Abstract
Leader–member exchange differentiation (LMXD) refers to the degree to which leaders develop varying quality relationships with team members. Due to a plague of mixed empirical results and misalignment in the LMXD literature, Buengeler et al. (2021) proposes a conceptual framework that suggests there are different types of LMXD—separation, variety, and disparity—each having distinct effects on team outcomes. This framework has yet to be empirically tested, so we conduct two studies to examine its utility. In Study 1, we conduct a constructive reproduction of Cobb and Lau (2015), reanalyzing their data with LMX disparity (instead of separation) to better align with justice-based theorizing. In Study 2, we extend Kim et al. (2023) by testing whether LMX separation predicts team potency and performance more effectively than team median LMX. Across both studies, LMXD explained incremental variance beyond central tendency scores and showed negative effects on team outcomes. However, LMX disparity outperformed LMX separation—even when theoretically misaligned—raising concerns about the empirical distinctiveness of LMXD types. Our findings support the conceptual value of differentiating LMXD types and theory-measurement alignment, but we highlight challenges in operationalizing them.
Team dynamics in organizations involve complex social settings, given the individual differences of each person and the unique relational dynamics among each dyad in the group. One area where this complexity is particularly salient is in the study of relationship quality among leaders and followers. Leadership scholars have long acknowledged that varying levels of relationship quality between leaders and their followers is an inevitable reality of leadership in group settings (Dansereau et al., 1975). Accordingly, the study of leader–member exchange (LMX) differentiation (LMXD), or the variability in relationship quality between a leader and team members within a single team (Yu et al., 2018), has been the focus of group-level leadership research for some time.
Scholars believed that understanding LMXD would clarify and strengthen LMX theory at the team level, but instead, LMXD research has left scholars conflicted, as paradoxical findings have been revealed (Li & Liao, 2014). Indeed, reviews indicate that LMXD research is contradictory, with logic and empirical evidence for both the benefits and the dangers of differentiation to teams (Buengeler et al., 2021; Liden et al., 2006; Martin et al., 2018). For example, on one hand, high levels of LMXD have been argued to harm teams because leaders having high-quality relationships with a few group members may incite competition and compromise cohesion (e.g., Hooper & Martin, 2008; Stewart & Johnson, 2009). On the other hand, a high level of differentiation has been suggested to benefit teams because it allows leaders to accommodate followers’ unique needs and skills (e.g., Henderson et al., 2009). Indeed, empirical research has found that LMXD is associated with both greater group effectiveness and worse group effectiveness (see Yu et al., 2018 for a meta-analytic review).
To address the issues of contradictory findings in the LMXD literature, Buengeler et al. (2021) proposed a theoretical framework that suggests the configuration of differentiation—not merely its degree—shapes team outcomes. Their framework categorizes LMXD into three distinct types: LMX separation, LMX variety, and LMX disparity. They argue that overlooking these configurations of LMXD as well as misalignment of theory, measurement, and outcomes may be responsible for the conflicting findings in the literature. While this framework offers a promising solution to the inconsistencies in LMXD research, it has not been tested empirically. Consequently, it is not yet clear whether aligning the LMXD types with their measurement can address the inconsistencies in the literature that the framework purports to address.
Therefore, the purpose of the current research is to empirically test the utility of the Buengeler et al. (2021) framework. Leadership scholars have recently emphasized the importance of theory refinement and theory testing to create a more cohesive understanding of leadership (Brouer et al., 2026). Replications, reproductions, and generalizability studies have recently been advocated as effective ways to accomplish theory testing and refinement (Köhler & Cortina, 2023; Kraimer et al., 2023). Indeed, Kraimer et al. (2023) suggest that one of the most direct ways to contribute to theory is through the empirical testing of existing theoretical models. Accordingly, to facilitate theory testing and enhance our current knowledge of LMXD, the current research tests Buengeler et al.'s (2021) theoretical framework across two empirical studies. In Study 1, we conduct a constructive reproduction of Cobb and Lau (2015), which was identified by Buengeler et al. (2021) as a study with theoretical and operational misalignment. In contrast to the original Cobb and Lau (2015) analyses, our constructive reproduction employs LMX Disparity measurement (i.e., operationalizing LMXD using the coefficient of variation (CV)), as prescribed by Buengeler et al.'s framework. In Study 2, we conduct a direct test of a portion of Buengeler et al.'s framework using an existing dataset from Kim et al. (2022). Notably, the original focus of Kim et al.'s (2022) paper was on LMX Mean (LMXM). However, we use the primary data from this study to posit new hypotheses. Specifically, we focus on LMXD instead of LMXM to understand the effects of LMX Separation on group outcomes. Together, our two studies allow for a robust empirical test of the Buengeler et al. framework, given our ability to focus on multiple LMXD configuration types across multiple samples.
The results of our research contribute to the leadership literature in three ways. First, we provide the first empirical test of Buengeler et al.'s (2021) conceptual LMXD framework by applying it to two samples in which we align the LMXD index with the theorizing and outcomes of the original study. As explained by Reiche and Schaffer (2024), it is crucial to balance the development of new theories with rigorous theory testing to build a cohesive body of scientific knowledge. Our test of the Buengeler et al. framework allowed us to examine whether the framework has the potential to address prior and prevent future inconsistencies. Our findings reveal that while LMXD types (e.g., separation and disparity) are conceptually distinct and show differential effects in certain contexts, they also tend to yield similar empirical patterns—particularly when variability is low or central tendency is strong. This suggests that the empirical distinctiveness of LMX types may be context-dependent, warranting further refinement and testing across diverse samples and outcomes.
Second, our research opens the door to future LMXD scholars to explore alternative approaches to clarifying LMX dispersion and configurations. Our study indicates that while the current indices provided by Buengeler et al. capture aspects of LMXD configurations, they may lack the fidelity needed to fully account for the nuanced impacts of LMXD on team dynamics. Our findings invite the development of new measures or methods that better capture the complexity of LMXD and its role in team functioning. Our work contributes to the broader goal of resolving contradictory findings in LMXD research and enhancing the theoretical and empirical coherence of leadership studies.
Third, our study helps refine the broader issues within foundational LMX theory, beyond LMXD specifically. Some scholars say there should be a moratorium on LMX research and others say it needs to be completely re-theorized (e.g., Gooty et al., 2012; Gottfredson et al., 2020). These issues stem from a more large-scale problem of clearly defining and measuring the construct at hand. Our study combats this issue by aligning theory with proper constructs, and their respective index, using Buengeler et al.'s framework. This helps us better understand the limitations of LMX theory at the group level by putting Buengeler et al.'s conceptual framework into practice and modeling how to use it for future research. When scholars align their research based on clear definitions and measurements, our future research will become more cohesive, advancing science.
Buengeler et al.’s Leader–Member Exchange Differentiation Framework
Buengeler et al. (2021) propose a comprehensive framework that reconceptualizes LMXD by drawing from the group diversity literature to advocate for the study of LMXD configurations. Whereas extant research on LMXD tends to explore the mean level of differentiation in groups—comparing groups with leaders who treat group members differently (i.e., high LMXD) to those with leaders who hold similar relationships with all group members (i.e., low LMXD)—Buengeler et al. suggest that focusing on the degree of differentiation ignores the pattern of LMX differences in a group and the meaning that those patterns of differentiation imply. For example, in a group where one individual has a very high-quality relationship with the leader and the remaining followers have low-quality relationships with the leader, the degree of differentiation is modest, yet the configuration has significant implications for team functioning. Accordingly, Buengeler et al. identify three distinct types of LMXD to capture and specify both the meaning and the configuration of LMX relationships in groups, which they conceptualize as LMX separation, LMX variety, and LMX disparity. They further advocate for the alignment of each LMXD type with respective theories, outcomes, and measurement properties to combat inconsistent findings. For the comprehensive framework of LMXD constructs, theories, predicted group outcomes, and recommended measurement indices, please refer to Buengeler et al.'s Table 1 in their original article. For a summary of this information and how it relates to our current studies, please see our table in OSF. 1
LMX separation refers to the spread of LMX quality, capturing the extent to which a team is divided into subgroups. That is, groups high in LMX separation would be characterized by subgroups with very high LMX (“in-group”) and very low LMX (“out-group”). In contrast, a group with low levels of LMX separation would be characterized by similar quality relationships among all leader–follower dyads. To capture LMX separation, statistical dispersion indices, such as standard deviation (SD), should be employed as they are maximized when subgroups form. Further, LMX separation is aligned with theories that explain in-group and out-group formation, including similarity-attraction and social identity theory. Social categorization theory and balance theory are also invoked for LMX separation because they tend to explain its associated negative outcomes such as reduced cohesion and increased conflict (Buengeler et al., 2021).
LMX variety reflects the diversity or range of LMX relationships within a team, capturing qualitative or categorical heterogeneity in relationship types (e.g., mentoring vs. transactional). Maximum LMX variety occurs when each group member holds a distinct type of relationship with the leader, whereas minimum variety reflects uniformity across all members. Because LMX variety is inherently categorical, Buengeler et al. (2021) argue that current LMX survey measures—typically continuous—are inadequate to meaningfully capture LMX variety. Indices that best represent variety, such as Blau's index or Teachman's entropy index, require categorical inputs, making their use with standard LMX scales problematic. If one were to attempt to operationalize LMX variety with a measure of LMX and a dispersion index, the survey items would need to be categorized in some way, introducing error which may not accurately capture LMX variety. It is for this reason that Buengeler et al. (2021, p. 281) “advocate for a measure of perceived LMX variety… [to] assess LMXD as a function of each member's contribution, role, or knowledge.” Theoretical perspectives aligned with variety include requisite variety, variation–selection–retention processes, and information-processing perspectives (e.g., Hinsz et al., 1997), all of which emphasize the value of diverse inputs for group functioning. Human and social capital theories (e.g., Becker, 1964) also predict that unique member expertise can enhance group performance. Accordingly, LMX variety is the only LMXD type that is expected to be positively related to group outcomes such as decision quality, creativity, and complex task performance.
LMX disparity refers to the inequality in LMX relationships within the team, with high-quality LMX relationships concentrated in one or very few team members. High disparity indicates an asymmetrical distribution in which a small minority receives a disproportionate share of relational resources—such as leader attention, support, and opportunities—while the majority remains at the lower end of the LMX continuum. Because LMX disparity captures inequality, it is best measured by indices that are sensitive to unequal distributions, most notably the CV, which accounts for the group mean and is maximized when one member's LMX is far higher than all others. Because LMX disparity is associated with a disparate spread of a valued resource, theories that capture inequality, injustice, and relative deprivation are often invoked for LMX disparity. In the context of high-LMX disparity, group members often feel unfairly treated, leading to various negative group outcomes such as reduced motivation, negative affect, and perceptions of injustice, particularly among low-LMX members.
As mentioned, the central argument posited by Buengeler et al. (2021) is that previous research has often treated LMXD too simplistically by failing to explore LMXD types, namely separation, variety, and disparity. Additionally, the authors call for alignment, arguing that the form of dispersion (i.e., LMXD type) must dictate the theoretical mechanisms and measurement approach. Misalignment—such as using SD (an index of separation) while invoking justice theories (which align with disparity)—can yield inconsistent or misleading findings. Such theoretical and empirical disconnects can result in false conclusions about how and why LMXD influences team outcomes. Indeed, their systematic review revealed that such misalignment is pervasive in the literature and may account for the field's contradictory results. Therefore, the Buengeler et al.'s LMXD framework serves as the theoretical foundation for the current paper. Specifically, we employ two studies to test the implications of aligning LMXD types with their associated operationalizations, construct definitions, theories, and group outcomes as suggested by Buengeler et al.'s framework.
The Current Research
This paper presents two studies designed to empirically evaluate the theoretical framework proposed by Buengeler et al. (2021). Across both studies, we aim to test the central premise of their work: that LMXD types are meaningful conceptualizations, and the LMXD types must be conceptualized and measured in theoretically consistent ways (i.e., aligned) to yield meaningful insights. Using existing datasets, 2 we adopt two complementary approaches to examine the framework across different samples, measures, and theoretical emphases, focusing on LMX disparity in Study 1 and LMX separation in Study 2.
Study 1 is a constructive reproduction, which Kraimer et al. (2023) define as a form of theory testing that reanalyzes original data but introduces theoretically justified improvements in measurement, modeling, or testing procedures. Accordingly, in Study 1, we reanalyzed Cobb and Lau's (2015) dataset, replacing their original separation-based index of LMXD (SD) with a disparity-based measure (CV) to better align with their theorizing. This constructive change reflects the core premise of the Buengeler et al. framework: the type of LMXD invoked by theory should guide measurement choice. Therefore, this study allows us to test whether measuring LMX disparity—which can capture hierarchical resource concentration, unlike LMX separation—clarifies or alters conclusions about the negative effects predicted by justice and stratification-based theories in Cobb and Lau's original article. In Study 2, we test a portion of the Buengeler et al. (2021) theoretical model (Kraimer et al., 2023), using secondary data originally published in Kim et al. (2022). While the original study focused on team median LMX and peer mentoring as predictors of team potency and performance, we repurpose the dataset—with permission—to evaluate new hypotheses derived from Buengeler et al.'s framework. Specifically, we test whether LMX separation negatively impacts team potency and, in turn, team performance. In contrast to Study 1, which focused on disparity and justice-based outcomes, Study 2 targets group effectiveness outcomes theoretically tied to separation-based mechanisms such as social categorization and team fragmentation. By controlling for team median LMX (the original independent variable in Kim et al.), we directly assess whether LMX separation offers added predictive value. Additionally, we compare LMX separation to disparity to test whether alignment following Buengeler et al.'s framework yields stronger predictive validity.
Together, these two studies offer a complementary and robust test of Buengeler et al.'s (2021) theory-measurement alignment framework for LMXD. Study 1 evaluates whether applying a theoretically appropriate measurement of LMXD (disparity instead of separation) to an existing study leads to potentially different conclusions. Study 2 builds on this by examining core propositions directly drawn from the LMXD framework in a different context with a different LMXD type (separation) as the construct of interest. By examining different LMXD types across diverse empirical settings, these studies advance both methodological clarity and theoretical precision in the LMXD literature and help evaluate the practical utility of Buengeler et al.'s proposed configurations.
Study 1 Theoretical Development
Background of Cobb and Lau’s Study
Cobb and Lau (2015) examined the effects of LMXD on team processes and justice climates above and beyond team mean of LMX (LMXM). Specifically, they found that LMXD had incremental negative effects—above and beyond LMXM—on co-worker communication, relationship conflict, team member exchange (TMX), and both the strength and level of three justice climates (interactional justice, procedural justice, and distributive justice). This study advanced the literature by showing that LMXD can have harmful effects on group outcomes. While the authors provided a novel contribution, Buengeler et al. (2021) identify this study as one example of LMXD research suffering from misalignment. Consequently, this provides the opportunity to apply the Buengeler framework to reexamine the Cobb and Lau (2015) data under conditions of alignment.
In the original paper, the authors invoke theory and explore group outcomes that align with both LMX separation (i.e., group TMX and relationship conflict) and disparity (i.e., group communication and group justice climate strength and level), but operationalize LMXD only using the SD of LMX, which is an index of LMX separation. Yet, the outcomes they targeted, such as communication and justice, are more appropriately linked to LMX disparity. In line with these LMX disparity outcomes, Cobb and Lau (2015) emphasized disparity-aligned theories such as social stratification (Grusky, 1994), equity and justice theories, and concepts of relative deprivation (Bolino & Turnley, 2009) in their theoretical rationale. Therefore, although Cobb and Lau operationalized LMXD using SD (i.e., an index of LMX separation), their theoretical rationale consistently emphasized core mechanisms better captured by LMX disparity. This misalignment presents a validity concern, as it may obscure the true mechanisms underlying team outcomes—analogous to committing a Type I or Type II error. For example, two teams could have the same SD of LMX (i.e., the same level of LMX separation) but very different CV values of LMX (i.e., different levels of disparity), particularly when the average level of LMX differs between teams. This distinction is critical for assessing whether the effects Cobb and Lau observed are better explained by inequality (CV) than general variability (SD).
It is important to note that Cobb and Lau only use disparity-related theorizing and outcomes for their original hypotheses 1, 4, and 5. In contrast, the authors invoke separation-aligned theorizing (i.e., using social distance logics) and separation-aligned outcomes (i.e., TMX and relationship conflict) for hypotheses 2 and 3—meaning these hypotheses were aligned with the operationalization of LMXD using the SD. Therefore, although we directly reproduce each of the five hypotheses, we only provide theoretical background and constructive reproduction analyses for the three hypotheses relevant to our paper's purpose (e.g., Obenauer & Kalsher, 2023), that is, those that were originally misaligned. We focus on Cobb and Lau's hypotheses regarding coworker communication (H1) and justice climate strengths (H4) and levels (H5) and constructively reproduce these analyses using LMX disparity operationalized using the CV. These hypotheses are reflected in the current reproduction study as Hypotheses 1–3. For a full list of original hypotheses and the associated constructive reproduction hypotheses, see our table in OSF.
LMX Disparity and Coworker Communication
LMX disparity represents unequal distribution of high-quality leader–member relationships within a team, where only one or a few members receive preferential treatment while others are left marginalized. This is inherently stratifying, creating distinct social hierarchies that foster negative team dynamics. Drawing on status hierarchy (Blau, 1977) and social stratification (Grusky, 1994), Cobb and Lau describe how unequal access to leadership resources creates rigid subgroup divisions, with high-status members forming communication cliques that exclude others. In such contexts, low-LMX members may withhold communication due to perceived exclusion or fear of judgment, while high-LMX members may gatekeep or avoid sharing information outside their privileged subgroup. LMX separation, which simply reflects the spread of LMX scores around the mean, lacks the power hierarchy implications central to these processes. Two teams may be equally separated in LMX scores but only one may exhibit actual status-based exclusionary dynamics—a distinction that only disparity captures. Thus, we expect that LMX disparity—not separation—will predict lower levels of coworker communication due to its stratifying and exclusionary consequences. Therefore, we use Cobb and Lau's exact hypothesis only conceptualizing LMXD as disparity instead of separation:
LMX Disparity and Justice Climate Strength
Justice climate strength is defined as agreement among members on the fairness of leader treatment. LMX disparity heightens disagreement in how justice is perceived. The inequality leaves only one or few members of the team with high-LMX, monopolizing resources. This leads the majority of the team to feel excluded and undervalued, encouraging perceived unfairness. Justice theories such as equity theory (Adams & Freedman, 1976) and the group-value model of procedural justice (Lind & Tyler, 1988) explain how disparity undermines fairness because unequal treatment from the leader triggers low-LMX members to perceive their input-to-reward rations as unfavorable. These justice concerns extend to all three climates of interactional, procedural, and distributive justice. Consequently, we use Cobb and Lau's exact hypothesis only conceptualizing LMXD as disparity instead of separation:
LMX Disparity and Justice Climate Level
Justice climate level is defined as the overall team perception of justice (i.e., group mean). Disparity can lower these levels because even high-LMX members may view their advantage as unfair, and low-LMX members certainly will (Degoey, 2000). This aligns with group-value theory (Lind & Tyler, 1988) and equity theory (Adams & Freedman, 1976), which suggest that inequality reduces perceived legitimacy of leader behavior. As Cobb and Lau hypothesized, this disconnect extends across all three justice dimensions. Therefore, we use Cobb and Lau's exact hypothesis only conceptualizing LMXD as disparity instead of separation:
Methods
In Study 1, we empirically test Buengeler et al.'s conceptual framework through a constructive reproduction (Kraimer et al., 2023) using data from Cobb and Lau (2015). Specifically, we align their theoretical focus on group-level inequality and stratification with a measurement approach that better reflects these dynamics—LMX disparity. While Cobb and Lau originally operationalized LMX differentiation using SD (i.e., LMX separation), we instead apply the CV, a dispersion index that captures relative inequality in LMX quality. This theoretically grounded shift in operationalization enables us to assess whether alignment between theorizing and measurement yields meaningful empirical differences between LMX separation and LMX disparity. In doing so, we examine the robustness of prior findings and evaluate the added value of the Buengeler et al. framework. This alignment clarifies the complex dynamics of LMXD and enhances the validity and interpretability of its measurement and implications.
Data and Sample
In line with constructive reproduction criteria, we did not collect our own data for Study 1. We use the primary data from Cobb and Lau (2015)'s original data collection as provided by the authors via email to reproduce the results of their study. Therefore, the data is not publicly available, and access is governed by the original authors. The data was collected by the authors via an online cross-sectional survey, collected from a Reserve Officer Training Corps at a large university in the United States. Teams had defined group structure with a formal leader and clear membership with interdependent tasks. Responses from 413 members grouped into 87 teams were received. There was an average of six followers per team with a response rate varying between 62.5 percent and 100 percent, with the exception of one group with a response rate of 50 percent. Participants were largely Caucasian/White (81.2%), male (84%), and average age was 19.59.
Measures
Because the authors provided only an individual-level dataset, we first aggregated the data to the group level to operationalize each of the group-level predictors and outcomes used in the primary study. We confirmed the accuracy and reliability of these aggregated variables by comparing the descriptive statistics and aggregation statistics in our aggregated dataset and those of the primary study and found only minor differences. Therefore, we proceeded with reproducing the results using the team-level variables we computed. Notably, we also generated one new measure of LMX disparity for the constructive reproduction. We detail the variables, and original measures used to capture these variables, used in this study below.
Analyses
Study 1 Correlation Table
Note: Team-level correlations appear below the diagonal. N = 87 teams. Individual-level correlations appear above the diagonal. n = 409–413 employees.
***p ≤ .001; **p ≤ .01; *p ≤ .05.
LMX, Leader–member exchange; SD, standard deviation.
Results
Direct Reproduction
The results of direct reproduction analyses largely mirrored the results reported in Cobb and Lau (2015), with only minor discrepancies in the magnitude, but not direction or significance, of effects. These minor discrepancies are likely attributable to differences in aggregation, software, or code specifications. This provided sufficient evidence for us to continue to the constructive reproduction. For full direct reproduction results and tables, please see our supplementary material in OSF.
Constructive Reproduction
Given the consistency of our reproduction results and the results reported in the original text, we then moved to constructive reproduction. In doing so, we examined the effects of LMX disparity—operationalized as the CV of team LMX—on the disparity-related outcomes of coworker communication (i.e., Hypothesis 1), strength of justice climates (i.e., Hypothesis 2), and level of justice climates (i.e., Hypothesis 3). Before testing our hypotheses, though, we first examined descriptive information regarding the LMX disparity variable. In this sample, the LMX disparity variable (operationalized using CV) had a sample mean of 0.23 (SD = 0.11). Scores on the LMX disparity variable ranged from 0.00 to 0.57, which is slightly range restricted, given that the full hypothetical range of CV for this sample is 0 to 2.15. 3 This is likely attributable to LMX being measured only on a 5-point scale along with the average LMX mean being fairly high across the board. Indeed, CV is maximized under conditions of low mean levels of the variable of interest (Buengeler et al., 2021). Upon inspecting each of the LMX disparity scores for each group, only 16 of the 87 teams were considered high in LMX disparity (i.e., had LMXD scores that met or exceeded +1 SD above the mean). To give an example of a team high in LMX disparity, we use team 23 from the sample. In this team, all six-group members reported their LMX, leading to a team average LMX of 2.60—which is almost two SDs below the sample mean of 3.48. However, inspecting the individual group member ratings of LMX reveals that one team member has an exceptionally high-quality relationship with the leader at 4.29—almost 2 SDs above the mean—whereas the remaining group members have ratings between 1.71 and 2.71, for an average of 2.26. As such, this group is high in disparity, in that one group member holds a high level of the valued asset of LMX quality and the remaining group members have low-quality relationships with the leader. This specific example highlights the utility of the LMX types, because the low group mean alone, without exploring the configuration, would obscure the solo-high configuration in this team.
Having detailed the descriptive statistics regarding the LMX disparity variable, we then tested our constructive reproduction hypotheses and report standardized betas to align with what the original authors report. Using hierarchical regression, we first explored whether LMX disparity explained variables in group communication, above and beyond LMXM. Results show that LMX disparity has a negative relationship with co-worker communications (β = −.38, p = .001), with the full model being significant, F(3, 83) = 6.81, p < .001, and supporting Hypothesis 1, explains significantly more variance in communication than LMX mean alone (ΔR2 = .12, ΔF(1, 83) = 12.69, p < .001). Results of hypotheses 1 appear in Table 2. We then examined the impact of LMX disparity on each of the justice climate strength (i.e., variability) variables and found a positive and significant effect of LMX disparity on the variability of the level of interactional justice climate (β = .58, p = .001), with the full model being significant, F(3, 83) = 13.03, p < .001, procedural justice climate (β = .54, p = .001), F(3, 83) = 10.36, p < .001, and distributive justice climate (β = .32, p = .007), F(3, 83) = 2.58, p = .059, meaning LMX disparity was negatively related to the strength of these climates. Moreover, LMX disparity explained significantly more variance in interactional (ΔR2 = .30, ΔF(1, 83) = 34.91, p < .001), procedural (ΔR2 = .25, ΔF(1, 83) = 28.60, p < .001), and distributive (ΔR2 = .09, ΔF(1, 83) = 7.68, p = .007) justice climate strength than LMX Mean alone. Therefore, Hypotheses 2abc were supported. Further, as predicted there was a significant and negative effect of LMX disparity on levels of interactional justice climate (β = −.26, p = .005), with the full model being significant, F(3, 83) = 19.63, p < .001, and LMX disparity explained significantly more variance in interactional justice climate levels (ΔR2 = .06, ΔF(1, 83) = 8.30, p = .005). However, in contrast to predictions (but consistent with the original results presented in Cobb & Lau, 2015), there was a positive effect of LMX disparity on distributive justice climate (β = .22, p = .026), with the full model being significant, F(3, 83) = 14.85, p < .001, and LMX disparity explained significantly more variance in distributive justice climate levels (ΔR2 = .04, ΔF(1, 83) = 5.14, p = .026). Finally, there was no effect of LMX disparity on procedural justice climate (β = .09, p = .302), although the full model was significant, F(3, 83) = 27.96, p < .001, and LMX disparity did not explain additional variance beyond LMX mean alone (ΔR2 = .01, ΔF(1, 83) = 1.08, p = .302). Therefore, these results support Hypothesis 3a but not 3b and 3c. Results for Hypotheses 2 and 3 appear in Table 3. Notably, these results are consistent with the pattern of effects of the primary study when using LMX separation.
Constructive Reproduction of LMXD Disparity (CV) on Co-Worker Communication (Study 1)
Note: n = 87; Standardized regression coefficients are reported. LMXM = LMX Mean. LMXD = LMX disparity (CV).
***p ≤ .001; **p ≤ .01; *p ≤ .05, † = ≤ .10.
LMXD, Leader–member exchange differentiation; CV, coefficient of variation.
Constructive Reproduction of LMXD Disparity (CV) on Justice Climate Variables (Study 1)
Note: n = 87; Standardized regression coefficients are reported. LMXM = LMX Mean. LMXD = LMX disparity (CV).
***p ≤ .001; **p ≤ .01; *p ≤ .05, † = ≤ .10.
LMXD, Leader–member exchange differentiation; CV, coefficient of variation.
Supplemental Analyses
Although not part of our primary hypotheses, we were also curious whether the LMXD types explained variance above and beyond one another to further test the Buengeler et al. framework. Therefore, we also tested whether LMX disparity explained variance in the disparity-aligned outcomes (communication, and justice climate strengths and levels) above and beyond LMX separation (operationalized as the SD of team LMX). This provides additional evidence into the empirical and theoretical differences among the LMXD types. To do so, we first compared the intercorrelations and descriptive statistics of the LMX disparity and LMX separation variables to understand their overlap and discrepancies. Interestingly, we found that the intercorrelation among LMX disparity and separation is r = .96, suggesting significant overlap between the two LMXD types. This correlation calls into question whether the indices meaningfully capture distinct phenomena and configurations within teams. We also explored how teams were categorized based on these indices and found that the disparity and separation variables operated similarly for most teams. Using the Mean and SD for the LMX disparity and separation variables, we categorized teams into low, average, and high-LMX disparity and separation. This grouping revealed 16 teams high in disparity and 13 teams high in separation. Of the 16 teams high in disparity, 11 of them were those also high in separation. Indeed, across all 87 teams, only 10 were categorized differently based on the separation and disparity groupings (in terms of low, average, and high levels). Again, these results call into question the discriminant validity of the two LMXD types.
To further explore the utility of LMX disparity and separation, we again conducted hierarchical regression, entering group size and LMX separation in step one, and then adding LMX disparity in step 2. Although our results showed that LMX disparity did not explain additional variance above LMX separation for communication or any of the three justice climate strength outcomes, we did find that LMX disparity explained additional variance in the justice climate levels outcomes. Indeed, LMX disparity explained 35% more variance in interactional justice climate than LMX separation, alone. Similarly, LMX disparity explained 36% more variance in procedural justice climate than LMX separation. Finally, LMX disparity explained 21% more variance in distributive justice climate than LMX separation. Notably, though, for each of these analyses, entering both separation and disparity into the hierarchical regression led the LMX separation construct to flip its sign from negative relations with the justice climate variables when modeled alone, to positive relations with the justice climate variables. We suspect the flipped sign is due to multicollinearity, given the extremely high correlation between LMX disparity and separation. Indeed, collinearity statistics (e.g., VIF > 10) suggest multicollinearity might be present in the regression models. Nevertheless, these results suggest that there may be important differences among the LMXD types, in terms of the variance they can explain in their respective outcomes, but also further question the true distinctiveness of the LMXD types. However, the current sample is homogenous and quite small in sample size at both levels 1 and 2, with relatively small group sizes and a sample of only 87 teams. It is possible, therefore, that a sample with greater variability in occupation, group size, and a larger sample of teams would allow for more variability in relationship types and associated team configurations that would allow for representation of the LMXD types in the sample.
Study 1 Discussion
The results of Study 1 provide an initial empirical test of Buengeler et al.'s (2021) theoretical framework by examining whether aligning LMXD theorizing with the appropriate measurement index meaningfully alters empirical conclusions from Cobb and Lau's (2015) original article. Our direct reproduction largely replicated the original study findings, reinforcing the robustness of their conclusions when LMXD is measured using SD (i.e., LMX separation). This provided a sound foundation for our constructive reproduction which assessed whether LMX disparity, operationalized as CV, clarified or altered the relationships Cobb and Lau originally investigated. Results offered partial support for the Buengeler framework. LMX disparity was significantly negatively related to coworker communication and justice climate strength, as predicted, and explained greater variance in these outcomes than LMX mean alone. This suggests that LMX disparity captures dimensions of team stratification and perceived inequality that are overlooked by central tendency scores. However, for justice climate level outcomes, the results were more mixed. Disparity negatively predicted interactional justice level as expected but had no effect on procedural justice and unexpectedly showed a positive association with distributive justice. Importantly, though, when compared to Cobb and Lau's original findings, the direction and significance levels for the procedural justice outcomes remained consistent for the LMX disparity analyses. This pattern raises a critical question: if the outcomes and conclusions remain largely unchanged when using a theoretically aligned LMXD index, does the specific LMXD type or measurement actually matter—at least in practice?
Our supplemental analyses further our understanding of the distinctiveness of the two LMXD types. Although the hierarchical regressions results demonstrate that LMX disparity generally explained more variance than LMX separation in the justice climate outcomes, which would suggest that the two indices are not interchangeable, the high correlation between the two does call the distinctiveness of the indices into question. It is possible, however, that these findings are sample-specific, given the homogeneity of teams (ROTC members) and small sample size. Even so, the results demonstrate that the strength of the effects were stronger for LMX disparity than for LMX separation, indicating that the distribution of LMX relationships may be impactful such that disparity is more detrimental for team outcomes than separation. While these differences in explained variance were small in magnitude, we recognize that even small effects can have meaningful cumulative consequences in organizational settings (Abelson, 1985; Bosco et al., 2015). Thus, modest gains in explanatory power from using theory-consistent LMXD types may still carry practical relevance.
Notably, though, these differences seen in variance explained between the LMXD types are not captured, or are at least muted, when controlling for or comparing against the mean level of LMX in the team. This highlights an important consideration for scholars: while LMXD configurations offer theoretical richness, their practical utility may be diminished when differentiation effects are overshadowed by central tendency. Study 1 thus provides a qualified validation of the Buengeler et al. framework. Aligning theory and measurement did not radically alter conclusions, but it did enhance explanatory power in some outcomes and highlighted conceptual considerations that may help resolve contradictions in the literature. These findings set the stage for Study 2, where we examine whether a different LMXD type, specifically LMX separation, predicts theoretically aligned team outcomes in a broader organizational sample.
Study 2 Theoretical Development
Background of Kim et al.
Kim et al. (2022) examined how team-level LMX relationships and peer mentoring influence team potency and team performance. Their focal independent variable was Team LMX which they operationalized as the median LMX score within a team. Their findings suggested that higher Team LMX enhanced team potency and ultimately team performance, especially when peer mentoring was high. The authors used LMXD as a control variable and therefore LMXD has not been theoretically developed in the original study. This presents an opportunity to theoretically extend Kim et al.'s work. While their approach of calculating team LMX scores using the median level of LMX within a team provides a useful starting point for understanding team dynamics, this method may not capture the full complexity of LMX relationships within teams. Therefore, we reframe Kim et al.'s model by theorizing LMXD as the focal construct of interest. Specifically, using best practices from Buengeler et al.'s (2021), we argue that LMX separation is the most appropriate type of differentiation to assess Kim et al.'s outcomes because it reflects broad perceptions of overall relationship quality and aligns with the effectiveness-related outcomes of interest. High-LMX separation signals fractures in shared perceptions of leadership and collective capability, making it theoretically relevant for predicting team effectiveness outcomes such as potency and performance.
In Study 2, we test a portion of Buengeler et al.'s theoretical model using secondary data from Kim et al. (2022). We develop and test novel hypotheses derived from the LMXD framework, positioning LMX separation, rather than median LMX, as the focal predictor of team outcomes. In doing so, we examine whether a theory-consistent operationalization of LMXD enhances our understanding of team functioning beyond what central tendency measures alone can reveal. Conducting this theoretical extension allows us to extend the scope of the original study and further evaluate the utility of the Buengeler et al. (2021) framework in a different empirical context.
LMX Separation and Team Effectiveness
Social categorization theory (Tajfel & Turner, 2004) suggests that individuals inherently classify themselves and others into social groups based on salient characteristics. In team contexts, LMXD creates identifiable subgroups within teams based on perceived relationship quality with the leader. LMX median overlooks these subgroup dynamics because it aggregates perceptions into a single score, masking the potential for how the team is divided. LMX separation—SD of LMX—has the ability to capture how the team is divided, which can allow us to theorize using social categorization. When LMX separation is high, members with high-quality LMX relationships may be perceived as the in-group, while those with lower quality relationships form the out-group (Li & Liao, 2014). These categorizations can undermine team cohesion, increase relationship conflict, and inhibit effective communication, ultimately impeding team processes and outcomes. Team potency, defined as shared confidence in the team's ability to perform across contexts (Guzzo et al., 1993), is particularly vulnerable to these dynamics. When subgroups are formed and perceived, the lack of collective identity and cohesion disrupts the team's ability to function as a unified entity, reducing its confidence. This diminished team potency, in turn, adversely impacts team performance, as the teams’ ability to align efforts and achieve goals is compromised. Additionally, balance theory (Heider, 1958) emphasizes the discomfort individuals experience in imbalanced or asymmetrical relationships. LMX separation introduces the concept of imbalance in leader–member relationships, which creates tension and discomfort, eroding team confidence (team potency) and hindering performance. These nuances are absent in a median score, which assumes uniformity in leader–member interactions. Therefore, we hypothesize:
Team Potency as a Mediator
While LMX separation is likely to impair team performance directly, team potency is a theoretically grounded mechanism that explains how and why this effect occurs. Team potency is a shared psychological resource that enables collective confidence in facing challenges and achieving goals thus serving as a critical psychological resource for achieving high team performance (Guzzo et al., 1993). When potency is high, team members are more likely to coordinate effectively, persist through challenges, and take ownership of their collective success. However, the relational dynamics introduced by LMX separation can undermine the foundation required for this shared confidence. High-LMX separation signals a fragmented social environment where access to leader support is uneven. Drawing on social categorization theory, these internal divisions foster perceptions of in-group and out-group membership, making it difficult for teams to cultivate a unified identity. Without a strong sense of cohesion or inclusion, team members may feel less committed to the group's success, diminishing their belief in the team's capability. This reduction in potency has downstream consequences such as the team may fail to coordinate effectively, experience reduced collective motivation and ultimately perform at a lower level. Thus, team potency serves as a mechanism explaining why LMX separation may indirectly influence team performance. Therefore, we propose the following mediation hypothesis:
LMX Separation Beyond LMX Disparity
A central proposition of Buengeler et al.'s (2021) LMXD framework is that the type of LMX differentiation must match the theoretical mechanism and the outcome of interest. Thus, we examine whether LMX separation explains more variance than other LMXD types. In this case, we are only able to test this hypothesis against LMX disparity, because LMX scale (LMX-MDM) produces continuous scores that do not support categorical coding schemes needed for LMX variety indices (e.g., Blau's index requires discrete categories).
Misalignment—such as using a measure of LMX disparity to test outcomes predicted by LMX separation—can produce misleading or diluted findings. In this study, we focus on team potency and team performance, which are group-level outcomes that reflect collective functioning, shared confidence, and psychological cohesion. According to Buengeler et al., such outcomes are theoretically aligned with LMX separation, because it reflects the extent to which a team is fractured along lines of relational quality. High-LMX separation signals divisions in the team's shared experience of leadership, which undermines unity and the team's belief in its ability to work effectively—a core element of team potency. In contrast, LMX disparity is conceptually tied to fairness concerns, status stratification, and social comparison processes. It reflects whether a select few members receive disproportionately high-quality relationships with the leader, not whether the team as a whole is divided. As such, LMX disparity is better suited to predicting outcomes like justice climate or resentment, but is less directly tied to shared psychological constructs like team potency. Thus, while both are valid forms of LMXD, only LMX separation aligns conceptually with the mechanisms underlying team cohesion and confidence. As Buengeler et al. (2021) note, such structural differences are not interchangeable and failing to align construct and theory can obscure meaningful effects.
To more fully test the Buengeler et al.'s assumption that theoretically aligned LMXD types should explain more variance in relevant outcomes, we compared the indirect effect of LMX separation on team performance via potency to the indirect effect of LMX disparity. We chose to examine the mediation-only model because we were interested in the differences among LMX separation and disparity, alone, as opposed to their dual effects with peer mentoring. To do so, we tested a multiple-mediation model for the supervisor-rated and the top-manager-rated performance outcome. If the framework holds, LMX separation should better account for the indirect effects on performance via potency than the non-theoretically aligned LMXD type of LMX disparity. Therefore, we hypothesize:
Methods
Data and Sample
Study 2 utilized a primary dataset previously published in Kim et al. (2022). The first author provided the primary dataset (with data at both the individual and team levels) via email and granted permission for us to test our new LMXD hypotheses using the dataset. The data is not publicly available, but may be accessed by contacting the original authors. This dataset included responses to time-lagged surveys representing 592 subordinates nested in 111 teams from 25 diverse organizations across various industries in China. Each team reported to a single supervisor and a single upper-level manager, each of whom provided survey responses regarding their respective team's performance. Thus, Study 2 is based on dyadic leader–follower data that captures vertically linked relationships nested within work teams. Teams had an average size of 6.33 members, ranging from 3 to 19 members. The average team response rate was 82.96%. Of the subordinates, 36.7% were female, the average age was 34.06, and average organizational tenure was 9.14 years. Of the supervisors, 28.2% were female, average age was 38.96, and average organizational tenure was 12.22 years.
Measures
We confirmed the accuracy and reliability of the team-level variables reported in Kim et al. (2022) by reconstructing a team-level dataset from the individual-level dataset and found the descriptive statistics and aggregation statistics to match those of the team-level dataset provided by the authors. Therefore, we proceeded to test our hypotheses using the team-level variables provided by the original author team. Notably, though, we did generate one new measure of LMX separation. We detail the variables used in this study, below.
Analyses
Results
Direct Reproduction Results
Because direct reproduction was not the primary goal of the current research, we summarize our findings briefly and report the full results of the direct reproduction in our OSF repository. 4 Notably, we began with direct reproduction to demonstrate that the methodological approach used and statistical conclusions made in the current study are valid. In reproducing the results of the full mediation and mediated moderation models tested in Kim et al. (2022), we were largely able to reproduce the results reported in the original manuscript. 5 As such, the direct reproduction results indicated that our methods and analyses were similar to those used in the primary study, allowing us to move forward with our hypothesis tests.
Hypothesis Testing
Having established the validity of our analyses, we proceeded to test our new primary hypotheses derived from the Buengeler et al. (2021) framework. But, as in Study 1, before testing the hypotheses we first inspected the descriptive information regarding the LMX separation variable. In this sample, LMX separation (operationalized using SD) had an average group mean of .50 (SD = .38) with a range of 0 to 1.49. The theoretical range of LMX separation for this sample is 0 to 2.08, suggesting the presence of range restriction5. Because the maximum value for range is calculated using the scale mean, the number of scale anchors should not affect range restriction as might be expected with CV (i.e., such as in Study 1), but similar to Study 1, the average LMX in teams was quite high for the sample. This could contribute to range restriction because high separation is more likely under the conditions of LMX that is low or at the mid-point. Upon inspection of the LMX separation scores for each group, only 18 of the 111 teams were considered high in LMX separation (i.e., had LMXD scores that met or exceeded +1 SD above the mean). To give an example of a team high in LMX separation, we illustrate using team 19. This team comprised of 18 members had an average LMX of 4.85—less than half a SD below the mean. However, inspecting the individual group member ratings of LMX reveals two subgroups—one of higher-than-average LMX relationships and one of lower-than-average LMX relationships. Indeed, 11 of the 18 team members had LMX relationships ranging from 2.33 to 4.75, with an average LMX of 4.13 (i.e., almost two SDs below the mean). In contrast, seven group members has LMX relationships ranging from 5.33 to 6.58, with an average LMX of 5.98 (i.e., almost 1.5 SDs above the mean). This team demonstrates the quintessential bimodal distribution characterizing LMX separation, such that there is an “in-group” with high levels of LKMX relationships and an “out-group” with low levels of LMX relationships. As in the example provided in Study 1, examining the mean alone would obscure the bimodal configuration present in this team. Full descriptives for Study 2 appear in Table 4.
Study 2 Correlation Table
Note: Team-level correlations appear below the diagonal. n = 111 teams. Individual-level correlations appear above the diagonal. n = 538–592 employees.
***p ≤ .001; **p ≤ .01; *p ≤ .05.
LMX, Leader–member exchange; SD, standard deviation.
Having detailed the descriptive statistics of our constructs at hand, we then tested our primary hypotheses and report unstandardized betas. Hypothesis 1 posited that LMX separation would have a negative relationship with team potency and team performance (as rated by supervisors and top managers). Results demonstrated a negative relationship between LMX separation and team potency (b = −.20, SE = .12, p = .086), but this effect was only marginally significant. Further, LMX separation was negatively related to team performance as reported by supervisors (b = −.55, SE = .21, p = .009), but was not significantly related to team performance as reported by top managers (b = −.22, SE = .24, p = .342). Notably, LMXM had significant positive relationships with each of the three outcomes, and the addition of LMX separation as a predictor explained 4% additional variance in potency, 7% additional variance in supervisor-rated team performance, and 3% additional variance in top manager-rated team performance. Accordingly, Hypothesis 1 received marginal support. Full results appear in Table 5.
Hypothesis 1: Direct Effects of LMX Separation on Team Potency and Team Performance (Study 2)
Note: As in Kim et al., unstandardized regression coefficients are reported and SE in ().
***p ≤ .001; **p ≤ .01; *p ≤ .05; † = ≤ .10.
LMX, Leader–member exchange; LMXD, Leader–member exchange differentiation; SD, standard deviation.
Hypothesis 2 explored the indirect effect of LMX separation on team performance via team potency. Despite weak direct effects between LMX separation and the effectiveness variables, the indirect effects of LMX separation on team performance rated by supervisors (ab = −.09†, CI = [−.36, −.01]), and top managers (ab = −.09†, CI = [−.37, −.01]) via team potency were negative and significant, supporting Hypothesis 2. Notably, the inclusion of LMX separation as a predictor in the model explained only an additional 0.2% of variance in supervisor-rated team performance and 1% of variance in top manager-rated team performance above LMXM. Accordingly, results of Hypotheses 1 and 2—shown in Table 6—demonstrate that although LMX separation had the expected direct and indirect effects on group effectiveness, these effects are dwarfed by the central tendency scores.
Hypothesis 2: Indirect Effects of LMX Separation on Team Performance via Team Potency (Study 2)
Note: As in Kim et al., unstandardized regression coefficients are reported and SE in (). Confidence intervals for indirect effects were generated using Monte Carlo simulation.
***p ≤ .001; **p ≤ .01; *p ≤ .05; † = ≤ .10.
LMX, Leader–member exchange; SD, standard deviation.
Hypothesis 3 explored the effects of LMX separation on group effectiveness as compared to LMX disparity. Given the multicollinearity issues present in the supplemental analyses of Study 1, we first examined the intercorrelation between LMX separation and LMX disparity (operationalized using CV), and, as in Study 1, found a large intercorrelation of r = .97. Consistent with the effects seen for LMX separation in Hypothesis 2, LMX disparity also had a significant and negative indirect effect with supervisor-rated (ab = -3.29, CI = [-13.15, −0.93]) and top manager-rated performance (ab = -3.10, CI = [-12.90, −1.00]) via team potency. However, when modeling the indirect effects of both LMX separation and LMX disparity simultaneously, the indirect effect for LMX separation became positive and remained significant. As in Study 1, we believe the change in direction of the effects is due to multicollinearity given the high intercorrelation between LMX separation and LMX disparity. As such, the significant negative indirect effects are likely more reliable estimates of the true effects of separation on the effectiveness outcomes. Notably, the addition of LMX disparity as a predictor explained an additional 2% of variance in team potency, an additional 5% of variance in supervisor-rated team performance, and an additional 2% of variance in top-manager-rated performance. Accordingly, Hypothesis 3 was not supported, given that LMX disparity demonstrated stronger negative effects with the group effectiveness outcomes, and explained additional variance in the outcomes, than LMX separation. Full results are reported in Table 7. As in Study 1, these results again underscore that there may be small, nuanced differences among the LMXD types that explain additional variance in group outcomes. However, these findings are again overshadowed by central tendency scores and continue to call into question the utility of the various LMXD types proposed in the Buengeler et al. framework.
Hypothesis 3: Indirect Effects of LMX Separation and LMX Disparity on Team Performance via Team Potency (Study 2)
Note: As in Kim et al., unstandardized regression coefficients are reported and SE in (). Confidence intervals for indirect effects were generated using Monte Carlo simulation.
***p ≤ .001; **p ≤ .01; *p ≤ .05; † = ≤ .10.
LMX, Leader–member exchange; LMXD, Leader–member exchange differentiation; SD, standard deviation; CV, coefficient of variation.
Study 2 Discussion
Results of Study 2 demonstrate a similar pattern of effects as those seen in Study 1. Specifically, results demonstrated that LMX separation was negatively related to potency and performance, though these results were weak at times. Importantly, LMX separation explained incremental variance in team outcomes beyond team-level LMX median, suggesting that dispersion in LMX relationships carries explanatory value above and beyond central tendency. These findings diverge from Study 1 in one important respect: whereas Study 1 emphasized the theoretical alignment of measurement and outcome, Study 2 illustrates that both the level and the distribution of LMX relationships matter for group effectiveness. Higher median LMX scores were consistently associated with better team outcomes, yet greater LMX separation sometimes undermined these benefits—indicating that central tendency and dispersion are simultaneously predictive but may push team outcomes in different directions.
Moreover, when comparing the effects of LMX separation and disparity on group effectiveness outcomes, we found significant but small (in effect size), differences between these two types of LMXD regarding their relationships with team outcomes. Notably, though, the pattern of effects differed from our predictions. LMX disparity was more strongly related to the group effectiveness outcomes than LMX separation and explained incremental variance in outcomes over LMX separation. This failed to support the test of the Buengeler et al. (2021) framework, which proposes that alignment between theoretical mechanism and LMXD type should yield stronger effects. Although we selected LMX separation as the theoretically aligned construct for this context, LMX disparity showed greater predictive utility—suggesting potential limits to the framework's empirical traction or highlighting measurement challenges. Regardless, these results suggest that the configurations themselves matter, in that groups with a solo-high configuration of LMX relationships (i.e., groups characterized by LMX disparity) might perform worse than those with a bimodal distribution of LMX relationships (i.e., LMX separation). Nevertheless, the high intercorrelations among LMX separation and disparity continue to call the discriminant validity of the LMXD types into question.
General Discussion
LMXD is a critical, but complex, construct that captures the variability in leader–member relationships within teams. While prior research has demonstrated that such differentiation can have significant implications for team functioning, findings have often been contradictory—partly due to misalignment between theoretical mechanisms and measurement strategies. To address this, we tested Buengeler et al.'s (2021) conceptual framework, which proposes that LMXD comprises distinct types—separation, disparity, and variety—that should be matched to theory-aligned outcomes and measurement indices. Across two studies, we evaluate one of the core premises of the Buengeler et al. framework: that theoretical alignment between construct and measurement enhances explanatory power. In Study 1, we constructively reproduce Cobb and Lau (2015), shifting the operationalization from LMX separation to LMX disparity to match the study's justice-based theorizing. In Study 2, we test the predictive utility of LMX separation on team outcomes from Kim et al. (2022), guided by social categorization and balance theory. Our findings contribute to the ongoing debate regarding LMXD's utility and offer insights into the empirical distinctiveness of its types proposed by Buengeler and colleagues.
Our findings provide partial support for this premise. Across both studies, LMXD types explained incremental variance in team outcomes above central tendency measures (i.e., mean and median), and the configuration of LMX quality—whether bimodal (separation) or solo-high (disparity)—had meaningful consequences for team functioning. However, empirical distinctions between LMXD types were weaker than anticipated. Even so, we recognize that even small effects can have meaningful cumulative consequences in organizational settings (Abelson, 1985; Bosco et al., 2015). In Study 2, the theoretically misaligned index (disparity) outperformed the aligned index (separation), challenging the assumption that theory-consistent measures will yield superior predictive utility. These inconsistencies suggest that existing indices may not adequately capture the nuances of differentiation—or that other factors (e.g., team mean LMX) may drive the effects attributed to LMXD. Indeed, our findings raise concerns about measurement redundancy and discriminant validity, particularly given the extremely high correlations (r > .95) between separation and disparity across both studies. This convergence may stem from range restriction in LMX scores or from overlapping computational properties of SD and CV. Taken together, our research affirms that LMXD is not a monolithic construct, but also underscores that more work is needed to clarify how and when its different forms matter. While our studies advance theoretical precision and empirical insight, they also signal the need for improved measurement practices, broader sampling, and continued refinement of the LMXD framework. Nonetheless, it remains beneficial for the literature to practice alignment between theory and measurement as Buengeler et al. assert.
Theoretical Implications
Our findings offer several theoretical contributions. First, we reinforce the central proposition of Buengeler et al. (2021)—that different types of LMXD should be conceptualized and measured according to their unique theoretical mechanisms. Both studies demonstrated that configuration matters: dispersion in LMX quality can negatively affect team outcomes even when central tendency is high, supporting the need to go beyond average LMX scores.
However, our results also raise significant challenges to the framework. First, the empirical distinctiveness between LMXD types was not as strong as theoretically proposed. In both studies, LMX disparity consistently outperformed LMX separation, even when separation was the theoretically aligned type. This consistent pattern suggests that disparity may offer greater explanatory utility, potentially because its computation—SD divided by the mean—captures relative dispersion, thereby amplifying nuances that SD alone may miss. While this enhances the practical value of disparity, it challenges the framework's assumption that measurement–theory alignment will yield superior predictive accuracy. Further, given the high correlations between LMXD types across two very different samples, our results call into question the empirical distinctiveness of the constructs. Given that the indices only differ at high levels of differentiation, perhaps instead of predictors, they operate more effectively as moderators of central tendency scores (e.g., Seo et al., 2018). Across both studies, central tendency variables (mean or median LMX) consistently exert stronger effects, meaning that LMXD may serve to qualify or condition those relationships. Finally, our work raises theoretical concerns about LMX variety as a construct. Given that LMX variety requires categorical indices that our datasets could not support, we were unable to evaluate it empirically. As a result, it remains unclear whether variety represents a distinct type of differentiation or functions better as an antecedent or contextual factor that shapes LMX formation. Theoretical clarity on this point is needed to complete the LMXD typology.
Practical Implications
This research also carries practical lessons for leaders and organizations. The most consistent finding across both studies was that LMX disparity—solo-high configurations—is particularly harmful for team effectiveness. When one team member enjoys a high-quality relationship with the leader while others do not, the resulting stratification can erode perceptions of fairness, reduce communication, and undermine performance. Even in teams with high average LMX, disparity can create divisions that damage cohesion. Leaders should be cautious of developing close ties with only one or two subordinates, especially if others perceive those relationships as exclusive or privileged. Interventions that focus on improving lower quality LMX relationships—rather than further strengthening already high ones—may be most effective. Additionally, leaders should avoid behaviors that signal favoritism, such as over-relying on a single “go-to” person, which may create resentment or disengagement among other team members. From a systems perspective, organizations should consider monitoring LMXD patterns during leader evaluations, talent reviews, or team assessments. If disproportionate investment in certain team members is identified, leadership development efforts should emphasize balanced relationship-building and inclusive engagement practices.
Limitations and Future Research
Although our study offers robust empirical testing of the Buengeler et al. (2021) framework, several limitations warrant consideration. Our reliance on existing datasets constrained our ability to shape sample characteristics and limited the breadth of measurement options (e.g., insufficient variability in team compositions or low response rate per teams). At low levels of LMXD, separation and disparity appear mathematically indistinguishable (Buengeler et al., 2021). Such limitations may constrain the empirical utility of LMXD and blur the uniqueness of specific types. Therefore, future theory tests should seek samples that maximize variability in LMX patterns to better distinguish among differentiation types. We acknowledge the limited generalizability of our findings due to the characteristics of the samples. For instance, in Study 1, we analyzed data from ROTC squads within a large university. This context offers clear advantages for testing LMXD—such as formal team structures and designated leaders—but it may not reflect the relational dynamics found in more complex, less hierarchical, or more diverse organizational settings. Although this was secondary data, we recognize the need for replication across other occupational, cultural, and team environments to establish the broader applicability of Buengeler et al.'s framework. Additionally, both studies relied on non-experimental, observational data, so our findings may be subject to endogeneity concerns. Our reliance on secondary data further prevented us from implementing formal endogeneity corrections or experimental manipulation of LMXD. Accordingly, causal interpretations should be made with caution. Future research could explore how different types of LMXD emerge and evolve over time, potentially through longitudinal or experimental designs. Future studies using purposefully design data collection strategies with specific types of LMXD in mind—especially with larger teams and full response coverage—are needed to isolate these constructs more precisely.
Further, we join recent calls for more open science practices in management research. Greater access to high-quality and diverse datasets would allow for stronger reproductions, replications, generalizability studies, and theoretical framework tests. This would allow for better understanding of complex phenomenon, such as LMXD, and strengthen our field as a whole. Although our study is not without limitations, it highlights the value of rigorous theory testing in advancing leadership research. Buengeler et al.'s framework has conceptually unpacked complex configurations, now it is scholars’ duty to provide sufficient empirical testing to strengthen our confidence in such framework. Future research should continue to examine the boundary conditions of LMXD types, refine their measurement in line with theoretical definitions, and replicate findings across diverse contexts. Doing so allows LMXD to remain a meaningful, valid, and practically useful construct in the study of leadership and team dynamics.
Footnotes
Author Note
We would like to express our sincere gratitude to Dr. Tae-Yeol Kim and Dr. Anthony Cobb for generously providing access to their datasets, which was instrumental in conducting this research. We would also like to thank the JOMSR “Pub” Crawl for providing early exposure and feedback. Supplemental material for this article is available at 
.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.