The Leadership Arena-Reputation-Identity (LARI) Model: Distinguishing Shared and Unique Perspectives in Multisource Leadership Ratings

Abstract

Multisource leadership ratings rely on the assumption that –in addition to the leader’s self-evaluation– different rater groups (i.e., subordinates, peers and superiors) bring in unique perspectives and thus provide a more well-rounded analysis of the leader’s behavior. However, the way in which multisource data are typically treated in research offers little information about the precise levels of overlap and uniqueness that are encapsulated in these different perspectives. Drawing on the Trait-Reputation-Identity (TRI) model, we propose a model that conceptualizes these shared and unique perspectives in terms of latent factors reflecting respectively (i) the consensus about the leader (i.e., the Leadership Arena), (ii) the impressions conveyed to others that are distinct from self-perceptions (i.e., the leader’s Reputation), and (iii) the unique self-perceptions of the leader (i.e., the leader’s Identity). This Leadership Arena-Reputation-Identity (LARI) model is formalized by means of bifactor modeling, which allows to statistically decompose the variance captured by multisource ratings. The LARI model was tested against five alternative models in two large multisource samples (N1 leaders = 537, N1 observers = 7,337; N2 leaders = 1,255, N2 observers = 15,777), each using different leadership instruments. In both samples, the LARI bifactor model outperformed the alternative models. A subsequent variance decomposition showed that each rater source indeed provides unique information about the target’s behavior, although in varying degree. Across all leadership dimensions in both samples, superiors consistently provided the largest share of unique information among the three observer groups. Implications and future directions are discussed.

Full text

The Leadership Arena-Reputation-Identity (LARI) Model: Distinguishing shared and
unique perspectives in multisource leadership ratings
Jasmine Vergauwea*, Joeri Hofmansb, and Bart Willea
aGhent University, Belgium
bVrije Universiteit Brussel, Belgium
Accepted for publication in the
Journal of Applied Psychology (January 30, 2022)
*Address correspondence to: dr. Jasmine Vergauwe, Department of Developmental,
Personality, and Social Psychology, Ghent University. H. Dunantlaan 2, B-9000 Gent.
Belgium. Jasmine.Vergauwe@ugent.be Tel.: +32 9 264 64 29
Funding: This work was supported by the Research Foundation – Flanders (FWO), grant
number 1200919N.
Acknowledgements: We would like to thank Robert B. Kaiser (Kaiser Leadership Solutions)
and Patrick Vermeren (PerCo) for enabling the use of their multisource data in this article
(i.e., Sample 1 and 2 respectively). Further, we thank Louis Tay for his valuable and
constructive feedback on an earlier version of the manuscript.
© 2022, American Psychological Association. This paper is not the copy of record and may
not exactly replicate the final, authoritative version of the article. Please do not copy or cite
without authors' permission. The final article will be available, upon publication, via its DOI:
10.1037/apl0001012

THE LARI MODEL
Abstract
Multisource leadership ratings rely on the assumption that –in addition to the leader’s self-
evaluation– different rater groups (i.e., subordinates, peers and superiors) bring in unique
perspectives and thus provide a more well-rounded analysis of the leader’s behavior. However,
the way in which multisource data are typically treated in research offers little information
about the precise levels of overlap and uniqueness that are encapsulated in these different
perspectives. Drawing on the Trait-Reputation-Identity (TRI) model, we propose a model that
conceptualizes these shared and unique perspectives in terms of latent factors reflecting
respectively (i) the consensus about the leader (i.e., the Leadership Arena), (ii) the impressions
conveyed to others that are distinct from self-perceptions (i.e., the leader’s Reputation), and
(iii) the unique self-perceptions of the leader (i.e., the leader’s Identity). This Leadership Arena-
Reputation-Identity (LARI) model is formalized by means of bifactor modeling, which allows
to statistically decompose the variance captured by multisource ratings. The LARI model was
tested against five alternative models in two large multisource samples (N1 leaders = 537, N1
observers = 7,337; N2 leaders = 1,255, N2 observers = 15,777), each using different leadership
instruments. In both samples, the LARI bifactor model outperformed the alternative models. A
subsequent variance decomposition showed that each rater source indeed provides unique
information about the target’s behavior, although in varying degree. Across all leadership
dimensions in both samples, superiors consistently provided the largest share of unique
information among the three observer groups. Implications and future directions are discussed.
Keywords: multisource leadership ratings, 360-degree feedback, self-other agreement, unique
leadership perceptions, bifactor modeling

THE LARI MODEL 1
The Leadership Arena-Reputation-Identity (LARI) Model: Distinguishing shared and
unique perspectives in multisource leadership ratings
Across various disciplines of psychology, researchers and practitioners have shown great
interest in how people’s self-perceptions converge with perceptions by others (Connelly & Ones,
2010; Fleenor et al., 2010). In the leadership domain, in particular, it is typically argued that
comparing self-ratings to other-ratings on key leadership dimensions highlights areas of
agreement and disagreement (i.e., “blind spots”), which can be used to inform leadership training
and development (e.g., Day et al., 2014). In addition, it allows tapping leaders’ levels of self-
awareness regarding their leadership behaviors and capabilities, which has also been linked to
important performance outcomes (Atwater & Yammarino, 1992; Fleenor et al., 2010). Because
of these reasons, multisource rating procedures and their feedback programs, often referred to as
360s, are widely used in leadership assessments across the globe (Slater & Coyle, 2014).
At the same time, however, research has questioned the use of 360s in organizational
contexts (e.g., LeBreton et al., 2003; Mount et al., 1998; Scullen et al., 2000). Critics refer to the
administrative burden and relatively high implementation costs associated with having multiple
people rating one and the same leader. Central to this discussion is the question of whether and to
what extent 360s indeed deliver what they promise. Do they really allow uncovering different
perspectives on a target leader’s behavior? Because only in the case that this is true,
organizations can justify the implementation of burdensome and costly 360s as an investment for
the future.
Multisource leadership ratings are based on the deep-rooted belief that, next to the leader
themselves, individuals from different rater groups (i.e., supervisors, peers, subordinates) provide
unique perspectives on a leader’s behavior and performance (Borman, 1997). This is known as

THE LARI MODEL 2
the ‘discrepancy hypothesis’ and is purported to occur because of a variety of reasons, including
leaders not acting similarly around all rater groups, and different rater groups having different
opportunities to observe leaders (Borman, 1997; Fleenor et al., 2010; LeBreton et al., 2003; Lee
& Carpenter, 2018; Tornow, 1993). For about 40 years already, leadership scholars have been
testing this assumption by comparing self- and other-ratings and evaluating their levels of
convergence (e.g., Conway & Huffcutt, 1997; Harris & Schaubroeck, 1988; Lee & Carpenter,
2018; Mount, 1984). Consistent with the idea that these ratings represent a mixture of overlap
and uniqueness, meta-analyses show moderate correlations between different observer sources
(i.e., ranging between .22 and .32 for managerial jobs; Conway & Huffcutt, 1997), as well as
moderate correlations between leader- and observer-ratings of leader behavior (i.e., in the .30
range across leadership dimensions, as well as across types of observers; Lee & Carpenter,
2018).
The relevance of multiple rater groups has also been extensively examined in research on
the structure and reliability of multisource performance ratings (MSPRs) (e.g., Hoffman et al.,
2010; Lance et al., 2008; Mount et al., 1998; Scullen et al., 2000; Viswesvaran et al., 2005). A
large body of literature has relied on analytical approaches such as (second-order) factor analyses
(e.g., Hoffman et al., 2010) and/or multitrait-multirater analyses (e.g., Woehr et al., 2005) to
identify the sources of variance underlying MSPRs. As the most recent example in this research
tradition, Jackson et al. (2020) used Bayesian generalizability theory to get an estimate of the
proportion of true score variance in multisource ratings associated with ratee-, rater-, source-,
and dimension-related effects. In contrast to earlier studies that downplayed the importance of
source effects (Mount et al., 1998; Scullen et al., 2000), Jackson et al. (2020) demonstrated that
large portions of variance in MSPRs are attributable to source-related effects, cumulatively

THE LARI MODEL 3
explaining between 30.90 and 58.06% of the variance in ratings across samples. Such large
source effects again underscore the idea that different rater sources provide unique perspectives
on the leader.
Taken together, the literatures on self-other agreement and on the structure of MSPRs
have a long history in applied psychology, and they seem to converge on the notion that different
rater sources can indeed contribute both unique and shared information.Yet, very little is known
about the exact nature of these different source effects and their size relative to each other. This
raises the fundamental question: “How important are each of the four typical rater sources (i.e.,
self, peers, subordinates, superiors) in multisource leadership ratings?”
The Relevance of Scrutinizing Unique Rater Source Effects
Studying the existence and magnitude of these various rater source effects is important
for theoretical reasons, while it might also provide crucial insights for leadership research and -
practice. First, from a theoretical point of view, it aligns well with the fundamental idea of the
attribution theory of leadership, which assumes that leadership is determined, to a certain extent,
by the perceptions of people surrounding the focal leader (e.g., Foti et al., 1982; Hogan, 1996;
Lord et al., 1984, 2020). For example, research has shown that through socialization and earlier
experiences with leaders, people develop cognitive structures or prototypes of leadership
categories (i.e., implicit leadership theories), and that people rely on those cognitive structures or
prototypes when evaluating a particular leader (Lord et al., 2020). Importantly, being perceived
as a leader influences both self- and observers’ assessments of leadership, and ultimately matters
for individual- (e.g., career progress), team- (e.g., cohesion), and organizational outcomes (e.g.,
performance; Lord et al., 2020). Similarly, socio-analytic theory (Hogan, 1996) emphasizes the
importance of people’s reputations (i.e., observers’ perceptions) on work outcomes such as

THE LARI MODEL 4
career success (Hogan & Blickle, 2018). Although a great deal has already been written about
leadership being in the eye of the beholder (e.g., Lord et al., 2020; Nye, 2002), researchers seem
to focus on these observer perceptions without taking into account potential overlap with leaders’
self-perceptions. Separating observer perspectives from the leader’s self-perception could
potentially advance these attributional theories by shedding light on the exact extent to which
leadership lies uniquely ‘in the eye of the beholder’. Although part of these observer perceptions
will be rater-specific, there will also be commonality, capturing the leader’s general “reputation”.
A subsequent question is then whether and to which extent this unique observer perspective can
be further decomposed in different source-specific reputation factors that lie uniquely in the eyes
of peers, subordinates, and superiors, respectively.
Further, knowledge about the uniqueness of each rater source can also inform basic
leadership research. As mentioned earlier, one of the central premises of 360-degree assessments
is that different rater groups contribute unique information to the leader's overall assessment
(Borman, 1997). However, the traditional treatment of multisource data does not allow for an
empirical test of this central assumption. Most obviously, research aggregating the ratings across
the different rater sources (e.g., Kaiser et al., 2015) represents a rudimentary way to account for
the variability in rater perspectives. Although the resulting observer score might be ‘rich’ in the
sense that all different perspectives contribute to this score, this approach neglects the possibility
that the different sources diverge in their perceptions to an important extent (e.g., Jackson et al.,
2020). A second, more common, approach in leadership research is to compare the different
observer perspectives separately to the self-rating (e.g., Atwater et al., 2009), or to relate them
separately to external variables of interest (e.g., Aramovich & Blankenship, 2020; Atwater &
Brett, 2006). A crucial difficulty here is that, besides unique variation, each rater group also

THE LARI MODEL 5
shares variation with the other rater groups (Conway & Huffcutt, 1997). Because the exact
composition of shared and unique variation between rater groups is unknown, it is unclear what
the ‘source-specific’ effects actually reflect. Knowing these exact variance compositions does
not only allow testing the central premise of 360s (i.e., the discrepancy hypothesis), it may also
provide new insights regarding the way multisource ratings should be treated in research (e.g.,
aggregation methods), while revealing new perspectives on the construct of self-other agreement.
Finally, from a practical point of view, knowing which percentage of variance is captured
by the self and the various observer groups gives an idea of the usefulness or even the necessity
of including these different rater groups in multisource leadership ratings. Organizations can
approach the use of these procedures in terms of a trade-off between collecting as much
information as possible and keeping the implementation cost (and administrative burden) as low
as possible. Knowing that the unique input provided by one or more rater groups is small or even
negligible may help organizations in deciding which rater groups to include or not.
In sum, although source effects are “alive and well” (Hoffman et al., 2010; Jackson et al.,
2020), much remains to be learned about the degree to which rater groups involved in 360s
converge and diverge in their perspectives. To fully exploit the benefits of multisource leadership
ratings, both theoretical and methodological innovations are needed. Theoretically, an advanced
application of the Trait-Reputation-Identity (TRI) model (McAbee & Connelly, 2016) is
proposed that can facilitate our conceptual understanding of these different rater source effects
by separating consensus about the leader, unique self-perceptions of the leader, and unique
observer-perceptions. Methodologically, we provide a tool to distinguish and quantify these
shared and unique perspectives in multisource data. Before outlining our framework in greater

THE LARI MODEL 6
detail, we provide an overview of the different theoretical mechanisms that may explain why
various rater groups may hold unique perspectives on a focal leader.
Theoretical Perspectives on Leader Self- and Observer Ratings
There are several accounts of why raters from different groups can hold unique
perspectives on a target’s behavior, and those accounts can broadly be categorized into those that
refer to (i) role differences, (ii) informational differences, and (iii) motivational differences. First,
different rater groups might be viewing the focal leader through different lenses due to
evaluating the leader in varying roles (Atwater & Yammarino, 1997; Yammarino, 2003). In
particular, superiors rate them as followers, peers rate them as coworkers, subordinates rate them
as managers, and leaders may rate themselves as a blend of all of these roles (Lee & Carpenter,
2018). Each of these different roles may come with different ideas and beliefs about what
constitutes “leadership”, which could create differences between these perspectives (e.g.,
Hooijberg & Choi, 2000).
Second, particular rater groups can have more knowledge of specific leader behaviors
because there is simply more information at their disposal to evaluate these particular behaviors
(i.e., information availability) or because there is a higher chance of noticing particular behaviors
(i.e., information detection) (Funder, 2012; Rothstein, 1990; Vazire, 2010). Self- and other-
perceptions are, for instance, likely to differ in quantity and type of information available. The
self has a major advantage over others because “no one else has access to more information” (p.
277; Paulhus & Vazire, 2007). By the same token, although –theoretically– the self can observe
most of its own behaviors, it is unlikely to detect many of these behaviors. As oneself is not as
salient in one’s visual field as it is in others’, some behaviors will not be detected. Moreover,
even if people detect their own overt behavior, the self usually places more weight on thoughts

THE LARI MODEL 7
and feelings than on overt behaviors when forming self-perceptions. This effect is usually
reversed when forming perceptions of others (Andersen, 1984). Against this background, the
self-other knowledge asymmetry (SOKA) model (Vazire, 2010) expects that the leader
themselves knows more than others about low-observable behaviors. Highly observable leader
behaviors, on the other hand, are at least equally known to others as compared to the self (Vazire,
2010).
In a similar vein, informational differences can also exist between the different observer
groups (i.e., superiors, peers, subordinates). This idea aligns with Funder’s (1995, 2012) realistic
accuracy model, which holds that observer accuracy depends on the extent to which behavioral
cues are available to observation and the extent to which they are detected by an observer. In
general, subordinates may have more knowledge of leadership behavior than would superiors,
given that subordinates are the direct recipients of leader behaviors, whereas superiors are
usually not (Hansbrough et al., 2015; Lee & Carpenter, 2018). However, some behaviors are
easier to witness for particular (and not all) groups of observers, as leaders may not act similarly
around all rater groups (Borman, 1997; Yukl, 2010). Hence, it is likely that some type of leader
behaviors can only be witnessed in certain contexts. Subordinates may, for instance, have more
knowledge of interpersonal leader behavior than would superiors, whereas superiors may have
more unique knowledge of strategic or business-related leader behavior (Hiller et al., 2011).
Finally, social information processing theory (Salancik & Pfeffer, 1978) predicts that
different motivational sources may drive leader- and the various observer ratings. Leaders
themselves can be susceptible to self-enhancement bias and inflate their ratings in order to
present themselves in a favorable light to others or to protect their own self-image (Atwater et
al., 1998). Moreover, leaders may be well aware that the impression they make on others has

THE LARI MODEL 8
implications for how others perceive and evaluate them. Therefore, leaders can be motivated to
control the impressions others form of them, behaving in ways that create certain impressions in
others’ eyes (Leary & Kowalski, 1990). Observers, in turn, may also be vulnerable to biasing
motives, as they may intentionally provide inflated ratings of leaders to avoid punishment from
leaders or other forms of backlash (Lee & Carpenter, 2018). Not surprisingly, raters who pursue
to (a) identify the target’s weaknesses or (b) strengths, (c) provide fair ratings or (d) motivate the
target, eventually provide different ratings of the same behavior. So raters pursuing different
goals may actually give different ratings (Murphy et al., 2004). Hence, shared intentions among
raters of the same group would also contribute to unique perspectives on the target’s behavior
(e.g., subordinates may want to emphasize the leader’s strengths, whereas superiors may want to
uncover developmental possibilities).
The Leadership Arena-Reputation-Identity (LARI) Model
The objective of the current paper is to propose a framework for conceptualizing and
quantifying the different rater source effects in multisource leadership ratings. To this end, we
turn to recent developments in personality psychology, a field that has amassed a substantial
body of knowledge about how strong consensus is across raters and how self- versus other-
ratings of personality uniquely contribute to the prediction of relevant outcomes at work and
beyond (e.g., Connelly & Ones, 2010; Connolly et al., 2007).
Recently, McAbee and Connelly (2016) developed a formal framework for studying
multi-rater personality data, teasing apart the different ways in which a person’s personality is
constructed by oneself and by others. The basis of their model is the Johari window (Luft &
Ingham, 1955), which maps information known versus unknown to the self and observers in a
two-by-two grid (see left part of Figure 1). The “Arena” reflects information that is shared

THE LARI MODEL 9
between the self and others (i.e., known to the self and observers). The “Façade” captures self-
knowledge that is not shared with others (i.e., known to the self but unknown to observers). The
“Blind-Spot” represents aspects of a target that others see, but the target is unaware of or does
not endorse (i.e., known to observers but unknown to the self). Finally, the Johari window
acknowledges that some information remains “Unknown” to (or not perceived by) both the self
and observers (Luft & Ingham, 1955). McAbee and Connelly (2016) relabeled the quadrants of
the Johari window as Trait (Arena), Reputation (Blind-Spot), and Identity (Façade) because these
terms aligned better with long traditions of personality research. The “Trait” label reflects a
historical emphasis on corroborating traits through consensual validation (Campbell & Fiske,
1959). In contrast, the distinction between “Reputation” and “Identity” has roots in symbolic
interactionism (Blumer, 1986) and socioanalytic theory (Hogan, 1996). In short, the Trait-
Reputation-Identity (TRI) Model (see right part of Figure 1) was a breakthrough in personality
science by separating the variability in multisource personality ratings into (i) consensus about
the person (i.e., Trait), (ii) unique self-perceptions (i.e., Identity), and (iii) impressions conveyed
to others that are distinct from self-perceptions (i.e., Reputation).
-------------------------------------------Insert Figure 1 about here-------------------------------------------
As recognized by McAbee and Connelly (2016), much of the logic behind the TRI model
can be directly transferred to the context of multisource leadership ratings. There are certain
aspects of a leader’s behavior that everyone (i.e., self and all others) agrees upon. There are also
certain aspects that will be unique to the perception of the leader or to the perception of others.
Yet, a truly unique feature of multisource leadership ratings is that the others or ‘observers’ are
not interchangeable. They instead cluster in conceptually meaningful subgroups, such as (but not
limited to) work peers, supervisors, and subordinates. As explained above, distinguishing these

THE LARI MODEL 10
different observer groups is important because differences in roles, information, and motivations
will likely result in higher homogeneity between people belonging to the same rater groups and
more divergence between individuals belonging to different rater groups (e.g., Borman 1997;
Lee & Carpenter, 2018). Hence, applying the TRI model to multisource leadership ratings
requires a significant extension to the model that allows each of these observer groups to form
their own unique reputation of a target leader.
One particular extension that has been proposed by McAbee and Connelly (2016) adds
multiple rater contexts to the Reputation factor, acknowledging that personality manifestations
can be different in different environments (e.g., at home, at work, among friends; see Figure 4 on
p. 583). Although rater contexts represent rater groups in reference to multisource leadership
ratings, our proposed LARI model provides a similar extension by separating the variance in
multisource leadership ratings into six distinguishable latent factors: The Arena, the shared
Reputation, the Reputation by peers, by subordinates, and by supervisors, and the Identity (see
Figure 2). A formal description of these LARI factors along with an overview of elements
contributing to the respective factors is provided in Table 1. The Arena factor captures
information about the target’s leadership behavior that is shared between the self and all others.
Going back to the original Johari window, we prefer the term “Arena” over “Trait” (cf. TRI) as
the former reflects better what this means in a leadership context. Specifically, this factor
captures the features of the target’s behavior on which all raters –including the leader
themselves– agree. This arena is the place where private leader identity and public leadership
reputation overlap.
-------------------------------------------Insert Figure 2 about here------------------------------------------
-------------------------------------------Insert Table 1 about here-------------------------------------------

THE LARI MODEL 11
A second factor is called general Reputation and refers to information that is shared by
(all) observer groups but is unique from the perspective provided by the leader themselves. This
external reputation can reflect (i) information not available to the self, (ii) information
intentionally not shared by the self (but picked up by all observers), or (iii) systematic bias
shared across all observer groups (e.g., physical appearance stereotypes or a leniency bias).
Importantly, this general reputation factor reflects only what is shared across the different
observer groups, whereas three source-specific reputation factors capture perceptions that are
unique to the perspectives of raters from the same source. This feature of the LARI model is
based on the idea that a leader’s subordinates, peers, and supervisors each have unique
interactions with and expectations of the leader (Lee & Carpenter, 2018).
Specifically, the Reputation by subordinates reflects what is unique in the perspective
provided by subordinates, and not shared with the views provided by superiors, peers, or the
leader themselves. Considering that much of a leader’s behavior involves and is directed towards
subordinates (Hansbrough et al., 2015), and subordinates have many opportunities to witness
their leader in a leadership role (Conway et al., 2001), subordinate-ratings are probably the most
common source of observer-rated leadership. Moreover, compared to the leader’s peers and
superiors, subordinates should have a unique perspective on various leader behaviors including
their motivational style and individualized consideration (Hiller et al., 2011). Peers, relative to
subordinates, likely have less opportunities to observe leader behavior (i.e., in part because they
are themselves leaders). On the other hand, peers may also have the chance to observe a range of
leader behaviors that other individuals in the organization rarely see (Braddy et al., 2014), such
as behavior related to alignment, positioning, and boundary spanning (Hiller et al., 2011).
Further, the leader’s superiors arguably have even fewer occasions to witness leader behavior,

THE LARI MODEL 12
relative to subordinates and peers (Pollack & Pollack, 1996). However, it is likely that superiors
monitor different aspects of a leader’s functioning compared to the other two sources.
Supervisors might be more motivated to notice leader behaviors aimed at realizing the
company’s strategic goals. Also, because of their hierarchical position, superiors may have
different information at their disposal to evaluate or describe a focal leader. The performance of
a target leader’s unit relative to that of other units might, for instance, be better judged by
superiors in the higher echelons.
Finally, the leader Identity factor represents self-perceptions of leadership that are not
shared with others. Leadership research on self-other agreement has referred to the notion of
“bias” in self-ratings to explain self-other discrepancies. This bias represents inaccurate self-
perceptions that may arise for several reasons (Fleenor et al., 2010), including the limited
opportunities to receive feedback from followers, thus limiting (dis)confirmation, and the lack of
motivation to use followers’ perceptions as relevant feedback on their behavior. In addition to
these explanations that refer to “bias” or “inaccuracies”, the uniqueness of leaders’ self-
perceptions may also reflect behaviors or attitudes that are simply low in observability (cf.
SOKA model; Vazire, 2010), such as private goals and strivings that are not fully expressed.
Such “hidden knowledge” thus captures relevant variance in leader behavior that observers are
simply never exposed to (McAbee & Connelly, 2016). Indeed, coworkers probably witness only
a portion of the leader’s full repertoire of behavior (e.g., Allen et al., 2000), and are more likely
to only witness and remember the results of the leader’s behaviors (DeNisi et al., 1984).
Research Questions
Drawing on various theoretical perspectives on self-other and other-other (dis)agreement
in leadership perceptions, it becomes clear how each of the informant groups in multisource

THE LARI MODEL 13
leadership ratings may bring in perspectives that are partly distinctive and partly overlapping.
The proposed LARI model scrutinizes the different rater source effects by disentangling these
shared and unique perspectives for various leadership dimensions. As a first research question
(RQ1), the viability of this model will be explored.
RQ1: Does the presented LARI model, decomposing the shared and unique perspectives
of the four typical rater sources, fit multisource leadership data well?
In a second research question (RQ2), we will focus on the relative size of the different
LARI factors. Grounded in the realistic accuracy model (Funder, 2012), the relative size of the
unique observer perceptions of subordinates is generally expected to be larger, compared to the
unique perceptions of peers and superiors, as subordinates would have more opportunities to
observe and interact with focal leaders (e.g., Conway et al., 2001; Hansbrough et al., 2015; Hiller
et al., 2001; Lee & Carpenter, 2018). Importantly, however, in the current study, a number of
leadership dimensions will be taken into account, each focusing on substantially different aspects
of leadership. A relevant related question then is whether the relative size of the different LARI
factors depends on the specific leadership dimension being assessed (RQ3). On the one hand,
different theories would expect this to be the case. Building on the SOKA model (Vazire, 2010),
a crucial factor in this regard is the observability of those leadership dimensions. For highly
observable dimensions, the shared perceptions (i.e., among everyone (Arena) and among
observers (general Reputation)) are expected to be relatively large, whereas the unique self-
perceptions of the leader (Identity) are expected to be relatively small. For dimensions that are
more difficult to observe, a reversed pattern could be expected. Further, as leaders do not act
similarly around members of different rater groups (Borman, 1997; Yukl, 2010), a larger relative
size of the Reputation by subordinates might be particularly likely for interpersonal leadership

THE LARI MODEL 14
dimensions, whereas peers and superiors could be more knowledgeable about strategic or
business-related leadership dimensions (Hiller et al., 2011). These expectations are further in line
with the theoretical perspective offered by Guion (1965), which holds that source variation may
occur because sources systematically differ in the dimensions they use to evaluate people. In
contrast to these theoretical arguments, however, Jackson et al. (2020) recently provided
evidence that different source perspectives do not depend on dimension-based evaluations.
Instead, their evidence suggested that different rater sources “essentially bypass specific
dimensions altogether and form source-dependent overall impressions of ratees” (p. 325). As
such, the final two research questions are:
RQ2: What is the relative size of the different LARI factors? And does this generalize
across samples and instruments?
RQ3: Is the relative size of the different LARI factors dependent on the specific
leadership dimension being assessed?
Method
We analyzed two large multisource datasets, each of which was operated by a different
consultancy firm specialized in leadership assessment, and each charting a different set of
leadership dimensions. A first international sample rated four leader behaviors (using the
Leadership Versatility Index (LVI); Kaiser et al., 2010), and a second –mainly Belgian– sample
rated eight leadership styles (using the Circumplex Leadership Scan (CLS); Redeker et al.,
2014). As further detailed below, the CLS exclusively maps interpersonal leadership dimensions,
whereas the LVI covers both interpersonal and organizational leadership dimensions. Moreover,
the CLS and the LVI also differ in terms of the rating scale format that is used to measure these
leadership dimensions. Whereas the CLS uses a more traditional frequency-type Likert scale

THE LARI MODEL 15
ranging between 0 (never) and 4 (always), the LVI uses a less conventional “too little/too much”
(TLTM) scale ranging from -4 (much too little), over 0 (the right amount) to +4 (much too
much). Due to variance from the “too much” range (from 0 to +4), TLTM ratings of leadership
provide incremental validity over Likert ratings in predicting performance (Vergauwe et al.,
2017). Hence, the TLTM scale’s ability to capture behavioral excess can be assumed to provide
‘broader’ assessment output with regard to the underlying leadership dimension. In sum, a
number of differences can be observed between the two study samples, including the
(inter)nationality of the targets, the type of leadership dimensions, and the rating scale format.
However, cross-validating the LARI model using these different samples and leadership
instruments is critical to this study as it will allow to strengthen our conclusions regarding the
viability of the presented LARI framework. All research conducted in this study was approved
by the Ethical Committee of the Faculty of Psychology and Educational Sciences of Ghent
University (Study title: Shedding new light on 360° leadership assessment: A multi-rater
framework for studying leadership; Reference number: 2018/54).
Transparency and Openness
Existing data from two consultancy firms were re-analyzed. We describe the study
samples based on the available information in the Sample sections below. We also describe all
data exclusions and all measures in the study, and we adhered to the Journal of Applied
Psychology methodological checklist. Data were analyzed using Mplus 8.4 (Muthén & Muthén,
1998-2017). The analysis code (Mplus syntaxes) for testing all models (RQ1) and for the
variance decomposition of the LARI factors (RQ2) is available on the Open Science Framework
(OSF) via this link, as are the detailed results of these analyses in both study samples. Data, as
well as the measures described in the method section, are not available due to their proprietary

THE LARI MODEL 16
nature. The multisource data of Sample 1 partially overlap with two other publications, as they
represent an extension of the multisource data in Kaiser et al. (2015) and Study 3 of Vergauwe et
al. (2018). Specifically, more target leaders went through a development center, so the sample
size is almost twice as large. The study design, research questions, and the analyses were not
preregistered.
Sample 1
Participants. A first multisource dataset was obtained from a U.S.-based consultancy
firm involved in leadership assessments across the globe. Next to self-ratings of the leaders (N =
537), an average of 14 raters (min. 3; max. 55) rated each leader in terms of leader behaviors,
including at least one subordinate, one peer, and one superior. A total of 7,337 observers,
including 1,142 supervisors, 2,695 peers, and 3,500 subordinates provided ratings in the context
of a development center. Target leaders were on average 45.59 years old (SD = 7.97) and 72.5%
were male. The geographic region of employment was highly diverse. Seventy-five percent of
the leaders worked in North America, 17% in Western Europe, 3% in Africa, 3% in East Asia,
and a smaller percentage was active in Latin America (0.4%), India (0.4%), Caribbean (0.4%),
Australia (0.2%) and the Pacific Islands (0.2%). In terms of organizational level, the targets
operated as supervisor (21.4%), middle manager (12.1%), functional head (28.1%), C-level
executive (6.5%), or general manager (16.6%), while 15.3% indicated ‘other’ in this regard. The
leaders were active in a variety of industries (e.g., IT, banking, aerospace, construction).
Measure. Target leaders, as well as the observers, provided ratings on the 48 items of the
original English version of the Leadership Versatility Index (LVI), tapping into four leader
behaviors: forceful, enabling, strategic, and operational (Kaiser et al., 2010). Whereas forceful
and enabling leadership represent interpersonal leadership dimensions, strategic and operational

THE LARI MODEL 17
leadership represent organizational leadership dimensions. Each of the four leader behaviors
includes three subscales. Forceful leadership is defined as assuming authority and expecting a lot
from other people by (a) taking charge, (b) declaring themselves, and (c) pushing for
performance. Enabling leadership concerns creating conditions for others to contribute through
(a) empowerment, (b) participation, and (c) support. Strategic leadership is defined as
positioning the team for the future by (a) setting direction, (b) stimulating growth, and (c)
supporting innovation. Operational leadership, finally, refers to guiding the team to execute near-
term goals by (a) specifying the details of implementation (‘execution’), (b) focusing resources
(‘efficiency’), and (c) managing in a process-oriented way (‘order’). Each of the 12 subscales
were measured by means of four items that were rated on a 9-point scale ranging between -4
(much too little), 0 (the right amount), and +4 (much too much) (Kaiser et al., 2010). Table 2
shows that both the self-ratings of the LVI scales as well as the observer ratings showed a high
level of internal consistency (α’s between .69 and .84 for the self-ratings; and between [.72 -
.89], [.75 - .90], and [.70 - .88] for the supervisor-, peer-, and subordinate ratings respectively).
To obtain model indicators for each of the four leader behaviors, subscale scores were
aggregated across all individuals belonging to a specific rater group, such that 12 aggregated
subscale scores were obtained for each of the three observer groups (e.g., a score on ‘taking
charge’ for the subordinate-, peer-, and supervisor group). The rwg(j) interrater agreement
coefficient (IRA; James et al., 1984) was computed for each subscale within superior, peer, and
subordinate groups. The results in the Appendix (Table A) indicate that the level of similarity
within the different rater groups was sufficiently high to support aggregation (LeBreton &
Senter, 2008). All descriptive statistics, correlations, and internal consistencies of the variables in
Sample 1 are reported in Table 2.

THE LARI MODEL 18
-------------------------------------------Insert Table 2 about here-------------------------------------------
Sample 2
Participants. A second multisource dataset was obtained from a Belgian HR consultancy
firm. Leaders (N = 1,863) were assessed for developmental purposes, and a subset of 1,255
leaders that were rated by all three observer groups was retained for this study. In about 40% of
the cases, demographical information is missing, as these were optional items. To have an idea of
the sample characteristics, however, an estimate was made based on the complete cases. The
mean age of the target leaders was 44.17 (SD = 8.80), and 75.1% was male. The targets’ country
of origin was largely Belgium (89.3%) and the Netherlands (9%), and a small percentage (1.7%)
originated from other countries (i.e., 0.5% from France, 0.5% from South-Africa and Niger, and
0.7% from the United Arab Emirates, Spain, Croatia, Turkey and Italy). Their current
organizational level was described as non-management (22.7%), front-line management
(supervisor) (24.8%), middle management (manages managers) (29.7%), senior management
(14.7%), executive (reports to CEO) (5.6%), and CEO (2.6%). In terms of industries, the sample
was highly heterogeneous (e.g., 30.2% manufacturing, 10% information and communication,
9.1% human health/social work). An average of 12 raters (min. 3; max. 39) rated each leader,
including at least one subordinate, one peer, and one superior. In total, 15,777 observers
participated in this study (i.e., 2,175 supervisors, 5,068 peers, and 8,534 subordinates).
Measure. Both the target leaders and the observers provided ratings on the Circumplex
Leadership Scan (CLS; Redeker et al., 2014), comprising 116 items that tap into (positive and
negative) leadership behavior. The large majority of the participants (90.2% targets; 89.5%
observers) provided ratings on the validated Dutch version of the CLS (Redeker et al., 2014),
whereas a smaller percentage provided ratings in French (8.8% targets; 9.1% observers) or in

THE LARI MODEL 19
English (1% targets; 1.4% observers). Redeker et al. (2014) report that the original Dutch items
“were translated in both English and French by professional linguists” (p. 438). Contact with one
of the authors further clarified that two professional linguists translated the CLS using
translation-back translation procedures (Brislin, 1970), until quasi complete agreement was
reached. Table B in the Appendix shows the results of CFA models testing measurement
invariance across languages (Dutch, French, and English), showing full metric and partial scalar
measurement invariance across languages for each of the CLS leadership styles. Two dimensions
span the circular ordering of the circumplex: ‘communion’ (affiliation; the horizontal axis), and
‘agency’ (control/dominance; the vertical axis). Eight leadership styles were assessed, each
representing a different octant in this model: Coaching (e.g., “makes positive comments”),
inspirational (“has strong character”), directive (“puts employees in their place”), authoritarian
(“is bossy”), distrustful (“is suspicious”), withdrawn (“is isolated”), yielding (“wants to please
everybody”), and participative leadership (“accepts other approaches”). Items were rated on a
Likert scale ranging from 0 (never) to 4 (always). All descriptive statistics, correlations, and
internal consistencies of the variables in Sample 2 are reported in Table 3. As can be seen in
Table 3, both the self-ratings of the CLS scales as well as the observer ratings showed a high
level of internal consistency (α’s between .74 and .87 for the self-ratings; and between [.80 -
.91], [.80 - .82], and [.75 - .93] for the supervisor-, peer-, and subordinate ratings respectively).
Within each observer group, 116 aggregated item scores were calculated (e.g.,
CLS_item1 was aggregated across the leader’s subordinates, across the leader’s peers, and across
the leader’s supervisors) to use as (model) indicators. The rwg(j) IRA (James et al., 1984) was
computed for each item within superior, peer, and subordinate groups. The results in the

THE LARI MODEL 20
Appendix (Table C) indicate that the level of similarity within rater sources is sufficiently high to
support aggregation (LeBreton & Senter, 2008).
-------------------------------------------Insert Table 3 about here-------------------------------------------
Statistical Analyses
To address RQ1 (i.e., Does the LARI model fit multisource leadership data well?), we
tested the LARI model using an analytical approach similar to the one proposed by McAbee and
Connelly (2016). In essence, the statistical counterpart of our LARI model is a bifactor model
(Reise, 2012). In bifactor modeling, the factor indicators simultaneously load on general and
specific factors (which themselves are uncorrelated). Such a model reflects the idea that variation
in the indicators is believed to be caused by these different general and specific sources. In the
LARI bifactor model (see Figure 3, Panel A), self-ratings load on the Arena and Identity factor,
while other-ratings load on the Arena, the general Reputation and a source-specific Reputation
factor. This model adheres to the idea that self-ratings comprise a mixture of perceptions shared
by everyone (i.e., Arena), self-perceptions not shared with others (i.e., Identity), and
measurement error. Other-ratings, in turn, are a blend of perceptions shared by everyone (i.e.,
Arena), perceptions shared by all external observers (i.e., general Reputation), perceptions
unique to the specific group one belongs to (e.g., Reputation by peers)
1
, and measurement error.
When testing the LARI bifactor model, we also allowed for residual correlations between
identical indicators across observer groups (see also McAbee & Connelly, 2016; Olsen & Kenny
2006). These residual correlations represent consensus in aspects of those indicators not captured
by the Arena and reputation factors.
1
Note that when a single rater from a given group is used to represent a source-specific factor—rather than
aggregated ratings across raters of that group—it is not possible to separate source-specific variance from rater-
idiosyncrasy.

THE LARI MODEL 21
Apart from testing the LARI bifactor model (Panel A), we also compared it to a set of
five alternative models, including a LARI bifactor model without a general Reputation factor
(Panel B), a LARI bifactor model without source-specific Reputation factors (Panel C), a higher-
order factor model (Panel D), a higher-order model including a general Reputation factor (Panel
E), and a correlated-factors model (Panel F; see Figure 3 for an overview). In the LARI bifactor
model, the general Reputation factor captures commonality in the other-ratings that is not shared
with self-ratings. Hence, finding that model fit is not improved in the LARI bifactor model
relative to the model without a general Reputation factor (see Panel B) would suggest that there
is little communality in the perceptions uniquely shared by the external observers. Finding that
model fit is not improved relative to the model without source-specific factors (Panel C), on the
other hand, would suggest that the variance that is shared among raters from the same source is
negligible. As such, support for this model would question the existence of source effects in
multisource leadership ratings. The higher-order factor model (see Panel D) represents a model
in which source-specific factors define a higher-order factor. Hence, finding comparable fit for
the higher-order factor model than for the LARI bifactor model would suggest that the source-
specific factors (i.e., the Reputation and Identity factors in the LARI model) and the higher-order
factor (i.e., the Arena factor in the LARI model) are isomorphic (McAbee & Connelly, 2016).
Compared to a higher-order factor model, a higher-order model including a general Reputation
factor (see Panel E), in which the three external rater sources load on a general Reputation factor,
is closer to the proposed LARI model, as it also separates Identity (self) from Reputation (all
external observers). Finally, the correlated-factors model (see Panel F) tests the existence of
source-specific factors without assuming a specific higher-order factorial structure in the source-
specific factors. Hence, finding that model fit is not improved in the LARI bifactor model

THE LARI MODEL 22
relative to the correlated-factors model would suggest the existence of source-specific
perceptions without requiring the specific LARI patterns of loadings.
-------------------------------------------Insert Figure 3 about here-------------------------------------------
To address RQ2 (i.e., What is the relative size of the different LARI factors?), the LARI
bifactor model was scrutinized in terms of explained variance by each of the LARI factors. To
this end, an extension of the Explained Common Variance (ECV) statistic (see Reise, 2012) was
used. Specifically, we calculated the proportion of variance explained by each of the LARI
factors as the sum of the squared factor loadings for the respective factor divided by the sum of
the factor loadings across all LARI factors
2
(see also McAbee & Connelly, 2016). Note that this
approach disregards residual variances (and correlated residuals), which means that ECV
captures the percentage of the explained variance (and not the percentage of the total variance)
that is accounted for by the different LARI factors.
Finally, to address RQ3 (i.e., Is the relative size of the different LARI factors dependent
on the leadership dimension?), these variance decompositions among the LARI factors will be
compared between leadership dimensions. In the interest of transparency, these comparisons will
be described as exploratory.
All models were tested in Mplus 8.4 (Muthén & Muthén, 1998-2017) using Bayesian
estimation and relying on the default Mplus priors
3
. To evaluate model fit, we evaluated the
Comparative Fit Index (CFI) and the Tucker–Lewis Index (TLI), for which values ≥. 90 suggest
2
Because in the bifactor model, for identification purposes the factor variances are fixed to one rather than fixing the
loading of a marker item, the factor loadings in the LARI model are standardized with respect to the latent variables.
3
Working within a Bayesian framework, we did not report degrees of freedom (note that they are not part of the
Mplus output when working with Bayesian estimation). However, the degrees of freedom can easily be calculated
from the information that is reported. For the LARI model for forceful behavior (see Table 4), for instance, we have
75 estimated (or free) parameters (i.e., 33 loadings, 18 residual covariances between like items, 12 residual
variances, and 12 item intercepts) and we have 90 data points (i.e., 78 variances and covariances + 12 item
means/intercepts). This means that this model has 15 degrees of freedom.

THE LARI MODEL 23
an adequate model fit. Moreover, we use the Root Mean Square of Error of Approximation
(RMSEA), with values of ≤ .10 pointing to an acceptable fit, ≤ .08 to an approximate fit, and ≤
.05 to a good model fit (Chen et al., 2008). For model comparison purposes, we used the
Deviance Information Criterion (DIC), which is the Bayesian alternative of the Akaike
Information Criterion (AIC). Similar to AIC, DIC is as a measure that balances model fit and
model complexity, which is why it allows comparing non-nested models. Lower values on DIC
refer to a better fit-complexity ratio and thus (relatively) better models (Spiegelhalter et al.,
2014). Finally, although the same analytical models were tested on the data of both study
samples, one important difference between the two measures is that the LVI (Sample 1) is a
hierarchically structured leadership instrument in which the four leadership dimensions (e.g.,
enabling) each have three underlying leadership facets (e.g., empowerment, participation,
support), whereas the CLS (Sample 2) is not hierarchically structured. Therefore, the one
difference between the analytical models that were tested on the two samples is that the four
leader behaviors in Sample 1 were modeled by their respective subscales, whereas the eight
leadership styles in Sample 2 were modeled by their respective items. In other words, in Sample
1, subscale scores were used as factor indicators, whereas the factor indicators were item scores
in Sample 2.
Results
Sample 1
Testing the LARI model. Table 4 summarizes the model fit of all tested models
4
. First,
the LARI model without source-specific Reputation factors does not fit the data well (e.g., CFI =
.783, .858, .713, and .819; RMSEA = .191, .123, .193 and .107 for forceful, enabling, strategic,
4
The detailed results for all tested models, in both Sample 1 and Sample 2, can be found on OSF via this link.

THE LARI MODEL 24
and operational respectively). Although the other alternative models fit the data well, they are all
outperformed by the LARI model for each of the four leader behaviors. This can be seen from
the fact that the LARI model has the best absolute fit values (e.g., CFI = .993, 1, 1, and .990;
RMSEA = .047, .00, .00, .023 for forceful, enabling, strategic, and operational respectively), but
more importantly that it also has the lowest DIC value (DIC = 8927.91, 7101.20, 5937.18, and
6089.06 for forceful, enabling, strategic, and operational respectively). Hence, it appears that,
among the tested models, the LARI model represents the data best.
-------------------------------------------Insert Table 4 about here-------------------------------------------
Variance explained by LARI factors. Variance decomposition of the LARI factors is
presented for each of the four leader behaviors (see Figure 4). Across these behaviors, even for
the smallest percentage of explained variance, the 95% credibility interval did not include zero
(i.e., 5.1%; 95% CI 0.4 - 9.1 for Reputation by peers in forceful behavior). As systematic
variation is captured by each LARI factor, each LARI factor turns out to be relevant.
For each of the four LVI leader behaviors, the leadership Arena explained the largest
proportion of the variance in leader behavior, ranging between 35% (for strategic) and 52% (for
enabling). Therefore, the general consensus about the leader, or the information on the targets’
leadership behavior that is shared between the self and all others, represents the largest source of
information. Although the Arena consistently explains the largest proportion of the variance, it is
noticeably larger for the LVI’s interpersonal leadership dimensions (forceful (46%); enabling
(52%)) compared to the organizational dimensions (strategic (35%); operational (41%)). Except
for forceful behavior, the second-largest share originates from the leader Identity, which are the
unique self-perceptions of the targets. Here, the Identity seems to be larger for the organizational
leadership dimensions –and strategic behavior in particular (31%)– compared to the

THE LARI MODEL 25
interpersonal dimensions (forceful (13%); enabling (20%)). Next to the relevance of the general
Reputation factor, which represents the shared perceptions across all observer rater groups,
Figure 4 also shows the relevance of the three source-specific Reputation factors that include
perceptions that are uniquely shared by raters from the same source. Among these three observer
groups, superiors consistently provide the largest amount of (unique) information across the four
leader behaviors (i.e., 8 to 12%), whereas peers provide the smallest amount of (unique)
information (i.e., 5 to 8%).
When adding up the explained variance of the Arena and the general Reputation factor,
an average percentage of 53.7% is obtained across the four leader behaviors (%= 63.2, 58.2,
45.3, and 48.1 for forceful, enabling, strategic, and operational respectively). The remaining
(average of) 46.3% is represented by unique views of the four rater sources. As such, the shared
views on leader behavior (i.e., shared among all raters, and shared among observers) reveal about
half of the story, while the other half is captured by the unique source-specific perspectives.
-------------------------------------------Insert Figure 4 about here-------------------------------------------
Sample 2
Testing the LARI model. Table 5 summarizes the model fit of all tested models in
Sample 2. Across leadership styles, the LARI model outperformed the other models, with
superior absolute fit indices (CFI, TLI, RMSEA) as well as the lowest DIC values. Next in rank
is the LARI model without general Reputation factor, followed by the two hierarchical models
and the correlated factors model. Consistent with Sample 1, the LARI model without source-
specific Reputation factors fitted the data worst. For most of the leadership styles (5/8), the LARI
model provided a good fit to the data, while the other models did not (i.e., for inspirational,
authoritarian, distrustful, withdrawn, and participative). For example, CFI for inspirational

THE LARI MODEL 26
leadership was .904 for the LARI model, .878 for the LARI model without Reputation, .832 for
both the higher-order models and the correlated-factors model, and .530 for the LARI model
without source-specific Reputation factors. For coaching leadership (1/8), the alternative models
(except for the one without source-specific factors) also fit the data well, but the LARI model
outperformed the other models with higher absolute fit values and a lower DIC. For the
remaining two leadership styles, directive and yielding, none of the models fit the data well.
However, the LARI model did relatively better than the other models (e.g., CFI = .842 and .739
for the LARI and the higher-order model of directive leadership, respectively).
-------------------------------------------Insert Table 5 about here-------------------------------------------
Variance explained by LARI factors. Variance decomposition of the LARI factors is
presented for each of the eight leadership styles (see Figure 5). Consistent with our findings in
the first sample, even for the smallest percentage of explained variance, the 95% credibility
interval did not include zero (i.e., 1.5%; 95% CI 1.0 - 4.3 for Reputation by subordinates in
withdrawn behavior), supporting the idea that all LARI factors capture systematic variance.
Compared to Sample 1, the leadership Arena’s share of explained variance is smaller
(between 12 and 48%), whereas the general Reputation’s share is relatively larger (between 5
and 29%). So a larger share of “blind spots” –on which all observers agree– is captured in the
general Reputation factors of the leadership styles, compared to the four central leader behaviors
in Sample 1. Moreover, more variability was captured by the general Reputation factor, and the
Reputation factors by peers and subordinates in specific. In case Reputation by peers was small
(e.g., distrustful and participative), Reputation by subordinates was relatively larger (and the
other way around). Yet again, the shared perspectives on leadership (i.e., shared among all raters
and among observers) reveal about half of the story, while the other half is captured by unique

THE LARI MODEL 27
source-specific perspectives. Specifically, an average percentage of 44.09% is obtained across
the eight dimensions by adding up the explained variance of the Arena and the general
Reputation factor (% = 35.8 for coaching, 35.6 for inspirational, 51 for directive, 53.7 for
authoritarian, 39.2 for distrustful, 38.5 for withdrawn, 54.2 for yielding, and 44.7 for
participative). The remaining (average of) 55.91% is represented by unique perspectives of the
four rater sources. When evaluating the findings per leadership style, particularly the large Arena
(48%) for the authoritarian leadership style flags up.
The unique self-perceptions of the targets (leader Identity) account for 14 to 21% of the
explained variance in leadership behavior. Compared to Sample 1, in which these percentages
ranged between 13 and 31%, there seems to be less variability with regard to the explained
variance of this Identity factor in Sample 2. Among the three observer groups, superiors
consistently provide the largest share of (unique) information across the eight leadership styles
(i.e., 14 to 27%); a finding that is consistent with the findings in Sample 1. The smallest share of
(unique) information originates from peers in 2/8 styles (distrustful and participative) and from
subordinates for the remaining six styles.
-------------------------------------------Insert Figure 5 about here------------------------------------------
Discussion
Multisource leadership assessments continue to be a popular method for obtaining a well-
rounded analysis of leaders’ strengths and weaknesses (e.g., Day et al., 2014; Slater & Coyle,
2014). In spite of the obvious advantages that multisource perspectives have to offer in terms of
describing and understanding a focal leader, important questions remain as to the existence and
the magnitude of the various source effects in traditional 360s (i.e., self, superiors, peers,
subordinates). Although source effects, in general, have shown to be alive and well (e.g.,

THE LARI MODEL 28
Hoffman et al., 2010; Jackson et al., 2020; Lance et al., 2008), it has proven to be highly
challenging to disentangle the shared and unique perspectives in multisource data. By
investigating each rater source separately, one neglects the possibility that the different rater
sources’ perceptions show overlap (e.g., Conway & Huffcutt, 1997). Alternatively, by combining
all ratings into a single score, one neglects the possibility that the different sources diverge in
their perceptions to an important extent (e.g., Jackson et al. 2020). The consequence is that, at
present, it remains unknown to what extent each rater source offers truly unique information
about a leader (cf. the ‘discrepancy hypothesis’ in 360-research, e.g., Borman, 1997).
Against this background, we argued that exploiting the full potential of multisource
leadership ratings requires a shift in theory and methodology. We proposed the LARI model as a
framework that can be used to conceptualize the shared and unique perspectives in multisource
ratings. Methodologically, we formalized the LARI model using bifactor modeling. The viability
of the LARI model was tested against five alternative models in two samples, each using
different leadership instruments. In general, the LARI models showed a superior fit to the 360
data relative to the alternative models, even when taking into account model parsimony. Further,
variance decompositions of the LARI factors showed that systematic variation is captured by
each of the LARI factors. Especially the non-trivial variance explained by the Reputation factors
further supports the viability of the LARI model. In addition, our findings suggest that the shared
perspectives on leadership (i.e., Arena + general Reputation) reveal about half of the story, while
the other half is captured by unique source-specific perspectives.
Finally, we found tentative evidence for the idea that the relative size of the different
LARI factors depends on the specific leadership dimension being rated. Consistent with the
SOKA model (Vazire, 2010), the observability of the leadership dimension in question might

THE LARI MODEL 29
account for the size of the unique self-perceptions of the leader (Identity) relative to the size of
the shared perceptions on leadership (i.e., Arena + general Reputation). In particular, whereas the
LVI (cf. Sample 1) tapped into leader behaviors, the CLS (cf. Sample 2) covered (interpersonal)
leadership styles. Assuming that concrete behaviors—compared to styles—have a higher
likelihood of being observed (by others), the finding that the explained variance of the shared
perspectives on leadership (Arena + general Reputation) was on average 53.7% in Sample 1 and
44.1% in Sample 2, is consistent with the expectations of the SOKA model. Yet, from an
overarching perspective, strategic and operational leader behaviors (rated in Sample 1) were the
only non-interpersonal leadership dimensions in this study. Especially the large Identity factor
for the strategic behavior dimension might be explained by the fact that this type of behavior
may include cognitions, goals and aspirations that are not necessarily shared nor visible to others
(e.g., “Thinks strategically” or “Anticipates the need to change direction--looks ahead”). In
contrast, the Arena is particularly large for highly observable leadership dimensions (e.g.,
enabling (Sample 1); authoritarian (Sample 2)).
Although a few exceptions could be observed (such as the relatively large % of
subordinates’ unique perceptions for distrustful and participative leadership in Sample 2), the
pattern of results generally does not support the idea of source × dimension interactions (cf.
Guion, 1965; Hiller et al., 2011), such that subordinates would be particularly knowledgeable of
interpersonal dimensions, whereas superiors and peers would be more knowledgeable of
strategic or business-related leadership dimensions. Overall, and consistent with Jackson et al.’s
(2020) findings, no meaningful differences were observed in this regard. For instance,
subordinates’ unique perceptions on interpersonal dimensions like forceful, enabling, coaching,
inspirational, and authoritarian leadership were generally smaller –and not larger– compared to

THE LARI MODEL 30
the unique perceptions of superiors. More notably, among the three observer groups, superiors
consistently provided the largest amount of (unique) information across leadership dimensions
and study samples.
Implications
Our findings have important implications pertaining to leadership (i) theory, (ii) research,
and (iii) practice. As a first key implication, our study results support the viability of the
conceptual framework we introduced. Inspired by recent developments in personality
psychology (McAbee & Connelly, 2016), we conceptualized the shared and unique perspectives
in multisource leadership ratings into the Leadership Arena-Reputation-Identity model. While
the Arena factor is complementary to existing concepts in the leadership domain (e.g., self-other
agreement), the remaining LARI factors, tapping into the unique self- and observer-perceptions,
shed new light on multisource leadership ratings.
Apart from the fact that the LARI model outperformed all of the alternative models, it is
also noteworthy that the alternative model without source-specific factors (Panel C in Figure 3)
fitted the data worst. Combining these findings lends credence to the existence of important
source effects in the data (cf. Jackson et al., 2020). Further supporting this idea, the LARI model
and the associated variance decompositions showed that each of the LARI factors captured
systematic variation. As a set, these findings support the core assumption of 360s that each rater
source (i.e., self, superiors, peers, subordinates) provides unique information with regard to the
target’s behavior (cf. the discrepancy hypothesis). These findings align with the ecological
perspective on multisource ratings (Lance et al., 2008). In contrast to traditional psychometric
theory that treats source effects as systematic bias, the ecological perspective emphasizes the

THE LARI MODEL 31
essential accuracy of perception-based knowledge, recognizing different perceptions as distinct
views on a focal leader that may be equally valid (Lance et al., 2008; Landy & Farr, 1980).
As a third theoretical implication, disentangling the various rater perspectives sheds light
on what constitutes “leadership”. On the one hand, our results are consistent with the idea that
leadership is indeed an attribution based on the leader’s reputation, and therefore lies “in the eye
of the beholder” (cf. Howell & Shamir, 2005; Lord et al., 2020; Nye, 2002). On the other hand,
our results are also congruent with the idea that leadership is characteristic to the leader
themselves. This is reflected in the Arena of the LARI model and accords to what classic trait
psychologist would refer to as “true trait variance”, or judgements which everyone agrees upon
(Funder, 1995). Thus, our findings suggest that leadership is not “merely an attribution” that
people make about an individual (e.g., Haslam et al., 2001; Meindl, 1995). Instead, both the trait
theory of leadership (e.g., Kirkpatrick & Locke, 1991) and the attribution theory of leadership
(e.g., Green & Mitchell, 1979; Martinko et al., 2007) align well with these results.
For research purposes, studies on self-other agreement in leadership could use the LARI
model as a new way to examine the exact composition of shared and unique variation between
rater groups. There has been a long tradition of focusing on the level of agreement between the
leader and others (cf. Fleenor et al., 2010; Lee & Carpenter, 2018). Although different data-
analytic techniques in self-other agreement research (e.g., categories of agreement, polynomial
regression, within and between analysis; WABA) reveal different insights, none of the existing
techniques allows to present a complete picture of what is shared versus unique across the
multisource perspectives. The LARI (bifactor) model could therefore deliver new insights
regarding self-other agreement in leadership, that may accompany the widely accepted methods
(see Fleenor et al., 2010 for an overview). Researchers might consider the use of the Arena

THE LARI MODEL 32
factor, including the perspectives on which everyone agrees, as an indicator of self-other
agreement that could be related to external variables of interest (e.g., leaders’ performance,
subordinates’ job satisfaction). In addition, the remaining LARI factors offer insights on the
shared perspectives among all observers, unique from the leader’s self-report (general
Reputation), and whether (truly) unique information is captured in the perspectives of the four
traditional rater sources.
Further, our results contribute to the ongoing debate on whether self-ratings should be
compared to multiple rater groups seperatly or to all rater groups combined. Both correlation-
based (e.g., Pearson correlations, intraclass correlations (ICCs)), and non-correlation-based (e.g.,
rwg(j) interrater agreement coefficient) techniques have been used to estimate levels of rater
agreement within and between rater groups (Fleenor et al., 2010). In this context, LeBreton et al.
(2003) argued that multisource ratings are generally restricted in range (e.g., due to central
tendency and leniency bias), which in turn attenuates correlations within and between rater
groups. They further argued that, because of these range restrictions, past studies using
correlation-based statistics erroneously concluded that rating similarity between rater groups is
low. By applying correlations, ICCs, and rwg to multisource ratings, LeBreton et al. (2003) found
support for a ‘restriction of variance hypothesis’ as the correlations and ICC’s were relatively
small, whereas the rwg revealed high levels of agreement both within and between rater groups.
In light of these findings, the meaningfulness of the tradition in self-other agreement research of
separately comparing self-ratings to the ratings of superiors, peers, and subordinates has been
questioned (Fleenor et al., 2010). Drawing on our findings, however, we recommend scholars not
to aggregate observer ratings of different rater groups, given that next to the perspective they all
share (i.e., captured in the Arena and the general Reputation factor), they each have valuable

THE LARI MODEL 33
unique views on leadership. Therefore, an “overall” observer score obscures perspectives that are
not shared, i.e., perspectives that might as well be opposites. This recommendation is also in line
with recent insights on the magnitude of source-related effects in MSPRs (Jackson et al., 2020).
Finally, from a practical point of view, knowing how much of the explained variance is
due to self- and the various observer groups might inform us about their usefulness in
multisource leadership ratings. As leader behavior is mostly directed towards subordinates, the
most common source of observer-rated leadership is subordinate-ratings (Hiller et al., 2011).
However, our results reveal that subordinates are usually not the rater group providing the largest
share of unique information. In both our samples, superiors were. Although information
availability might be lower for superiors (Funder, 2012; Pollack & Pollack, 1996), our findings
support the notion that, perceived from their role, superiors evaluate a leader’s functioning in a
unique (and thus different) way compared to peers and subordinates (cf. Atwater & Yammarino,
1997). Interestingly, this was not only the case for leadership dimensions for which superiors
might be especially motivated to keep track (e.g., strategic behaviors), as would be predicted
from social information processing theory (Salancik & Pfeffer, 1978). Instead, this particular
finding transcended specific leadership dimensions or styles in both study samples. One potential
explanation refers to superior’s age and experience in evaluating others, and their vertical access
to effectively evaluate the focal leader (Harris & Kuhnert, 2008). Organizations can approach the
use of our procedures (cf. Analysis code on OSF) in terms of a trade-off between gathering as
much information as possible and keeping the implementation cost (and administrative burden)
as low as possible. For instance, knowing that superiors' unique input is consistently larger
compared to peers and subordinates, and that the latter rater groups provide more overlapping

THE LARI MODEL 34
information regarding leadership behavior, organizations may want to reconsider the use of peer
evaluations in their leadership assessments when cost-saving actions are in place.
Limitations and Future Avenues
Although the LARI model reveals a wealth of information, we did not examine
relationships between the LARI factors and external variables of interest. In other words, we did
not yet test the unique predictive validity of the LARI factors. Therefore, an interesting future
avenue could be to test whether the unique and shared perspectives as offered by multisource
data provide unique predictive power. For instance, from an attributional perspective on
leadership, one can expect that the general Reputation factor adds significantly to the prediction
of promotion decisions, above and beyond the Arena. In this regard, the LARI model could be
particularly useful to further advance attributional theories on leadership. Socioanalytic theory
(Hogan, 1996), for instance, generally uses people’s reputation to predict important life
outcomes (e.g., performance, career success), because the best predictor of future behavior is
past behavior, and because reputations reflect a person's past behavior (Hogan & Blickle, 2018).
The LARI model could be used to test this core assumption of socioanalytic theory (Hogan,
1996). One could further delve into this matter by testing whether unique perceptions (e.g., from
superiors) are particularly relevant in this regard. Further, a substantial body of research has
shown that specific leader behaviors relate to leader effectiveness (e.g., Judge et al., 2004;
Vergauwe et al., 2017). Consistent with research on self-other agreement (Fleenor et al., 2010), a
positive relationship could be expected between the Arena and leader effectiveness. In contrast,
the leader Identity might negatively predict effectiveness, as self-enhancing tendencies (that are
part of the Identity) have been linked to lower performance (Connelly & Hülsheger, 2012). Note
that objective effectiveness criteria are probably preferred in this context, given that subjective

THE LARI MODEL 35
indicators like multisource ratings of leader effectiveness can also be decomposed in shared and
unique rater perspectives on the criterion side.
Although we considered cross-validating the LARI model using different samples and
leadership instruments to be critical to this study, allowing us to strengthen our conclusions
regarding the viability of the presented LARI framework, it remains an open question whether
some of the inconsistencies between the results were due to these differences. The general
Reputation factor was, for instance, generally smaller in Sample 1 compared to Sample 2. It is
possible that the level of blind spots was lower in Sample 1 because (more overt) leader
behaviors were measured compared to (more abstract) leadership styles in Sample 2, or because
of the TLTM rating format, which allows raters –and thus the leaders themselves – to reflect on
their behaviors on a broader (and maybe deeper) level (cf. Vergauwe et al., 2017). Future
research can further explore the conditions that influence the relative size of the different LARI
factors.
Finally, the LARI factors can be expected to be dynamic, in the sense that leader
identities as well as the perceptions others have about a leaders can change over time (Lord et al.,
2020). Indeed, both leader identities as well as observers’ perceptions may develop through
ongoing interactions at work (e.g., leader-follower interactions) and changing contexts (e.g.,
financial crises). Understanding the complex ways in which leaders’ self-definitions and
observers’ perceptions develop, change, and are influenced by interactions and contexts, could
provide unique insights on the drivers of leader behaviors and actions (Epitropaki et al., 2017). In
this context, examining the LARI model using longitudinal multisource leadership data could
provide a wealth of information.

THE LARI MODEL 36
Conclusion
In the present paper, we proposed the LARI model, a model that allows disentangling of
the shared and unique perspectives in multisource leadership ratings. Cross-validating empirical
tests of the LARI model across two large samples and across leadership instruments supports this
novel framework's viability. Our study results are promising, and we hope that they enthuse
other scholars to explore the further possibilities of the LARI framework.

THE LARI MODEL 37
References
Allen, T. D., Barnard, S., Rush, M. C., & Russell, J. E. A. (2000). Ratings of organizational
citizenship behavior: Does the source make a difference? Human Resource Management
Review, 10, 97–114. doi: 10.1016/S1053-48229900041-8.
Andersen, S. M. (1984). Self-knowledge and social inference: II. The diagnosticity of
cognitive/affective and behavioral data. Journal of Personality and Social Psychology, 46,
294–307. doi: 10.1037/0022-3514.46.2.294
Aramovich, N. P., & Blankenship, J. R. (2020). The relative importance of participative versus
decisive behavior in predicting stakeholders’ perceptions of leader effectiveness. The
Leadership Quarterly, 31(5), 101387. doi: 10.1016/j.leaqua.2020.101387
Atwater, L. E., & Brett, J. F. (2006). 360-Degree feedback to leaders: Does it relate to changes
in employee attitudes? Group & Organization Management, 31(5), 578–600. doi:
1059601106286887
Atwater, L. E., Ostroff, C., Yammarino, F. J., & Fleenor, J. W. (1998). Self–other agreement:
Does it really matter? Personnel Psychology, 51(3), 577–598. doi:10.1111/j.1744-
6570.1998.tb00252.x
Atwater, L. E., Wang, M., Smither, J. W., & Fleenor, J. W. (2009). Are cultural characteristics
associated with the relationship between self and others’ ratings of leadership? Journal of
Applied Psychology, 94(4), 876–886. doi:10.1037/a0014561
Atwater, L. E., & Yammarino, F. J. (1992). Does self–other agreement on leadership
perceptions moderate the validity of leadership and performance predictions?
Personnel Psychology, 45, 141−164. doi:10.1111/j.1744-6570.1992.tb00848.x
Atwater, L. E., & Yammarino, F. J. (1997). Self–other rating agreement: A review and model.
In G. R. Ferris (Ed.), Research in personnel and human resources management, 15. (pp.
121−174). JAI Press.
Blumer, H. (1986). Symbolic interactionism: Perspective and method. Berkeley, CA: University
of California Press.
Borman, W. C. (1997). 360 degree ratings: An analysis of assumptions and a research agenda
for evaluating their validity. Human Resource Management Review, 7(3), 299-316. doi:
10.1016/S1053-4822(97)90010-3

THE LARI MODEL 38
Braddy, P. W., Gooty, J., Fleenor, J. W., & Yammarino, F. J. (2014). Leader behaviors and
career derailment potential: A multi-analytic method examination of rating source
and self–other agreement. The Leadership Quarterly, 25, 373–390. doi:
10.1016/j.leaqua.2013.10.001.
Brislin, R. W. (1970). Back-translation for cross-cultural research. Journal of Cross-Cultural
Psychology, 1, 185–216.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the
multitrait-multimethod matrix. Psychological Bulletin, 56, 81–105. doi:
10.1037/h0046016
Chen, F., Curran, P. J., Bollen, K. A., Kirby, J., & Paxton, P. (2008). An empirical evaluation of
the use of fixed cutoff points in RMSEA test statistic in structural equation models.
Sociological Methods and Research, 36, 462–494. doi: 10.1177/0049124108314720
Connelly, B. S., & Hülsheger, U. R. (2012). A narrower scope or a clearer lens for personality?
Examining sources of observers’ advantages over self-reports for predicting performance.
Journal of Personality, 80, 603–631.
Connelly, B. S., & Ones, D. S. (2010). An other perspective on personality: Meta-analytic
integration of observers’ accuracy and predictive validity. Psychological Bulletin, 136,
1092–1122. doi: 10.1037/a0021212
Connolly, J. J., Kavanagh, E. J., & Viswesvaran, C. (2007). The Convergent Validity between
Self and Observer Ratings of Personality: A meta-analytic review. International Journal
of Selection and Assessment, 15(1), 110–117. doi:10.1111/j.1468-2389.2007.00371.x
Conway, J. M., & Huffcutt, A. I. (1997). Psychometric properties of multisource performance
ratings: A meta-analysis of subordinate, supervisor, peer, and self-ratings. Human
Performance, 10(4), 331-360. doi: 10.1207/s15327043hup1004_2
Conway, J. M., Lombardo, K., & Sanders, K. C. (2001). A meta-analysis of incremental validity
and nomological networks for subordinate and peer rating. Human Performance, 14(4),
267–303. doi: 10.1207/S15327043HUP1404_1
Day, D. V., Fleenor, J. W., Atwater, L. E., Sturm, R. E., & McKee, R. A. (2014). Advances in
leader and leadership development: A review of 25 years of research and theory. The
Leadership Quarterly, 25, 63-82. doi: 10.1016/j.leaqua.2013.11.004
DeNisi, A. S., Cafferty, T. P., & Meglino, B. M. (1984). A cognitive view of the performance

THE LARI MODEL 39
appraisal process: A model and research propositions. Organizational Behavior
and Human Performance, 33(3), 360–396. doi: 10.1016/0030-5073(84)90029-1
Epitropaki, O., Kark, R., Mainemelis, C., & Lord, R. G. (2017). Leadership, followership and
identity processes: a multilevel review. The Leadership Quarterly, 28, 104–29. doi:
10.1016/j.leaqua.2016.10.003
Fleenor, J. W., Smither, J. W., Atwater, L. E., Braddy, P. W., & Sturm, R. E. (2010). Self-other
rating agreement in leadership: A review. The Leadership Quarterly, 21, 1005-1034.
doi:10.1016/j.leaqua.2010.10.006
Foti, R. J., Fraser, S. L., & Lord, R. G. (1982). Effects of leadership labels and prototypes on
perceptions of political leaders. Journal of Applied Psychology, 67(3), 326–333. doi:
10.1037/0021-9010.67.3.326
Funder, D. C. (1995). On the accuracy of personality judgment: A realistic approach.
Psychological Review, 102(4), 652. doi: 10.1037/0033-295X.102.4.652
Funder, D. C. (2012). Accurate personality judgment. Current Directions in Psychological
Science, 21, 177–182. doi: 10.1177/0963721412445309
Guion, R. (1965). Personnel testing. New York, NY: McGraw-Hill.
Green, S.G., Mitchell, T.R. (1979). Attributional processes of leaders in leader–member
interactions. Organizational Behavior and Human Performance, 23, 429-458. doi:
10.1016/0030-5073(79)90008-4
Hansbrough, T. K., Lord, R. G., & Schyns, B. (2015). Reconsidering the accuracy of follower
leadership ratings. The Leadership Quarterly, 26(2), 220–237. doi:
10.1016/j.leaqua.2014.11.006
Harris, L. S., & Kuhnert, K. W. (2008). Looking through the lens of leadership: A constructive
developmental approach. Leadership and Organization Development Journal, 29, 47–67.
doi: 10.1108/01437730810845298
Harris, M. M., & Schaubroeck, J. (1988). A meta-analysis of self-supervisor, self-peer, and
peer-supervisor ratings. Personnel Psychology, 41, 43−62. doi: 10.1111/j.1744-
6570.1988.tb00631.x
Haslam, S. A., Platow, M. J., Turner, J. C., Reynolds, K. J., McGarty, C., Oakes, P. J., Johnson,
S., Ryan, M. K., & Veenstra, K. (2001). Social identity and the romance of leadership:

THE LARI MODEL 40
The importance of being seen to be ‘doing it for us’. Group Processes & Intergroup
Relations, 4(3), 191-205. doi: 10.1177/1368430201004003002
Hiller, N. J., DeChurch, L. A., Murase, T., & Doty, D. (2011). Searching for outcomes of
leadership: A 25-year review. Journal of Management, 37, 1137–1177. doi:
org/10.1177/0149206310393520.
Hoffman, B., Lance, C. E., Bynum, B., & Gentry, W. A. (2010). Rater source effects are alive
and well after all. Personnel Psychology. 63(1), 119-151. doi: 10.1111/j.1744-
6570.2009.01164.x
Hogan, R. (1996). A socioanalytic interpretation of the Five-Factor Model. In J. S. Wiggins
(Ed.), The Five-Factor model of personality (pp.163–179). New York, NY: Guilford
Press.
Hogan, R., & Blickle, G. (2018). Socioanalytic theory: Basic concepts, supporting evidence and
practical implications. In V. Zeigler-Hill & T. K. Shackelford (Eds.), The SAGE handbook
of personality and individual differences: The science of personality and individual
differences (pp. 110–129). Sage Reference. doi: 10.4135/9781526451163.n5
Hooijberg, R., & Choi, J. (2000). Which leadership roles matter to whom? An examination
of rater effects on perceptions of effectiveness. The Leadership Quarterly, 11, 341–364.
doi: 10.1016/S1048-9843(00)00044-8
Howell, J. M., & Shamir, B. (2005). The role of followers in the charismatic leadership process:
Relationships and their consequences. The Academy of Management Review, 30, 96–112.
doi: 10.5465/AMR.2005.15281435
Jackson, D. J. R., Michaelides, G., Dewberry, C., Schwencke, B., & Toms, S. (2020). The
implications of unconfounding multisource performance ratings. Journal of Applied
Psychology, 105(3), 312–329. doi: 10.1037/apl0000434
James, L. R., Demaree, R. G., & Wolf, G. (1984). Estimating within-group interrater reliability
with and without response bias. Journal of Applied Psychology, 69, 85–98. doi:
10.1037/0021-9010.69.1.85
Judge, Piccolo, & Ilies, (2004). The forgotten ones? The validity of consideration and initiating
structure in leadership research. Journal of Applied Psychology, 89, 36-51. doi:
10.1037/0021-9010.89.1.36

THE LARI MODEL 41
Kaiser, R. B., LeBreton, J. M., & Hogan, J. (2015). The dark side of personality and extreme
leader behavior. Applied Psychology: In International Review, 64, 55-92. doi:
10.1111/apps.12024
Kaiser, R. B., Overfield, D. V., & Kaplan, R. E. (2010). Leadership Versatility Index version 3.0
facilitator’s guide. Greensboro, NC: Kaplan DeVries Inc.
Kirkpatrick, S. & Locke, E. (1991). Leadership: Do traits matter? Academy of Management
Perspectives, 5(2), 48-60. doi: 10.5465/ame.1991.4274679
Lance, C. E., Hoffman, B. J., Gentry, W. A., Baranik, L. E. (2008). Rater source factors
represent important subcomponents of the criterion construct space, not rater bias. Human
Resource Management Review, 18, 223–232. doi: 10.1016/j.hrmr.2008.03.002
Landy, F. J., & Farr, J. L. (1980). Performance rating. Psychological Bulletin, 87, 72−107. doi:
10.1037/0033-2909.87.1.72
Leary, M. R., & Kowalski, R. M. (1990). Impression management: A, literature review and two-
component model. Psychological Bulletin, 107, 34-47. doi: 10.1037/0033-2909.107.1.34
LeBreton, J. M., Burgess, J. D., Kaiser, R. B., Atchley, E. K., James, L. R. (2003). The
restriction of variance hypothesis and interrater reliability and agreement: Are ratings from
multiple sources really dissimilar? Organizational Research Methods, 6, 80–128.
doi:10.1177/1094428102239427
LeBreton, J. M., & Senter, J. L. (2008). Answers to 20 questions about interrater reliability and
interrater agreement. Organizational Research Methods, 11, 815–852. doi:
10.1177/1094428106296642
Lee, A., & Carpenter, N. C. (2018). Seeing eye to eye: A meta-analysis of self-other agreement
of leadership. The Leadership Quarterly, 29(2), 253–275. doi:
10.1016/j.leaqua.2017.06.002
Lord, R. G., Epitropaki, O., Foti, R. J., & Hansbrough, T. K. (2020). Implicit leadership
theories, implicit follower theories, and dynamic processing of leadership information.
Annual Review of Organizational Psychology and Organizational Behavior, 7, 49-74. doi:
10.1146/annurev-orgpsych-012119-045434
Lord, R. G., Foti, R. J., & de Vader, C. L. (1984). A test of leadership categorization theory:
Internal structure, information processing, and leadership perceptions. Organizational
Behavior & Human Performance, 34(3), 343–378. doi:10.1016/0030-5073(84)90043-6

THE LARI MODEL 42
Luft, J., & Ingham, H. (1955). The Johari window: A graphic model of interpersonal awareness.
Paper presented at the Proceedings of the Western Training Laboratory in Group
Development, LA, CA: UCLA Extension Office.
Martinko, M. J., Harvey, P., & Douglas, S. C. (2007). The role, function, and contribution of
attribution theory to leadership: A review. The Leadership Quarterly, 18(6), 561–585. doi:
10.1016/j.leaqua.2007.09.004
McAbee, S. T., & Connelly, B. S. (2016). A multi-rater framework for studying personality: The
Trait-Reputation-Identity model. Psychological Review, 123, 569-591. doi:
10.1037/rev0000035
Meindl, J. R. (1995). The romance of leadership as a follower-centric theory: A social
constructionist approach. The Leadership Quarterly, 329-341. doi: 10.1016/1048-
9843(95)90012-8
Mount, M. K. (1984). Supervisor, self and subordinate ratings of performance and satisfaction
with supervision. Journal of Management, 10(3), 305-320. doi:
10.1177/014920638401000304
Mount, M. K., Judge, T. A., Scullen, S. E., Sytsma, M. R., & Hezlett, S. A. (1998). Trait, rater
and level effects in 360-degree performance ratings. Personnel Psychology, 51(3), 557–
576. doi: 10.1111/j.1744-6570.1998.tb00251.x
Murphy, K. R., Cleveland, J. N., Skattebo, A. L., & Kinney, T. B. (2004). Raters who pursue
different goals give different ratings. Journal of Applied Psychology, 89, 158-164. doi:
10.1037/0021-9010.89.1.158
Muthén, L. K., & Muthén, B. O. (1998-2017). Mplus User’s Guide. Eighth Edition. Los
Angeles, CA: Muthén & Muthén.
Nye, J. L. (2002). The eye of the follower - Information processing effects on attributions
regarding leaders of small groups. Small Group Research, 33(3), 337-360.
doi:10.1177/10496402033003003
Olsen, J. A., & Kenny, D. A. (2006). Structural equation modeling with interchangeable dyads.
Psychological Methods, 11, 127. doi: 10.1037/1082-989X.11.2.127
Paulhus, D. L., & Vazire, S. (2007). The self-report method. In R. W. Robins, R. C. Fraley, & R.
Krueger (Eds.), Handbook of research methods in personality psychology (pp. 224–239).
New York, NY: Guilford Press.

THE LARI MODEL 43
Pollack, D. M., & Pollack, L. J. (1996). Using 360 degree feedback in performance appraisal.
Public Personnel Management, 25, 507–528. doi: 10.1177/009102609602500410
Redeker, M., de Vries, R. E., Rouckhout, D., Vermeren, P., & De Fruyt, F. (2014). Integrating
leadership: The leadership circumplex. European Journal of Work and Organizational
Psychology, 23, 435-455. Doi: 10.1080/1359432X.2012.738671
Reise, S. P. (2012). The rediscovery of bifactor measurement models. Multivariate Behavioral
Research, 47(5), 667-696. doi: 10.1080/00273171.2012.715555
Rothstein, H. R. (1990). Interrater reliability of job performance ratings: Growth to asymptote
level with increasing opportunity to observe. Journal of Applied Psychology, 75(3), 322–
327. doi: 10.1037/0021-9010.75.3.322
Salancik, G. R., & Pfeffer, J. (1978). A social information processing approach to job attitudes
and task design. Administrative Science Quarterly, 23(2), 224–253. doi: 10.2307/2392563
Scullen, S. E., Mount, M. K., & Goff, M. (2000). Understanding the latent structure of job
performance ratings. Journal of Applied Psychology, 85(6), 956–970. doi: 10.1037/0021-
9010.85.6.956
Slater, R. & Coyle, A. (2014). The governing of the self/the self-governing self: Multi-
rater/source feedback and practices 1940-2011. Theory & Psychology, 24(2), 233-255.
doi: 10.1177/0959354313520087
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & van der Linde, A. (2014). The deviance
information criterion: 12 years on. Journal of the Royal Statistical Society: Series B
(Statistical Methodology), 76(3), 485–493. doi:10.1111/rssb.12062
Tornow, W. (1993). Perceptions or reality: Is multi-perspective measurement a means or an
end? Human Resource Management, 32, 221−230. doi: 10.1002/hrm.3930320203
Vazire, S. (2010). Who knows what about a person? The Self-Other Knowledge Asymmetry
(SOKA) Model. Journal of Personality and Social Psychology, 98, 281-300. doi:
10.1037/a0017908
Vergauwe, J., Wille, B., Hofmans, J., Kaiser, R. B., & De Fruyt, F. (2017). The "too little/too
much" scale: A new rating format for detecting curvilinear effects. Organizational
Research Methods, 20, 518-544. doi:10.1177/1094428117706534
Vergauwe, J., Wille, B., Hofmans, J., Kaiser, R. B., & De Fruyt, F. (2018). The double-edged
sword of leader charisma: Understanding the curvilinear relationship between charismatic

THE LARI MODEL 44
personality and leader effectiveness. Journal of Personality and Social Psychology, 114,
110-130. doi: 10.1037/pspp0000147
Viswesvaran, C., Schmidt, F. L., & Ones, D. S. (2005). Is there a general factor in ratings of job
performance? A meta-analytic framework for disentangling substantive and error
influences. Journal of Applied Psychology, 90(1), 108–131. doi:10.1037/0021-
9010.90.1.108
Woehr, D. J., Sheehan, M. K., Bennett, W. (2005). Assessing measurement equivalence across
ratings sources: A multitrait-multirater approach. Journal of Applied Psychology,
90, 592–600. doi: 10.1037/0021-9010.90.3.592
Yammarino, F. J. (2003). Modern data analytic techniques for multisource feedback.
Organizational Research Methods, 6, 6−14. doi: 10.1177/1094428102239423
Yukl, G. A. (2010). Leadership in organizations. Upper Saddle River, NJ: Prentice Hall.

THE LARI MODEL 45
Table 1
Conceptual Meaning of the Six Factors of the Leadership Arena-Reputation-Identity (LARI) Model and Elements Contributing to the LARI Factors
LARI factor
Conceptual meaning
Elements contributing to the factor
Leadership
Arena
Information about the leader’s behavior that everyone
agrees upon, including the leader themselves,
supervisors, peers, and subordinates
 Overt leader behavior congruent with a leader’s self-reports.
 The extent to which the leader conveys his/her desired reputation
both to others and in his/her self-description
 The leader’s level of self-awareness
 Observers’ motivation and ability to provide accurate ratings of the
leader
 The extent to which observers are exposed to the leader
 The extent to which the particular behaviors are observable
General
Reputation
Information about the leader’s behavior that is shared
by all observer groups but is unique from the leader’s
self-report
 Overt leader behavior that the leader in unaware of (“e.g., “blind
spots”)
 Overt leader behavior that is intentionally not shared by the leader,
but picked up by all observers (e.g., impression management by the
leader)
 Perceptions of the leader communicated about in the broad social
network of subordinates, peers and supervisors
 Systematic bias shared across observers (e.g., gender or physical
appearance stereotypes, leniency bias, halo effect)
Reputation by
subordinates
Information about the leader’s behavior that is unique
in the perspective provided by subordinates (and not
shared with the views provided by supervisors, peers,
or the leader themselves)
 Overt leader behavior visible only to subordinates (e.g., “managing
down”), and not represented in leaders’ self-reports
 Communication about the leader among subordinates
 Leniency bias among subordinates due to strategic responding
Reputation by
peers
Information about the leader’s behavior that is unique
in the perspective provided by peers (and not shared
with the views provided by supervisors, subordinates,
or the leader themselves)
 Overt leader behavior visible only to peers, and not represented in
leaders’ self-reports
 Communication about the leader among peers
 Friendship bias
Reputation by
supervisors
Information about the leader’s behavior that is unique
in the perspective provided by supervisors (and not
shared with the views provided by subordinates, peers,
or the leader themselves)
 Overt leader behavior visible only to supervisors (e.g., “managing
up”), and not represented in leaders’ self-reports
 Communication about the leader among supervisors
 Leniency bias

THE LARI MODEL 46
Table 1 (continued)
Leader Identity
Information about the leader’s behavior that is unique
in the perspective provided by the leader themselves
(and not shared with the views provided by the three
observer groups)
 Accurate leader behaviors or attitudes that are low in visibility and
remain private to the self
 Inaccurate self-perceptions of a leader’s behavior (e.g., self-
enhancement)
 Strategic/intentional self-presentation tactics that don’t overlap with
observers’ perspectives

THE LARI MODEL 47
Table 2
Descriptive Statistics and Variable Intercorrelations: Sample 1 (N = 537 target leaders)
M
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
1.Sexa
-
-
-
2.Age
45.59
7.97
.00
-
3.Forceful: self
.09
.61
-.03
.00
.82
4.Forceful: superiors
-.14
.56
.04
-.06
.41**
.89
5.Forceful: peers
-.03
.52
.03
.00
.39**
.61**
.90
6.Forceful: subordinates
.01
.43
.05
-.01
.44**
.48**
.56**
.87
7.Enabling: self
-.06
.46
-.01
-.02
-.38**
-.32**
-.40**
-.39**
.73
8.Enabling: superiors
-.11
.37
.08
.02
-.33**
-.62**
-.51**
-.40**
.29**
.84
9.Enabling: peers
-.22
.39
-.01
-.03
-.35**
-.48**
-.68**
-.47**
.39**
.56**
.87
10.Enabling: subordinates
-.25
.33
.00
-.03
-.33**
-.34**
-.39**
-.61**
.36**
.46**
.50**
.87
11.Strategic: self
-.16
.56
-.17**
-.10†
34**
.17**
.13*
.08
.11*
-.14*
-.10†
-.06
.84
12.Strategic: superiors
-.34
.40
-.02
-.13*
.14*
.39**
.21**
.18**
-.14*
-.13*
-.08
-.02
.30**
.86
13.Strategic: peers
-.28
.36
-.07
-.14*
.13*
.28**
.33**
.17**
-.08
-.09†
.03
-.03
.25**
.41**
.88
14.Strategic: subordinates
-.19
.30
-.03
-.16**
.19**
.28**
.29**
.39**
-.11†
-.07
-.11†
.05
.27**
.41**
.48**
.88
15.Operational: self
-.14
.45
.04
.02
.10†
-.06
-.05
.05
.08
.00
-.03
-.05
-.24**
-.22**
-.25**
-.19**
.69
16.Operational: superiors
-.06
.30
.08
.02
-.03
.02
.02
.06
-.03
.04
-.05
-.05
-.23**
-.17**
-.22**
-.15**
.31**
.72
17.Operational: peers
-.09
.28
.04
.06
-.02
-.05
.06
.06
-.02
.02
-.04
-.04
-.22**
-.18**
-.29**
-.17**
.32**
.43**
.75
18.Operational: subordinates
-.13
.25
.10†
.06
.03
-.05
-.02
.24**
-.10†
-.02
.03
-.04
-.29**
-.17**
-.20**
-.20**
.32**
.31**
.37**
.70
Note. Leader behaviors are rated on a scale ranging between -4 (much too little), 0 (the right amount), and +4 (much too much); Means, standard deviations and
correlations are based on N = 537 target leaders; Bold values on the diagonal show the internal consistencies (Cronbach alphas) of the relevant variables, which are
based on N = 537 for target leaders’ self-reports, N = 1142 for superior-reports, N = 2695 for peer-reports, and N = 3500 for subordinate-reports; aSex is dummy
coded such that 0 = male and 1 = female; †p < .05, *p < .01, **p < .001.

THE LARI MODEL 48
Table 3
Descriptive Statistics and Variable Intercorrelations: Sample 2 (N = 1255 target leaders)
M
SD
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
1.Sexa
-
-
-
2.Age
44.17
8.80
-.14*
-
3.Coaching: self
3.11
.35
-.01
.17**
.86
4.Coaching: superiors
2.93
.40
.18**
-.06
.21**
.91
5.Coaching: peers
2.89
.34
.11*
-.02
.18**
.37**
.92
6.Coaching: subordinates
2.95
.37
.07†
-.08†
.19**
.30**
.36**
.93
7.Inspirational: self
2.87
.41
-.08†
.23**
.64**
.10**
.03
.07†
.87
8.Inspirational: superiors
2.74
.46
.13**
-.08†
.07*
.66**
.19**
.16**
.23**
.91
9.Inspirational: peers
2.80
.36
.08†
.00
.08*
.24**
.65**
.21**
.21**
.44**
.90
10.Inspirational: subordinates
2.80
.37
.05
-.03
.10**
.22**
.20**
.72**
.26**
.38**
.42**
.90
11.Directive: self
2.19
.43
-.14**
.06
.16**
-.06†
-.12**
-.05
.53**
.13**
.11**
.20**
.74
12.Directive: superiors
2.10
.45
.08†
-.03
-.03
.11**
-.07†
-.02
.25**
.63**
.34**
.34**
.32**
.80
13.Directive: peers
2.18
.37
.00
-.03
-.04
-.05
.00
-.07†
.24**
.34**
.59**
.32**
.35**
.56**
.80
14.Directive: subordinates
2.12
.35
-.04
.02
-.07†
-.07†
-.13**
-.01
.25**
.26**
.29**
.53**
.43**
.50**
.58**
.78
15.Authoritarian: self
1.39
.43
-.06
.03
-.38**
-.17**
-.26**
-.19**
.05
.10**
.07†
.10**
.48**
.32**
.34**
.41**
.81
16.Authoritarian: superiors
1.36
.46
-.01
.12*
-.15**
-.46**
-.27**
-.18**
.12**
.11**
.15**
.14**
.23**
.61**
.47**
.41**
.39**
.87
17.Authoritarian: peers
1.41
.42
-.06
.09†
-.12**
-.26**
-.56**
-.25**
.19**
.15**
.05
.15**
.30**
.46**
.66**
.50**
.42**
.57**
18.Authoritarian: subordinates
1.30
.41
-.05
.12*
-.15**
-.21**
-.31**
-.59**
.16**
.14**
.11**
-.05
.31**
.41**
.47**
.66**
.46**
.49**
19.Distrustful: self
.96
.33
.06
-.09†
-.61**
-.16**
-.21**
-.19**
-.34**
.00
-.02
-.04
.15**
.15**
.18**
.21**
.65**
.23**
20.Distrustful: superiors
.98
.39
-.08†
.15**
-.09*
-.72**
-.30**
-.20**
.05
-.37**
-.08*
-.05
.14**
.19**
.22**
.19**
.22**
.70**
21.Distrustful: peers
1.03
.34
-.07†
.09†
-.10*
-.30**
-78**
-.29**
.10**
-.04
-.38**
-.05
.15**
.25**
.31**
.28**
.29**
.41**
22.Distrustful: subordinates
.94
.35
-.05
.12**
-.14**
-.25**
-.34**
-.82**
.06†
-.03
-.08*
-.45**
.18**
.20**
.26**
.33**
.30**
.33**
23.Withdrawn: self
1.02
.38
.08†
-.14**
-.52**
-.13**
-.04
-.12**
-.74**
-.24**
-.21**
-.29**
-.42**
-.23**
-.25**
-.26**
.06†
-.08*
24.Withdrawn: superiors
1.07
.45
-.14**
.13**
-.03
-.58**
-.18**
-.16**
-.17**
-.83**
-.40**
-.34**
-.14**
-.54**
-.31**
-.25**
-.11**
.01
25.Withdrawn: peers
1.03
.35
-.09†
.04
-.06†
-.24**
-.61**
-.22**
-.15**
-.40**
-.86**
-.39**
-.11**
-.28**
-.48**
-.26**
-.06†
-.08*
26.Withdrawn: subordinates
.98
.33
-.05
.04
-.09*
-.23**
-.25**
-.72**
-.17**
-.32**
-.39**
-.86**
-.12**
-.24**
-.25**
-.37**
-.04
-.05
27.Yielding: self
1.63
.38
.13**
-.08†
-.09*
-.02
.09*
.03
-.48**
-.27**
-.22**
-.24**
-.40**
-.33**
-.35**
-.32**
-.26**
-.24**
28.Yielding: superiors
1.58
.43
.04
.11*
.06†
-.10**
.07†
.02
-.21**
-.60**
-.35**
-.31**
-.23**
-.63**
-.46**
-.41**
-.29**
-.42**
29.Yielding: peers
1.51
.35
.08†
.02
.04
.01
.01
.04
-.23**
-.34**
-.56**
-.33**
-.23**
-.43**
-.63**
-.46**
-.30**
-.37**
30.Yielding: subordinates
1.49
.28
.05
-.02
.06
.01
.08*
.06†
-.25**
-.34**
-.35**
-.45**
-.26**
-.44**
-.48**
-.62**
-.32**
-.36**
31.Participative: self
2.72
.32
.11*
.01
.58**
.14**
.17**
.13**
.16**
-.08*
-.09*
-.09*
-.20**
-.22**
-.25**
-.27**
-.52**
-.25**
32.Participative: superiors
2.59
.35
.11*
-.04
.18**
.65**
.29**
20**
-.07*
.10**
-.06†
-.08*
-.22**
-.38**
-.35**
-.34**
-.34**
-.69**
33.Participative: peers
2.54
.31
.12*
-.05
.15**
.29**
.74**
.28**
-.14**
-.07†
.15**
-.09*
-.28**
-.35**
-.42**
-.42**
-.39**
-.45**
34.Participative: subordinates
2.65
.30
.09†
-.11*
.18**
.24**
.33**
.81**
-.09*
-.05
-.01
.33**
-.22**
-.26**
-.33**
-.39**
-.36**
-.35**

THE LARI MODEL 49
Table 3 (continued)
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
17.Authoritarian: peers
.89
18.Authoritarian: subordinates
.58**
.90
19.Distrustful: self
.25**
.26**
.78
20.Distrustful: superiors
.38**
.31**
.15**
.87
21.Distrustful: peers
.79**
.42**
.22**
.38**
.90
22.Distrustful: subordinates
.42**
.81**
.24**
.28**
.40**
.91
23.Withdrawn: self
-.15**
-.12**
.45**
.01
-.06†
.01
.82
24.Withdrawn: superiors
-.12**
-.09*
-.01
.49**
.09*
.06†
.23**
.89
25.Withdrawn: peers
.05
-.07†
.01
.13**
.50**
.12**
.18**
.42**
.89
26.Withdrawn: subordinates
-.04
.21**
.07†
.11**
.15**
.61**
.25**
.34**
.41**
.88
27.Yielding: self
-.33**
-.28**
.08*
-.10**
-.20**
-.14**
.59**
.25**
.17**
.16**
.78
28.Yielding: superiors
-.40**
-.34**
-.12**
-.04
-.20**
-.18**
.22**
.64**
.32**
.23**
.40**
.84
29.Yielding: peers
-.49**
-.38**
-.14**
-.16**
-.16**
-.20**
.22**
.32**
.56**
.26**
.42**
.56**
.83
30.Yielding: subordinates
-.44**
-.50**
-.14**
-.15**
-.21**
-.21**
.25**
.29**
.30**
.43**
.42**
.51**
.58**
.75
31.Participative: self
-.26**
-.28**
-.49**
-.14**
-.16**
-.16**
-.05
.10**
.08*
.06†
.35**
.21**
.22**
.25**
.76
32.Participative: superiors
-.46**
-.38**
-.21**
-.64**
-.35**
-.28**
.06†
-.07†
.03
.02
.23**
.42**
.32**
.32**
.29**
.84
33.Participative: peers
-.77**
-.49**
-.24**
-.34**
-.71**
-.37**
.13**
.07†
-.13**
.01
.31**
.34**
.47**
.38**
.30**
.44**
.86
34.Participative: subordinates
-.45**
-.81**
-.24**
-.25**
-.35**
-.82**
.06†
.02
-.01
-.37**
.22**
.24**
.29**
.43**
.28**
.34**
.45**
.88
Note. Means, standard deviations and correlations are based on N = 1255 target leaders; Bold values on the diagonal show the internal consistencies (Cronbach
alphas) of the relevant variables, which are based on N = 1255 for target leaders’ self-reports, N = 2175 for superior-reports, N = 5068 for peer-reports, and N =
8534 for subordinate-reports; aSex is dummy coded such that 0 = male and 1 = female; †p < .05, *p < .01, **p < .001.

THE LARI MODEL 50
Table 4
Model Characteristics of the LARI Model compared to Alternative Models in Sample 1 (N = 537 target leaders)
Model
Free
parameters
2.5% PP
limit
97.5%
PP limit
PP p-
value
CFI
TLI
RMSEA
RMSEA
90% CI
DIC
Forceful
LARI model
75
-14.861
56.612
.133
.993
.976
.047
.017
.068
8927.914
LARI model w/o G-Reputation
66
1.377
71.734
.024
.989
.969
.053
.039
.067
8937.837
LARI model w/o S-Reputation
55
662.573
734.413
<.001
.783
.597
.191
.189
.194
9587.641
Higher-order model
58
35.702
103.277
<.001
.978
.956
.063
.056
.072
8962.204
Higher-order with G-Reputation
58
34.586
106.447
<.001
.978
.955
.064
.056
.073
8963.244
Correlated factors model
60
24.024
97.353
<.001
.980
.958
.062
.053
.071
8957.106
Enabling
LARI model
75
-36.251
37.309
.518
1.000
1.000
.000
.000
.024
7101.204
LARI model w/o G-Reputation
66
-36.495
33.868
.512
1.000
1.000
.000
.000
.041
7150.892
LARI model w/o S-Reputation
55
250.829
318.703
<.001
.858
.736
.123
.119
.127
7426.435
Higher-order model
58
-23.658
45.871
.254
.994
.989
.025
.000
.041
7153.820
Higher-order with G-Reputation
58
-24.401
47.139
.244
.994
.988
.026
.000
.042
7154.433
Correlated factors model
60
-23.756
43.405
.259
.995
.988
.026
.000
.044
7156.796
Strategic
LARI model
75
-38.185
37.565
.523
1.000
1.000
.000
.000
.049
5937.181
LARI model w/o G-Reputation
66
-19.593
51.519
.191
.993
.981
.036
.000
.054
5946.525
LARI model w/o S-Reputation
55
678.188
747.467
<.001
.713
.470
.193
.191
.195
6631.464
Higher-order model
58
-14.889
54.288
.129
.992
.984
.034
.016
.047
5941.319
Higher-order with G-Reputation
58
-17.832
53.178
.137
.992
.984
.034
.013
.047
5940.315
Correlated factors model
60
-17.347
52.242
.147
.993
.985
.033
.006
.048
5941.439

THE LARI MODEL 51
Table 4 (continued)
Model
Free
parameters
2.5% PP
limit
97.5%
PP limit
PP p-
value
CFI
TLI
RMSEA
RMSEA
90% CI
DIC
Operational
LARI model
75
-23.199
48.038
.244
.990
.986
.023
.000
.038
6089.060
LARI model w/o G-Reputation
66
-23.422
46.621
.241
.989
.973
.031
.000
.048
6108.971
LARI model w/o S-Reputation
55
191.877
261.980
<.001
.819
.675
.107
.103
.112
6312.123
Higher-order model
58
-21.184
50.432
.210
.989
.978
.028
.000
.043
6102.529
Higher-order with G-Reputation
58
-23.147
47.444
.225
.990
.980
.027
.000
.043
6100.808
Correlated factors model
60
-22.082
47.688
.225
.989
.978
.028
.000
.045
6103.902
Note. w/o G-Reputation = without General Reputation; w/o S-Reputation = without Source-specific Reputation (i.e., Reputation by superiors, peers,
and subordinates); PP p-value = Posterior Predictive P-Value; PP limits represent the 95% confidence intervals for the difference between the
observed and the replicated Chi-Square values; N = 537 target leaders, N = 1142 superiors, N = 2695 peers, and N = 3500 subordinates.

THE LARI MODEL 52
Table 5
Model Characteristics of the LARI Model compared to Alternative Models in Sample 2 (N = 1255 target leaders)
Model
Free
parameters
2.5% PP
limit
97.5% PP
limit
PP p-
value
CFI
TLI
RMSEA
RMSEA
90% CI
DIC
Coaching
LARI model
375
2292.883
2554.724
<.001
.939
.929
.036
.036
.036
79539.677
LARI model w/o G-Reputation
330
2972.577
3224.931
<.001
.923
.912
.040
.040
.040
80164.897
LARI model w/o S-Reputation
271
18408.982
18671.134
<.001
.541
.499
.096
.095
.096
95547.546
Higher-order model
274
3905.733
4169.916
<.001
.900
.890
.045
.045
.045
81045.831
Higher-order with G-Reputation
274
3905.486
4161.453
<.001
.900
.890
.045
.045
.045
81043.023
Correlated factors model
276
3902.707
4166.861
<.001
.900
.890
.045
.045
.045
81048.336
Inspirational
LARI model
375
3671.524
3958.234
<.001
.904
.889
.045
.045
.045
99699.931
LARI model w/o G-Reputation
330
4732.049
4986.164
<.001
.878
.862
.050
.050
.050
100701.987
LARI model w/o S-Reputation
271
18714.598
18973.046
<.001
.530
.486
.096
.096
.096
114631.159
Higher-order model
274
6583.061
6847.375
<.001
.832
.816
.058
.057
.058
102508.039
Higher-order with G-Reputation
274
6585.827
6845.620
<.001
.832
.816
.058
.057
.058
102503.972
Correlated factors model
276
6583.455
6838.245
<.001
.832
.816
.058
.058
.058
102505.655
Directive
LARI model
300
3720.810
3941.524
<.001
.842
.808
.058
.057
.058
104153.366
LARI model w/o G-Reputation
264
4883.256
5110.729
<.001
.794
.759
.064
.064
.065
105286.041
LARI model w/o S-Reputation
217
9816.332
10029.988
<.001
.593
.544
.089
.089
.089
110161.325
Higher-order model
220
6244.608
6465.375
<.001
.739
.707
.071
.071
.071
106599.955
Higher-order with G-Reputation
220
6246.422
6468.524
<.001
.739
.706
.071
.071
.071
106601.215
Correlated factors model
222
6221.549
6440.460
<.001
.740
.707
.071
.071
.071
106575.084

THE LARI MODEL 53
Table 5 (continued)
Model
Free
parameters
2.5% PP
limit
97.5% PP
limit
PP p-
value
CFI
TLI
RMSEA
RMSEA
90% CI
DIC
Authoritarian
LARI model
375
2948.975
3209.815
<.001
.920
.907
.040
.040
.041
120625.284
LARI model w/o G-Reputation
330
3713.572
3965.244
<.001
.900
.887
.044
.044
.045
121343.903
LARI model w/o S-Reputation
271
13241.532
13502.219
<.001
.656
.624
.081
.081
.081
130817.487
Higher-order model
274
4086.224
4342.102
<.001
.891
.881
.046
.046
.046
121663.811
Higher-order with G-Reputation
274
4082.003
4339.747
<.001
.891
.881
.046
.045
.046
121659.990
Correlated factors model
276
4078.712
4335.855
<.001
.891
.881
.046
.046
.046
121658.642
Distrustful
LARI model
375
1973.340
2236.164
<.001
.929
.917
.033
.033
.034
101502.184
LARI model w/o G-Reputation
330
2530.880
2760.219
<.001
.912
.900
.037
.037
.037
101984.550
LARI model w/o S-Reputation
271
13049.381
13310.091
<.001
.561
.521
.081
.081
.081
112479.941
Higher-order model
274
3228.835
3483.319
<.001
.887
.877
.041
.041
.041
102657.065
Higher-order with G-Reputation
274
3225.880
3479.258
<.001
.888
.877
.041
.041
.041
102652.720
Correlated factors model
276
3220.018
3477.773
<.001
.888
.877
.041
.041
.041
102654.718
Withdrawn
LARI model
375
2242.166
2505.567
<.001
.922
.916
.034
.034
.034
107280.255
LARI model w/o G-Reputation
330
2779.826
3028.565
<.001
.905
.892
.039
.038
.039
107904.987
LARI model w/o S-Reputation
271
13381.326
13641.747
<.001
.561
.520
.082
.081
.082
118451.746
Higher-order model
274
3182.082
3448.127
<.001
.891
.881
.041
.040
.041
108256.610
Higher-order with G-Reputation
274
3182.322
3439.083
<.001
.891
.881
.041
.040
.041
108253.187
Correlated factors model
276
3172.859
3434.033
<.001
.892
.881
.041
.040
.041
108249.158

THE LARI MODEL 54
Table 5 (continued)
Model
Free
parameters
2.5% PP
limit
97.5% PP
limit
PP p-
value
CFI
TLI
RMSEA
RMSEA
90% CI
DIC
Yielding
LARI model
375
4776.711
5029.464
<.001
.823
.793
.051
.051
.051
122120.154
LARI model w/o G-Reputation
330
6184.928
6439.976
<.001
.772
.742
.057
.057
.057
123488.869
LARI model w/o S-Reputation
271
11491.757
11752.704
<.001
.582
.543
.076
.076
.076
128742.209
Higher-order model
274
7504.916
7759.778
<.001
.725
.699
.061
.061
.062
124752.455
Higher-order with G-Reputation
274
7499.794
7762.419
<.001
.725
.700
.061
.061
.062
124750.533
Correlated factors model
276
7502.024
7764.264
<.001
.725
.699
.062
.061
.062
124756.592
Participative
LARI model
350
1828.411
2064.027
<.001
.923
.909
.035
.034
.035
86712.048
LARI model w/o G-Reputation
308
2491.768
2735.815
<.001
.896
.881
.040
.039
.040
87344.392
LARI model w/o S-Reputation
253
9924.605
10166.639
<.001
.604
.564
.076
.076
.076
94717.464
Higher-order model
256
2870.386
3111.919
<.001
.881
.869
.042
.041
.042
87670.015
Higher-order with G-Reputation
256
2872.318
3114.201
<.001
.881
.869
.042
.041
.042
87674.006
Correlated factors model
258
2868.996
3110.010
<.001
.881
.869
.042
.041
.042
87670.577
Note. w/o G-Reputation = without General Reputation; w/o S-Reputation = without Source-specific Reputation (i.e., Reputation by superiors, peers,
and subordinates); PP p-value = Posterior Predictive P-Value; PP limits represent the 95% confidence intervals for the difference between the
observed and the replicated Chi-Square values; N = 1255 target leaders, N = 2175 superiors, N = 5068 peers, and N = 8534 subordinates.

THE LARI MODEL 55
Figure 1. The Johari window (Luft & Ingham, 1955) with LARI-labels in parentheses (Left), and the Trait-Reputation-Identity (TRI)
Model (Right), adapted from “A multi-rater framework for studying personality: The Trait-Reputation-Identity model” by S. T. McAbee and
B. S. Connelly, 2016, Psychological Review, 123(5), p. 570. Copyright 2016 by the American Psychological Association.

THE LARI MODEL 56
Figure 2. The Leadership Arena-Reputation-Identity (LARI) model.
Note. s = self, b = subordinates, p = peers, r = superiors in the item subscripts.

THE LARI MODEL 57
Figure 3. Panel A: The LARI model. Panel B: The LARI model without general Reputation.

THE LARI MODEL 58
Panel C: The LARI model without source-specific Reputation factors. Panel D: Higher-order factor model.

THE LARI MODEL 59
Par
Panel E: Higher-order factor model with Panel F: Correlated-factors model.
general Reputation.

THE LARI MODEL 60
Figure 4. Proportion of the explained variance accounted for by the LARI factors in Sample 1 (Leadership Versatility Index; Kaiser et
al., 2010).
0,456 0,518
0,346 0,408
0,176 0,064
0,107
0,073
0,132 0,196
0,309 0,257
0,113 0,084 0,117
0,094
0,051 0,065 0,057
0,08
0,08 0,08 0,07 0,081
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
forceful enabling strategic operational
leadership arena general reputation leader identity (self) reputation superiors reputation peers reputation subordinates

THE LARI MODEL 61
Figure 5. Proportion of the explained variance accounted for by the LARI factors in Sample 2 (Circumplex Leadership Scan; Redeker
et al., 2014).
0,124 0,12
0,221
0,483
0,176
0,237
0,383
0,219
0,234 0,236
0,289
0,054
0,216 0,148
0,159
0,228
0,193 0,212
0,206 0,142
0,165 0,18
0,162
0,173
0,241 0,233
0,139
0,153
0,216
0,266
0,171
0,201
0,171 0,14 0,087 0,088
0,039
0,15 0,098
0,021
0,038 0,057 0,056 0,08
0,187
0,016 0,026
0,156
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
coaching inspirational directive authoritarian distrustful withdrawn yielding participative
leadership arena general reputation leader identity (self) reputation superiors reputation peers reputation subordinates

THE LARI MODEL 62
Appendix
Table A
Inter-rater Agreement (rwg(j)) for the Leader Behavior Subscales (Leadership Versatility Index) in
Sample 1
Subordinates
Peers
Superiors
Forceful
Takes charge
.84
.83
.88
Declares themselves
.86
.83
.89
Pushes performance
.84
.82
.89
Enabling
Empowerment
.86
.85
.90
Participation
.88
.87
.92
Support
.87
.89
.93
Strategic
Direction
.88
.87
.90
Growth
.86
.85
.90
Innovation
.92
.90
.94
Operational
Execution
.87
.87
.93
Efficiency
.91
.90
.93
Order
.91
.90
.94
Mean
.88
.87
.91
SD
.03
.03
.02
Agreement
strong
strong
very strong
N for k = min. 2
530
523
297
Note. To account for central tendency bias (most ratings ranged between -2 and +2 on the scale ranging
from -4 to +4), a triangular null distribution was used in the computation of the IRA (see LeBreton &
Senter, 2008); Level of agreement according to LeBreton and Senter (2008).

THE LARI MODEL 63
Table B
CFA Models Testing Measurement Invariance across Languages (Dutch, French, and English) in
Sample 2
Χ2
df
p
CFI
RMSEA
Coaching
1. Invariance of the factor structure (i.e, configural invariance)
6595.69
270
<.001
.933
.067
2. Invariance of the factor loadings (i.e., metric invariance)
6698.23
298
<.001
.932
.064
3. Invariance of the item intercepts (i.e., scalar invariance)
7346.15
326
<.001
.926
.064
Inspirational
1. Invariance of the factor structure (i.e, configural invariance)
14669.35
270
<.001
.822
.101
2. Invariance of the factor loadings (i.e., metric invariance)
15089.23
298
<.001
.817
.097
3. Invariance of the item intercepts (i.e., scalar invariance)
16695.23
326
<.001
.797
.098
3a. Partial invariance of the item intercepts – item 15Dutch free
16275.72
325
<.001
.802
.097
3b. Partial invariance of the item intercepts – item 6French free
16009.13
324
<.001
.806
.096
3c. Partial invariance of the item intercepts – item 12French free
15766.41
323
<.001
.809
.095
Directive
1. Invariance of the factor structure (i.e, configural invariance)
13046.25
162
<.001
.667
.123
2. Invariance of the factor loadings (i.e., metric invariance)
13083.62
184
<.001
.667
.115
3. Invariance of the item intercepts (i.e., scalar invariance)
14835.28
206
<.001
.622
.116
3a. Partial invariance of the item intercepts – item 109French free
14373.42
205
<.001
.634
.115
3b. Partial invariance of the item intercepts – item 105Dutch free
14035.04
204
<.001
.643
.114
3c. Partial invariance of the item intercepts – item 107Dutch free
13761.34
203
<.001
.650
.113
3d. Partial invariance of the item intercepts – item 110Dutch free
13636.55
202
<.001
.653
.112
3e. Partial invariance of the item intercepts – item 112French free
13470.44
201
<.001
.658
.112
Authoritarian
1. Invariance of the factor structure (i.e, configural invariance)
7763.66
270
<.001
.903
.073
2. Invariance of the factor loadings (i.e., metric invariance)
8051.17
298
<.001
.900
.070
3. Invariance of the item intercepts (i.e., scalar invariance)
9546.95
326
<.001
.881
.073
3a. Partial invariance of the item intercepts – item 90French free
9061.54
325
<001
.887
.071
3b. Partial invariance of the item intercepts – item 104French free
8713.80
324
<.001
.892
.070
Distrustful
1. Invariance of the factor structure (i.e, configural invariance)
5758.85
270
<.001
.919
.062
2. Invariance of the factor loadings (i.e., metric invariance)
5877.17
298
<.001
.917
.060
3. Invariance of the item intercepts (i.e., scalar invariance)
6940.15
326
<.001
.902
.062
3a. Partial invariance of the item intercepts – item 88French free
6540.32
325
<.001
.908
.060
Withdrawn
1. Invariance of the factor structure (i.e, configural invariance)
5435.28
270
<.001
.911
.060
2. Invariance of the factor loadings (i.e., metric invariance)
5641.99
298
<.001
.907
.058
3. Invariance of the item intercepts (i.e., scalar invariance)
6812.97
326
<.001
.888
.062
3a. Partial invariance of the item intercepts – item 63Dutch free
6431.83
325
<.001
.894
.060
3b. Partial invariance of the item intercepts – item 62French free
6256.34
324
<.001
.897
.059
Yielding
1. Invariance of the factor structure (i.e, configural invariance)
16638.51
270
<.001
.631
.107
2. Invariance of the factor loadings (i.e., metric invariance)
16716.29
298
<.001
.630
.102
3. Invariance of the item intercepts (i.e., scalar invariance)
19447.19
326
<.001
.569
.106
3a. Partial invariance of the item intercepts – item 57Dutch free
18699.06
325
<.001
.586
.104
3b. Partial invariance of the item intercepts – item 52French free
18003.40
324
<.001
.602
.102
3c. Partial invariance of the item intercepts – item 51Dutch free
17531.43
323
<.001
.612
.101
3d. Partial invariance of the item intercepts – item 59French free
17309.25
322
<.001
.617
.100
3e. Partial invariance of the item intercepts – item 51French free
17151.62
321
<.001
.621
.100

THE LARI MODEL 64
Table B (continued)
Participative
1. Invariance of the factor structure (i.e, configural invariance)
5722.53
231
<.001
.905
.067
2. Invariance of the factor loadings (i.e., metric invariance)
5774.59
257
<.001
.905
.064
3. Invariance of the item intercepts (i.e., scalar invariance)
7425.98
283
<.001
.877
.069
3a. Partial invariance of the item intercepts – item 32French free
6944.23
282
<.001
.885
.067
3b. Partial invariance of the item intercepts – item 36French free
6540.03
281
<.001
.892
.065
3c. Partial invariance of the item intercepts – item 39French free
6381.99
280
<.001
.895
.064
Note. CFI refers to the Comparative Fit Index, and RMSEA to the Root Mean Square Error of Approximation. The
analyses were performed using the MLR estimator in Mplus 8.4 (Muthén & Muthén, 1998-2017) on the observer-
ratings (N = 15,777). Of those observers, 14,119 (89.5%) completed the questionnaire in Dutch, 1,439 (9.1%) in
French, and 219 (1.4%) in English. For each leadership style of the Circumplex Leadership Scan (CLS; Redeker et
al., 2014) the change in fit indices was acceptable after imposing equality constraints on the factor loadings (i.e., <
.015 for ∆RMSEA and < .010 for ∆CFI), implying that full metric invariance across languages holds. Additional
equality constraints on the item intercepts revealed that for Coaching full scalar invariance held, while for the other
CLS leadership styles partial scalar invariance could be established. For some dimensions (i.e., directive and
yielding), the fit indices for the configural invariance model suggest that a one-factor model does not fit the data
well. Although such model fit would be unacceptable in many cases, it is less of an issue for this particular study
because we are interested in how the proposed LARI model fits the data relative to alternative ways of modeling the
multisource data.

THE LARI MODEL 65
Table C
Inter-rater Agreement (rwg(j)) for the Leader Behavior Items (Circumplex Leadership Scan) in
Sample 2
Subordinates
Peers
Superiors
Coaching
Coaching item 1
.72
.76
.81
Coaching item 2
.66
.74
.77
Coaching item 3
.67
.73
.76
Coaching item 4
.57
.69
.74
Coaching item 5
.69
.75
.77
Coaching item 6
.61
.68
.74
Coaching item 7
.62
.67
.73
Coaching item 8
.68
.73
.77
Coaching item 9
.68
.73
.77
Coaching item 10
.59
.68
.74
Coaching item 11
.64
.73
.73
Coaching item 12
.66
.72
.77
Coaching item 13
.62
.69
.73
Coaching item 14
.67
.75
.78
Coaching item 15
.66
.74
.78
Inspiring
Inspiring item 1
.44
.63
.68
Inspiring item 2
.58
.71
.76
Inspiring item 3
.48
.67
.69
Inspiring item 4
.68
.72
.77
Inspiring item 5
.59
.69
.72
Inspiring item 6
.60
.69
.70
Inspiring item 7
.62
.72
.77
Inspiring item 8
.62
.68
.69
Inspiring item 9
.60
.67
.72
Inspiring item 10
.73
.72
.74
Inspiring item 11
.72
.73
.73
Inspiring item 12
.65
.69
.71
Inspiring item 13
.60
.66
.69
Inspiring item 14
.58
.72
.75
Inspiring item 15
.55
.63
.67
Directive
Directive item 1
.61
.68
.71
Directive item 2
.53
.58
.66
Directive item 3
.58
.66
.71
Directive item 4
.50
.60
.66
Directive item 5
.46
.55
.63
Directive item 6
.55
.62
.67
Directive item 7
.64
.70
.72
Directive item 8
.64
.69
.70
Directive item 9
.60
.66
.69

THE LARI MODEL 66
Table C (continued)
Directive item 10
.34
.46
.54
Directive item 11
.47
.59
.63
Directive item 12
.58
.64
.68
Authoritarian
Authoritarian item 1
.61
.63
.68
Authoritarian item 2
.59
.61
.70
Authoritarian item 3
.52
.59
.66
Authoritarian item 4
.54
.61
.68
Authoritarian item 5
.50
.60
.67
Authoritarian item 6
.62
.70
.73
Authoritarian item 7
.46
.56
.65
Authoritarian item 8
.49
.59
.64
Authoritarian item 9
.54
.61
.64
Authoritarian item 10
.49
.58
.64
Authoritarian item 11
.46
.56
.63
Authoritarian item 12
.67
.71
.75
Authoritarian item 13
.55
.59
.70
Authoritarian item 14
.52
.62
.67
Authoritarian item 15
.57
.63
.69
Distrustful
Distrustful item 1
.62
.68
.72
Distrustful item 2
.60
.68
.76
Distrustful item 3
.58
.64
.70
Distrustful item 4
.67
.71
.76
Distrustful item 5
.56
.61
.70
Distrustful item 6
.71
.71
.75
Distrustful item 7
.71
.71
.79
Distrustful item 8
.66
.70
.74
Distrustful item 9
.68
.69
.74
Distrustful item 10
.64
.70
.73
Distrustful item 11
.59
.63
.69
Distrustful item 12
.52
.66
.71
Distrustful item 13
.59
.66
.70
Distrustful item 14
.63
.68
.73
Distrustful item 15
.59
.67
.72
Withdrawn
Withdrawn item 1
.60
.66
.71
Withdrawn item 2
.64
.69
.75
Withdrawn item 3
.64
.68
.74
Withdrawn item 4
.55
.64
.69
Withdrawn item 5
.66
.72
.74
Withdrawn item 6
.63
.69
.71
Withdrawn item 7
.57
.67
.73
Withdrawn item 8
.68
.71
.72
Withdrawn item 9
.60
.65
.68
Withdrawn item 10
.59
.64
.71
Withdrawn item 11
.62
.68
.70

THE LARI MODEL 67
Table C (continued)
Withdrawn item 12
.64
.69
.71
Withdrawn item 13
.65
.68
.70
Withdrawn item 14
.53
.65
.67
Withdrawn item 15
.66
.70
.76
Yielding
Yielding item 1
.76
.76
.76
Yielding item 2
.57
.64
.66
Yielding item 3
.47
.58
.61
Yielding item 4
.57
.64
.66
Yielding item 5
.56
.63
.68
Yielding item 6
.66
.70
.73
Yielding item 7
.45
.53
.59
Yielding item 8
.52
.59
.65
Yielding item 9
.63
.69
.69
Yielding item 10
.67
.69
.69
Yielding item 11
.57
.62
.66
Yielding item 12
.49
.59
.63
Yielding item 13
.67
.71
.76
Yielding item 14
.61
.68
.70
Yielding item 15
.50
.62
.66
Participative
Participative item 1
.67
.70
.73
Participative item 2
.65
.68
.72
Participative item 3
.72
.76
.80
Participative item 4
.75
.80
.81
Participative item 5
.65
.70
.73
Participative item 6
.56
.61
.65
Participative item 7
.79
.82
.83
Participative item 8
.62
.66
.67
Participative item 9
.64
.68
.69
Participative item 10
.60
.68
.74
Participative item 11
.70
.75
.78
Participative item 12
.70
.74
.77
Participative item 13
.64
.70
.74
Participative item 14
.58
.68
.72
Mean
.60
.67
.71
SD
.08
.06
.05
Agreement
moderate
moderate
strong
N for k = min. 2
1198
1102
543
Note. To control for a slight skew (most ratings ranged between 1 and 4 on the 0 to 4 scale), the slightly
skewed random response null distribution was used in the computation of the IRA (see LeBreton &
Senter, 2008); Level of agreement according to LeBreton and Senter (2008).