundary conditions when authors make the necessary
information available.
Sixth, also in terms of future research, the approach that we used in the particular leadership behaviorfollower per-
formance domain can be used to extract meta-theoretical principles in other areas associated with leadership, but also
many other domains including individual as well as team and rm performance. For example, at the individual level
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
586 R. K. GOTTFREDSON AND H. AGUINIS
of analysis, there are numerous underlying mechanisms that have been posited as explanations for why organizational
commitment leads to individual performance. At the team level of analysis, there are several reasons why various
team characteristics are associated with team performance. Also, at the rm level of analysis, there are numerous
competing explanations for why different types of resources are associated with rm performance. There seem to
be clouds of fogin these and many other domains in organizational behavior, and related elds, and adopting
our dual deductive and inductive approach, combined with the use of meta-analytic structural equation modeling,
seems to be an approach that may be useful for changing the conversationin these other research domains.
Finally, also related to the MASEM approach we adopted in our manuscript, making the meta-analytically derived
correlation matrices in Tables 27 available will allow others to replicate and also extend our analyses. Specically,
given concerns about replicability and research misconduct (e.g., Bedeian, Taylor, & Miller, 2010), making these
data available allows others to conduct the exact same analyses we did, thereby increasing condence in our results
and conclusions. Moreover, taken together, these tables reporting approximately 90 meta-analytically derived corre-
lations based on more than 3000 studies and 900 000 observations can serve as input for future MASEM on leader-
ship but also on other domains as well such as trust, justice, commitment, task performance, OCB, and job
satisfaction that serve as building blocks for the eld of organizational behavior.
Conclusion
There has been a signicant amount of research aimed at identifying why leadership behaviors lead to follower per-
formance. While we have learned much from this literature, we now have a plethora of theoretical rationales and
mechanism used to explain the leadership behaviorsfollower performance relations, suggesting a lack of clarity
in our explanations for these relations. We adopted a dual deductiveinductive approach, compiled a large dataset
based on 35 meta-analyses involving 3327 primary-level studies and 930 349 observations, and used meta-analytic
structural equation modeling to prune and rene the theoretical landscape. We investigated competing theoretical
mechanisms that have been relied upon to explain the relations between four major leadership behaviors (i.e., con-
sideration, initiating structure, contingent rewards, and transformational leadership) and two types of follower per-
formance (task performance and OCB). Recognizing that there are methodological limitations of our data sets,
results indicated that LMX is the common and strongest mediating mechanism across the leadership behaviors, mak-
ing it a meta-theoretical principle, which indicates that greater emphasis should be given to relational leadership the-
ory in leadership behaviorsfollower performance relations. In short, the leaderfollower relationship, as perceived
by followers, is what seems to chart the pathway from leadership to follower performance, suggesting an important
shift in leadership theory and practice.
Author biographies
Ryan K. Gottfredson is an assistant professor of management in the Mihaylo College of Business at California
State University, Fullerton; and a consultant at Gallup, Inc. He has published over 15 journal articles (JOB, JOM,
ORM, AMLE), and his research interests include leadership, performance management, employee perceptions,
and research methods and analysis.
Herman Aguinis is the Avram Tucker Distinguished Scholar and Professor of Management in the George
Washington University School of Business. He has published ve books and more than 130 articles addressing hu-
man capital acquisition, development, deployment, and research methods and analysis. He is a fellow of the Acad-
emy of Management (AOM) and received the AOM Research Methods Division Distinguished Career Award for
lifetime contributions.
Copyright © 2016 John Wiley & Sons, Ltd. J. Organiz. Behav. 38, 558591 (2017)
DOI: 10.1002/job
LEADERSHIP BEHAVIORS AND FOLLOWER PERFORMANCE 587
References
Aguinis, H., Beaty, J. C., Boik, R. J., & Pierce, C. A. (2005). Effect size and power in assessing moderating effects of categorical
variables using multiple regression: A 30-year review. Journal of Applied Psychology,90,94107.
Aguinis, H., Gottfredson, R. K., & Joo, H. (2013). Best-practice recommendations for dening, identifying, and handling outliers.
Organizational Research Methods,16, 270301.
Aguinis, H., Joo, H., & Gottfredson, R. K. (2013).<