Abstract
In this study, we propose that the upward leader-leader exchange (LLX) relationship is an important moderating condition in predicting the consequences of leader-member exchange (LMX) differentiation within work groups. We assert that the structural and operating efficiencies created by LMX differentiation will depend on group members’ appraisal of the legitimacy of the within-group LMX disparity. Drawing on relative deprivation theory, we argue that the level of perceived legitimacy of LMX differentiation varies with levels of LLX. Using data collected from 579 subordinates and 74 managers in 74 work groups over two time periods, we propose and test hypotheses that the quality of a leader’s LLX relationship will moderate the mediated relationship between LMX differentiation, group-level teamwork, and team effectiveness. We find support for our moderated-mediation model and suggest implications for theory and practice.
Introduction
Vertical dyadic linkage theory, the precursor to leader-member exchange (LMX) theory, proposes that leaders develop distinctive types of exchange relationships with their subordinates. Enriched LMX relationships between a leader and subordinate extend beyond the formal contractual relationship to include mutually expanded exchanges of support and resources (Dansereau, Graen, & Haga, 1975). In contrast, low-quality exchanges are more circumscribed relationships based on contractual or economic resource exchanges and more limited interpersonal interactions. These differentiated exchange types are believed to evolve naturally in the early stages of employment and are influenced by both employee preference and supervisory evaluations of employee competence (Dansereau et al.; Graen, 1976; Graen & Cashman, 1975). LMX theory suggests that the efficacy of leader behavior is optimized by developing high-quality relationships with a small number of subordinates who can act as managerial surrogates (Dansereau et al.). These surrogates offer an expanded managerial presence and structural efficiencies argued to be important to group functioning and effectiveness (Dansereau et al.; Dienesch & Liden, 1986; Graen; Liden, Sparrowe, & Wayne, 1997). Implicit in this reasoning is the assumption that leaders will provide enriched LMX relationships to only a few within the work group—a process referred to as LMX differentiation (Henderson, Liden, Glibkowski, & Chaudhry, 2009; Liden, Erdogan, Wayne, & Sparrowe, 2006).
A robust literature has subsequently developed suggesting enriched LMX relationships are positively associated with employee job satisfaction, organizational commitment, access to information, better work assignments, career advancement, salary progression, and information access (Gerstner & Day, 1997; Graen & Uhl-Bien, 1995). In contrast, employees in low-quality exchanges are less satisfied, are given more mundane assignments, receive less support, have fewer advancement opportunities, and have a greater tendency to leave the organization (Dulebohn, Bommer, Liden, Brouer, & Ferris, 2012; Gerstner & Day; Graen & Uhl-Bien; Maslyn & Uhl-Bien, 2001; Vecchio, 1995). Thus, considerable benefits appear to accrue to employees involved in higher quality LMX relationships that do not accrue to those in lower quality exchange relationships. Moreover, differences in LMX quality are found to be common within work groups (Liden & Graen, 1980), and the relative LMX status of individual members is understood by the group (Duchon, Green, & Taber, 1986; Maslyn & Uhl-Bien; van Breukelen, Schyns, & Le Blanc, 2006). The apparent inequity that results has led researchers to suggest that another possible consequence of LMX differentiation is a negative impact on team dynamics via compromised fairness perceptions (Liden et al., 1997; Scandura, 1999; Sias & Jablin, 1995; Tyler, 1989). The ability to reconcile the proposed structural and efficiency benefits of LMX differentiation with the deleterious effects of social comparison processes on team dynamics represents an important challenge in LMX theory development beyond the leader-member dyad (Graen & Uhl-Bien; Henderson et al., 2009; House & Aditya, 1997; Liden et al.; Scandura).
The limited research investigating the relationship between LMX differentiation and individual and group outcomes suggests that effects of LMX differentiation are not straightforward (Boies & Howell, 2006; Erdogan & Bauer, 2010; Hooper & Martin, 2008; Liden et al., 2006; Wu, Tsui, & Kinicki, 2010). While research has failed to produce consistent main effects between LMX differentiation and group-level outcomes, a number of contextual moderating conditions have been identified that affect the direction and strength of the relationship (e.g., Erdogan & Bauer; Liden et al.). These studies position the effects of LMX differentiation as contingent upon contextual boundary conditions (Henderson et al., 2009). For example, a group’s median level LMX has been shown to moderate the relationship between LMX differentiation and team performance such that the relationship is positive only in low median LMX contexts (Le Blanc & González-Romá, 2012; Liden et al.). The purpose of the present paper is to contribute to these efforts by arguing that the quality of the leader’s relationship with his or her boss, or leader-leader exchange (LLX), represents an additional boundary condition for the effects of LMX differentiation in work groups.
While the LLX relationship as a contextual feature was of central importance in early vertical dyadic linkage theory (Cashman, Dansereau, Graen, & Haga, 1976; Graen, Cashman, Ginsburg, & Schiemann, 1977), it has been largely ignored in the more recent LMX empirical literature. Because leaders serve as “linking pins” or conduits of both social and material resources within organizational systems, the quality of their own LLX relationship is an important determinant of the resources available for distribution within their own work group (Argyris, 1964; Graen, Dansereau, & Minami, 1972; Likert, 1961, 1967; McGregor, 1960). We apply relative deprivation theory (Bolino & Turnley, 2009; Crosby, 1976; Folger, 1986; Folger & Martin, 1986) to suggest that LLX quality, and the consequent availability of resources, is an important influence on a work group’s appraisal of the fairness of the leader’s decision to differentiate, thereby yielding different expectations regarding the effects of differentiation on work groups led by low- and high-LLX leaders.
Using work group data collected from 579 employees and 74 managers in seven organizations in mainland China, we test hypotheses proposing that the impact of LMX differentiation on group-level teamwork and subsequent team effectiveness will be moderated by the relative quality of the leader’s LLX. We also address inconsistency in the measurement of differentiation and provide theoretical and empirical guidance to future research (Harrison & Klein, 2007). In doing so, we offer additional theoretical clarity to the effects of LMX differentiation at the group level and further illuminate those circumstances in which improved structural efficiencies or detrimental team dynamics might dominate.
Literature Review
The goal of this study is to better understand the role of the LLX relationship in shaping the effects of LMX differentiation on group-level teamwork helping behavior and team effectiveness. Teamwork helping behavior (or simply, “teamwork”) describes the level of helping supportive behavior among team members (Glaser, Zamanou, & Hacker, 1987; Salas, Rosen, Burke, & Goodwin, 2009). Because LMX differentiation is argued to be potentially harmful to social harmony within work groups, teamwork behavior and effectiveness are seen as theoretically and practically important outcomes that need to be examined in assessing the overall impact of LMX differentiation on performance (Boies & Howell, 2006; Hooper & Martin, 2008; Liden et al., 1997; Scandura, 1999; Sias & Jablin, 1995; Sparrowe & Liden, 1997; Tyler, 1989; van Breukelen et al., 2006).
We begin by reviewing the relationship between LMX and employee attitudinal and behavioral outcomes at the individual level found in previous literature. We then review progress and challenges in extending these results to the group level. On the basis of relative deprivation theory, we propose that LLX affects employee appraisals of LMX differentiation within work groups and moderates its impact on teamwork behavior and team effectiveness.
LMX and Teamwork Behavior at the Individual Level
Rooted in social exchange theory (Blau, 1964), LMX theory and research is based on the idea that the supervisor-subordinate relationship can be understood as an exchange of both material and social resources (Graen & Scandura, 1987; Graen & Uhl-Bien, 1995). The quality of the exchange is believed to evolve early in the employment relationship and is based on the mutual appraisal of the relative value of resources received with the exchange (Graen & Scandura). Consistent with researchers in past work investigating LMX differentiation (Boies & Howell, 2006; Erdogan & Bauer, 2010; Liden et al., 2006; Stewart & Johnson, 2009), here, we have conceptualized and studied LMX as a unidimensional higher order construct (Graen & Uhl-Bien). However, LMX has been shown to be composed of affective, loyalty, contribution, and respect dimensions (Greguras & Ford, 2006; Liden & Maslyn, 1998). Accordingly, the exchange relationship is argued to manifest through emotions—with affect, respect, and trust operating as important facets of the exchange (Cropanzano & Mitchell, 2005; Graen & Uhl-Bien). Exchanges in which the investment of resources is reciprocated by both parties tend to develop into more enriched, higher quality exchanges, while exchanges in which marginal increases in resources are unreciprocated (by leader or member) tend to evolve toward lower quality, economic exchange. In this way, LMX theory is a transactional model of leadership based on the assumption that the effects of LMX quality operate through the transfer of resources from leaders to member and vice versa (Yukl, 1989).
At the individual level, the quality of the exchange relationship has also been shown to manifest behaviorally in the form of increased willingness to engage in behavior supportive of the group or organization, such as helping coworkers, being cooperative, or sharing information that is not explicitly required in task performance (Deluga, 1994; Dulebohn et al., 2012; Ilies, Nahrgang, & Morgeson, 2007; Liden et al., 1997; Settoon, Bennett, & Liden, 1996; Wayne, Shore, & Liden, 1997). Again, consistent with social exchange theory logic, the performance of these discretionary behaviors represents opportunities to repay the benefits received via their high-quality LMX relationship.
LMX at the Group Level
While the consequences of LMX quality for individual employees are well established, the effects of a differentiated approach to leadership are less clear at the group level. LMX theory presumes that at the group level, the development of differentiated relationships with subordinates is both a natural and functional leader behavior (Dansereau, 1995; Dansereau et al., 1995). By effectively multiplying themselves through deep investments in select employees, leaders create an emergent hierarchy in which high-LMX employees act as managerial proxies in training and mentoring other group members (Dansereau et al., 1975). Through this process, efficiencies are achieved and group effectiveness is enhanced. Furthermore, to the degree that LMX quality is allocated in favor of the more competent and better performing employees, differentiation operates as a reward to these employees and an incentive in motivating poorer performers towards increased effort (Dansereau).
However, group-level research suggests that within work groups, LMX differentiation is likely to affect a host of comparative and social processes among group members that may negatively affect group-level effectiveness (Liden et al., 1997; Scandura, 1999; Sias & Jablin, 1995; Sparrowe & Liden, 1997; van Breukelen, Schyns, & Le Blanc, 2006). While the relative quality of the LMX relationship is theorized to result primarily from a leader’s observation and experience with an employee and an employee’s preferences (Graen & Uhl-Bien, 1995; Liden et al.), research suggests that employee perceptions of relative LMX quality among fellow group members are subject to attribution bias that affect employee fairness perceptions (Ansari, Hung, Aafaqi, 2007; Sias & Jablin; Vecchio, Griffeth, & Hom, 1986). Imbalances in LMX quality between two employees and a supervisor can spawn contempt between employees (Hu & Liden, 2013). For example, Sias and Jablin collected qualitative evidence that when individuals received unfavorable treatment from their supervisor, it was seen as unfair. In these circumstances, high-quality LMX relationships were attributed to ingratiation tactics by the subordinate towards the supervisor and adversely affected trust, liking, and communication within the work group. Furthermore, research demonstrates that within work groups, one’s relative standing in terms of LMX quality affects individual-level self-efficacy (Hu & Liden) and provides additional explanatory power to the relationship between LMX quality and performance outcomes (Vidyarthi, Liden, Anand, Erdogan, & Ghosh, 2010). Therefore, the expectation that LMX differentiation will result in improved group functioning is not automatic but, instead, depends on intermediate perceptual and comparative mechanisms that shape its effects. However supportive or harmful to group effectiveness differentiated LMX quality may be, it is unlikely that its impact can be understood simply by summing the quality of each dyadic relationship within the work group.
Because of the inherent ambiguity in understanding the main effect of LMX differentiation on group effectiveness, recent empirical studies have focused on contextual conditions and intermediate processes that moderate the impact of LMX differentiation on group outcomes. For example, Liden and colleagues (2006) reported no main effect for LMX differentiation on group performance but found that task interdependence moderated the relationship such that for teams with high task interdependence, the relationship between LMX differentiation and performance was positive. In support of this finding, Liden et al. (2006) argued that, under conditions of high task interdependence, there is a greater need for leaders to coordinate interactions among group members. They reasoned that LMX differentiation helps improve group performance when task interdependence is high by allowing the leader to more effectively facilitate group coordination by strategically distributing critical tasks to more competent group members. In addition, Liden et al. (2006) hypothesized and found that the group-level median LMX moderated the LMX differentiation and group performance relationship such that a positive relationship was found only under a low-LMX median condition. Similarly, Le Blanc and González-Romá (2012) concluded that LMX differentiation was positively associated with team performance and team commitment only when the group-level LMX median was low. In both studies, the authors argued that a leader’s failure to differentiate in a low median LMX circumstance would result in a group composed entirely of low-quality LMX relationships. Therefore, it is necessary to couple low median LMX level with high-LMX differentiation to ensure there are at least a few members with high-LMX quality who are willing to expend additional effort on critical team tasks. Finally, Erdogan and Bauer (2010) found that LMX differentiation was related to more negative work attitudes and coworker relations and higher levels of withdrawal behaviors only when justice climate was low. The finding that justice climate moderates the relationship between LMX differentiation and work outcomes supports previous claims that fairness perceptions are important in shaping the effects of LMX differentiation (Liden et al., 1997; Scandura, 1999; Sias & Jablin, 1995; Tyler, 1989).
Further obscuring the effects of LMX differentiation at the group level is the diversity of measures that are used to represent differentiation in empirical tests of its effects. For example, within-group variance (Erdogan & Bauer, 2010; Liden et al., 2006), agreement indices (Boies & Howell, 2006), and measures of disparity (Hooper & Martin, 2008; Wu et al., 2010) have all been used. Each of these measures provides different information regarding the characteristics of the distribution of LMX within work groups and imposes different requirements in modeling its impact empirically (Harrison & Klein, 2007). For this reason, it is vital that research select a measure of differentiation that is consistent with the nature and type of differentiation under investigation (Cole, Bedeian, Hirschfeld, & Vogel, 2011). We provide a fuller discussion of this issue and its implications for research within our Method section, below.
LLX
To date, the role of LLX has not been examined as a moderator in the relationship between LMX differentiation and group-level outcomes. The lack of attention to LLX is surprising because by facilitating flows of communication, influence, and rewards both between and within the groups, leaders’ upward exchange relationship with their own boss serves as an important “linking pin” mechanism through which the work group connects to the larger organization (Graen et al., 1977). The importance of understanding the upward exchange relationship as a critical contextual variable was emphasized in early work on vertical dyadic theory (Cashman et al., 1976; Graen et al.). Indeed, the quality of the LLX relationship was argued to serve as a critical “understructure” or “organizational climate” feature important to shaping the attitudes and performance of work group members (Graen et al.). Early research demonstrated that the LLX relationship was an important determinant of both material and social networks.
Although restricted access to material resources, as a function of relative quality of LLX, is easily understood, the relationship between LLX quality and stocks of social resources important to LMX quality is more complex. However, past research has offered that the availability of trust, support, and feedback would be limited because these resources have not been provided to the leader (Tangirala, Green, & Ramanujam, 2007; Zhou, Wang, Chen, & Shi, 2012). For example, leaders who are not trusted by their own supervisor would not be in a position to invest trust in others as a result of their own tenuous standing in the organization. Furthermore, evidence exists that supervisors often mimic the behaviors of their own bosses (e.g., Mayer, Kuenzi, Greenbaum, Bardes, & Salvador, 2009; Zohar & Luria, 2005), leading to “trickle down” leadership effects—making it unlikely that supervisors in whom little social/psychological resources are invested would in turn invest these resources in their own subordinates.
Thus, as Cashman and colleagues concluded, When a superior develops a [high-quality LLX] relationship with his boss those members reporting to such a superior share his good fortune. In contrast, when a superior fails to develop a [high-quality LLX] relationship with his boss, those members reporting to him also must suffer his misfortune. (1976: 293)
Furthermore, Cashman et al. found that this “fortune” or “misfortune” was experienced by the group members both tangibly (e.g., budgetary limitations) and intangibly (e.g., latitude and support). Therefore, high-quality upward exchanges should result in a relative abundance of resources that provide expanded opportunity for enriched LMX relationships between the leader and his or her subordinates (Tangirala et al., 2007; Zhou et al., 2012). Conversely, when leaders possess low-quality upward exchanges, they have fewer tangible and intangible resources to invest in the development of LMX relationships amongst their own subordinates (Cashman et al.; Graen et al., 1977; Tangirala et al.; Zhou et al.).
Despite the intuitive appeal and theoretic importance of understanding the nature and quality of the LMX relationship in the context of the LLX relationship, we are aware of only two empirical studies in which the LLX relationship was examined as a potential moderator. Both studies found that higher quality LLX relationships strengthen the main effects of LMX quality on employee individual attitudinal outcomes (Tangirala et al., 2007; Zhou et al., 2012). These findings were argued to be a consequence of the greater material and psychological resources that are brought by the supervisor to the individual-level LMX relationship. Consistent with social exchange theory, results from both studies showed improvements in the perceived value of the exchange by employees in relationship with higher LLX leaders (Tangirala et al.; Zhou et al.). Using this same logic, Zhou et al. also found support for the hypothesis that LLX is positively associated with group-level empowerment. While these findings underscore the importance of the LLX relationship in group members’ appraisal and reaction to their own LMX status, they do not provide direct insight into how relative levels of LLX might affect the consequences of LMX differentiation in group settings.
One reason for the lack of greater attention to LLX’s role at the group level is the absence of a coherent theoretical framework for understanding how different levels of LLX affect group-level functioning. In the following section, we employ relative deprivation theory to understand how LLX might shape employees’ cognitive appraisal of LMX disparity (LMXD) and lead to different conclusions regarding its effects on group-level outcomes for high- and low-LLX leaders.
Disparity in the Context of LLX
Relative deprivation theory attends to the affective reactions of employees when confronted with the denial of resources available to themselves or others in their environment (Bolino & Turnley, 2009). While relative deprivation theory’s empirical history is found primarily in the social psychology (e.g., Walker & Pettigrew, 1984) and political science (e.g., Abeles, 1976; Korpi, 1974) literatures, it is an especially useful framework for understanding the role of LLX because it accounts for the importance of perceptions of availability, salience, and value of resources, as well as resource distribution in group settings (Crosby, 1976; Folger & Martin, 1986).
According to relative deprivation theory, when individuals lack resources available to others, they experience a sense of deprivation that manifests in negative emotional and affective reactions (Crosby, 1976; Folger, 1986; Martin, 1981). The experience of deprivation, and consequent emotional reaction, is most pronounced when the denied resource is made more salient because it is possessed by others in their proximate environment (Crosby). This reasoning is consistent with the assertion of justice researchers that uneven allocation or treatment makes fairness assessments more salient in work groups (e.g., Colquitt, Zapata-Phelan, & Roberson, 2005; Roberson, 2006). The experienced deprivation is further amplified when a resource that is withheld from some group members is valued and the group members perceive no legitimate justification for its denial (Folger). Therefore, a foundational prediction of relative deprivation theory is that when valued resources are withheld from some individuals with a weak or nonexistent justification, these individuals will experience negative emotions and attitudes and a sense of detachment (Folger; Folger & Martin; Martin). Moreover, individuals have been found to be more sensitive to and aware of deprivation when it is a shared experience in a group context, leading to an increased likelihood for a group to coalesce and engage in cooperation action (Crosby).
As applied to the current study, relative deprivation theory suggests the possibility of very different group-level employee reactions to LMXD under high- and low-LLX conditions. In the high-LLX condition, the leader approaches the resource allocation decision from a position of relative abundance. By virtue of their high-LLX relationship with their own superior, these leaders bring both greater material and social stocks to the leadership of their work group (Goodwin, Bowler, & Whittington, 2009; Sparrowe & Liden, 1997; Tangirala et al., 2007; Zhou et al., 2012). Given the relative abundance of resources available to the leader when LLX is high, the decision to provide resources to a relative few employees may be more difficult to justify than when LLX is low. For example, a low-LLX manager with relatively few tangible or intangible resources to distribute can be expected by team members to make trade-offs—if one or a few employees receive a large number of resources from the manager, others must (by necessity) receive fewer. In contrast, the need for high-LLX managers to make trade-offs in allocating resources to employees may be less apparent to group members. In other words, if a leader possesses a well-stocked cache of relational and material resources, the denial of valued resources would not be seen by group members as a product of environmental necessity but as a result of managerial choice, which may be seen as more arbitrary, unnecessary, or unfair. At the same time, the presence of abundant resources is likely to raise group members’ expectations regarding the levels of resources to which they believe they are entitled (Hegtvedt, 2001; Sondak, Neale, & Pinkley, 1999). In addition, the perceived value associated with establishing a high-quality exchange relationship with a high-LLX leader is expected to increase—further exacerbating the sense of deprivation by group members who lack this high-quality exchange relationship.
Conversely, low-LLX managers approach the leadership of their group from a position of resource scarcity. Because a high-quality LMX relationship with a low-LLX leader may not be seen as valuable to group members, employees are less likely to prefer a high-quality relationship and, thus, they may be less likely to react to LMXD in their group with a sense of deprivation and jealousy (Goodwin et al., 2009; Venkataramani, Green, & Schleicher, 2010). Furthermore, the limited literature investigating allocation preferences in circumstances of abundance versus scarcity suggests that individuals may be less likely to attempt to maximize their own outcomes in circumstances of environmental scarcity, especially when the other beneficiaries are known to them and the scarcity is seen as a product of environmental conditions rather than member abuse (Hegtvedt, 2001; Sondak et al., 1999). Finally, research suggests that allocators are more likely to make competency-based allocation decisions in circumstances where resources are constrained than when resources are abundant (Rusbult, Campbell, & Price, 1990) and to adopt more stringent criteria on which to base the allocation (Ross & Ellard, 1986). This tendency may further affect the perceived legitimacy of within-group disparity when their leader’s LLX is low.
Hypotheses
As described, LMX differentiation has been theorized to have both positive and negative consequences to group functioning. First, researchers have argued that differentiation in LMX quality allows for the development of an emergent hierarchy important to the improvement of group coordination and efficiency (Dansereau et al., 1975; Dienesch & Liden, 1986; Graen, 1976; Liden et al., 1997). On the other hand, a commonly offered consequence of LMX differentiation is its potentially negative impact on group teamwork behavior via compromised fairness perceptions (Scandura, 1999; Sias & Jablin, 1995). Indeed, the dominant theoretical perspectives in organizational justice research suggest that norms of equality (i.e., equal treatment of all individuals) are important for social harmony within groups (e.g., Deutsch, 1975; Tyler, 1989; Tyler & Blader, 2000). When imbalances in the allocation of resources occur, there is a greater potential for enmity, jealousy, and competition within groups (Henderson et al.; Sherony & Green; Vecchio, 1995). As differences in LMX quality become more pronounced within work groups, the fairness of the distribution becomes a critical determinant of members’ responses (Erdogan & Bauer, 2010).
In this study, we focus specifically on how group members respond to differences in LMX quality in the form of intragroup teamwork behavior. Our measure of group teamwork behavior assesses the level of mutual support offered by team members in the performance of their work (Glaser et al., 1987). For example, the level of group teamwork behavior represents the degree to which members are willing to collectively communicate information, help one another, and engage in supportive coordinative actions not explicitly included by job requirements (Glaser et al.). Examples of voluntary teamwork behaviors include a variety of employee actions directed toward coworkers, such as orienting new employees, assisting others with heavy workloads, acting courteously towards coworkers, and engaging in actions that help to prevent or resolve interpersonal conflict (Organ, 1988). These behaviors represent discretionary behaviors on the part of the employee who would likely be affected by inequality in LMX relationships.
Given the inherent theoretical ambiguity and lack of consistent main effects reported in past research, we offer no a priori expectation related to the main effects of LMXD on teamwork behaviors. However, in the context of varying levels of LLX, relative deprivation theory informs our expectations related to the degree to which these comparative mechanisms might impair group functioning. Specifically, we propose that the extent to which intragroup disparity in the level of LMX is viewed as legitimate or fair by group members will vary on the basis of the level of LLX. When LLX is low, LMXD is expected to be perceived as more legitimate and a function of necessary trade-offs than under the high-LLX condition. On the other hand, in the high-LLX condition, the disparate allocation of resources would be more difficult to justify as necessary and legitimate. Deprived group members who believe that sufficient resources are available to meet their needs are expected to respond to their relative deprivation by being less likely to engage in behaviors supportive of the whole. Previous research provides indirect support for these assertions by demonstrating that justice climate moderates the relationship between LMX differentiation and group helping behavior such that it is most positive when levels of justice climate are high (Erdogan & Bauer, 2010).
The relative quality of LLX will affect the attractiveness or value associated with a high-quality relationship with a supervisor. In a high-LLX circumstance, a high-quality LMX relationship would likely be preferred, making the deprivation more acute. Alternatively, when LLX is low and LMX quality is less valued, we would expect that growing disparity would be less likely to evoke feelings of deprivation. For these reasons, we predict that the relative availability of perceived work group resources in the form of high and of low LLX will lead to different interpretations of the legitimacy and fairness of the differentiation decision and, therefore, play a significant moderating role in the impact of LMXD on group teamwork behaviors. Therefore, we expect that the theorized coordinative efficiencies associated with LMXD will be realized only in circumstances of low LLX.
Hypothesis 1: LLX will moderate the relationship between LMXD and group teamwork behavior such that the LMXD will be more positively related to group teamwork behavior in a low-LLX condition than in a high-LLX condition.
Considered at the group level, teamwork behavior has occupied a central position in most, if not all, of team effectiveness models (Gist, Locke, & Taylor, 1987; Guzzo & Shea, 1992; Hackman, 1983) and has served theoretically as a critical team-level process important to group effectiveness (McIntyre & Salas, 1995; Oser, McCallum, Salas, & Morgan, 1989; Stout, Cannon-Bowers, Salas, & Milanovich, 1999). For example, the input-process-output model emphasizes teamwork processes such as coordination, information sharing, team helping behavior, and problem solving as important determinants of the group’s effectiveness in transforming inputs into meaningful outcomes (e.g., Hackman & Morris, 1975; McGrath, 1964).
Meta-analytic examinations demonstrate consistent and meaningful positive relationships between measures of teamwork behavior and a variety of team effectiveness criteria (LePine, Piccolo, Jackson, Mathieu, & Saul, 2008). Organizational citizenship behavior directed towards teammates has also been shown to be positively associated with group-level effectiveness (Podsakoff, Whiting, Podsakoff, & Blume, 2009). Group cohesion, which reflects group members’ interpersonal attraction to one another and shared commitment to a task, has also been shown to be positively associated with group and team performance outcomes (Beal, Cohen, Burke, & McLendon, 2003; Carron & Brawley, 2000; Evans & Dion, 2012). Finally, stable negative relationships are reported in reviews of past research examining the impact of group-level relational and process conflict on group effectiveness outcomes (de Wit, Greer, & Jehn, 2012). For these reasons, there is substantial empirical and theoretical precedent for our expectation that teamwork behavior will be a critical intermediary mechanism that will be positively associated with overall team effectiveness. Therefore, we hypothesize:
Hypothesis 2: Aggregated group teamwork behavior will be positively associated with team effectiveness.
Given the central role of teamwork behavior as a process mechanism important to the transformation of inputs into team-level outcomes, we anticipate that the effects of LMXD on team effectiveness will be mediated by teamwork behavior. This expectation is consistent with LMX theory that suggests that LMX differentiation will have positive consequences on team effectiveness and performance through enhanced coordinative efficiencies (Dansereau et al., 1975; Dienesch & Liden, 1986; Graen, 1976; Liden et al., 1997). However, we do not expect the strength of this meditated relationship to be the same in high-LLX and low-LLX conditions. As described, we expect that the positive coordinative efficiencies theorized to emerge from LMX differentiation will be more likely to be realized in the low-LLX condition, in which disparity is perceived as more legitimate. Conversely, in the high-LLX condition, we expect that LMXD is less likely to have positive consequences on teamwork behavior. Because we expect that strength of any positive relationship between LMXD and teamwork will be diminished in the high-LLX condition, the strength of the meditated effect from LMXD to team effectiveness through teamwork will also be reduced in this condition. Therefore, we expect that the predicted mediated relationship from LMXD to team effectiveness through teamwork behavior will be conditional, such that:
Hypothesis 3: The strength of the mediated relationship between LMXD and team effectiveness through aggregated group teamwork behavior will be moderated by LLX.
Method
Research Setting
The study was conducted in China with the cooperation of seven organizations representing a diversity of industries, including manufacturing, high technology, health care, and hospitality. A total of 120 work groups were invited to participate in the study, with 95 groups consenting to participate. The sample included multiple types of work groups, such as administration, R&D, sales, manufacturing, and customer service.
Members of the research team in China visited each organization, administered the survey, and collected all survey data in order to ensure confidentiality of responses and minimize the burden of survey administration for the participating organizations. Because all measures were originally in English, we translated the questionnaires into Chinese and back-translated into English to check the accuracy of the translation (Brislin, 1980). Participation was voluntary, and survey administration was conducted during work time. To help minimize common source bias, we collected survey data from two different sources. Data on LMX and teamwork behavior were collected from employees in each work group, while work group managers provided data on the LMX with their boss (LLX) as well as data on team effectiveness.
To further minimize common method variance, we administered two waves of data collection (Time 1 and Time 2) 10 weeks apart. At Time 1, subordinates answered the questions about their relationship with their supervisor (LMX). At Time 2, employees were asked to complete the survey assessing teamwork behaviors, and supervisors were asked to assess the quality of their exchange relationship with their manager (LLX) as well as the effectiveness of their own work group. This process ensured that our mediating variable (teamwork behavior) and dependent variable (team effectiveness) were temporally separate from the collection of our independent variable (LMXD).
Of the collected surveys, we limited our sample to work groups with at least three subordinate responses, as well as groups whose members participated in both waves of data collection. The final sample consisted of 579 subordinates nested in 74 work groups for an average group size of 7.82 employees per group. Of the subordinate respondents, 38.6% were male. The average tenure was 4.78 years, and the age distribution of the sample was 31.3% under 25 years old, 42.3% between 25 and 34 years old, 16.2% between 35 and 44 years old, 8.6% above 45 years old, and 1.6% who gave no response regarding their age. Seventy-five managers responded to the LLX and team performance items, 56.8% of whom were male. The average tenure for managers was 12.26 years, and the age distribution of the sample was 8.1% under 25 years old, 33.8% between 25 and 34 years old, 43.2% between 35 and 44 years old, and 14.9% above 45 years old.
Measures
LMX/LLX
LMX and LLX were measured using LMX-7 (Graen & Uhl-Bien, 1995). LMX-7 includes items measuring three highly correlated dimensions, which include respect, trust, and obligation. These items have been validated and used in many previous LMX studies (Gajendran & Joshi, 2012; Scandura & Schriesheim, 1994; Schriesheim, Castro, & Yammarino, 2000). Sample items included, “How well does your supervisor understand your job problems and needs?” and “How well does your supervisor recognize your potential?” Respondents were asked to respond to these items in reference to the person to whom they reported directly. Subordinates’ responses were used to create LMX, whereas managers’ responses were used to create LLX. Responses ranged from strongly disagree (1) to strongly agree (5). Cronbach’s alpha for this measure was .85 and .76 for LMX and LLX, respectively.
LMXD
As noted earlier, a striking feature of the extant research examining LMX differentiation is the variability across research studies in the operational definition of LMX differentiation—including within-group variance (Erdogan & Bauer, 2010; Liden et al., 2006), agreement indices (Boies & Howell, 2006), and measures of disparity (Hooper & Martin, 2008; Wu et al., 2010). Each of these operational definitions represents different assumptions regarding the meaning and underlying distribution of the within-group differentiation (Harrison & Klein, 2007). When the selected measure of differentiation is inconsistent with the nature and type of differentiation under investigation, the detection of the focal effect is more difficult (Cole et al., 2011). Therefore, empirical precision and theoretical progress is possible only when research explicitly describes, substantiates, and captures the compositional pattern of the theorized distribution being investigated (Harrison & Klein). For example, differentiation conceptualized as disparity indicates “differences in concentration of valued social assets or resources among unit members . . . that, at their extreme, privilege a few over many” (Harrison & Klein, 2007: 1200). Commonly measured using the coefficient of variation (CV), disparity is maximized in circumstances where valued social or material resources are offered to a small minority of group members (Harrison & Klein). For this reason, disparity (as measured by CV) is commonly employed in research directed towards understanding the distribution of pay and other resources (e.g., Bloom, 1999; Pfeffer & Langton, 1993; Siegel & Hambrick, 2005). In contrast, differentiation conceptualized as separation or agreement (e.g., within-group variance) indicates variability and is maximized when the distribution is equally split at opposite ends of the distribution continuum (Harrison & Klein).
We propose that the conceptualization of group-level LMX differentiation as disparity is the most appropriate way to understand the impact of LMX differentiation on group-level outcomes. For example, arguments that efficiency gains are achieved through the development of managerial surrogates presume that high-quality LMX relationships are developed with only a small number of employees (Dansereau et al., 1975). At the same time, providing high-quality LMX to only a few group members is the decision most likely to provoke feelings of injustice among the less privileged majority (Scandura, 1999; Sias & Jablin, 1995; Tyler, 1989; Tyler & Blader, 2000). Therefore, the tension between the beneficial and harmful effects of LMX differentiation is likely to be most acute when LMX quality is allocated to a relative few within the work group, to the exclusion of the majority. Measures of disparity (CV) are maximized in this type of circumstance.
Prior to using CV, we conducted a number of tests to determine whether CV could be justified empirically in this case. The reason for additional empirical tests is that CV is measured as a ratio (group-level standard deviation/group-level mean). Therefore, one must determine whether the distinct parts of the ratio (namely, standard deviation and 1/mean) as well as their interaction are jointly significant (Bradshaw & Radbill, 1987; Firebaugh & Gibbs, 1985; Harrison & Klien, 2007; Liermann, Steel, Rosing, & Guttorp, 2004; Liu & Schutz, 2003; Schuessler, 1974). To determine the level of joint significance, we conducted the Wald test with (1) standard deviation of group-level LMX, (2) 1/mean of group-level LMX, and (3) the CV of group-level LMX included in the regression analysis. We found that group-level standard deviation of LMX and 1/mean of LMX were jointly significant predictors of the criterion variable (teamwork, in this case), whereas CV (the interaction term) was not a significant predictor when the other two variables were included in the regression model. On the basis of the results of this test, we concluded that using CV as a measure of LMXD was statistically inappropriate in our case.
Post hoc analysis revealed that the reason for that lack of additional variance explained by CV after standard deviation and 1/LMX mean are entered into the ordinary least squares regression equation is that there is a nearly perfect correlation between CV and standard deviation in our sample (r = .97). Thus, standard deviation and CV were found to provide nearly identical (redundant) information concerning the distribution of LMX in our sample. Further examination revealed that the reason for the high correlation between group-level LMX variation measured using standard deviation and CV is that the vast majority of our groups had average LMX scores in the middle of the 1-to-5 range of possible responses for the LMX variable (M = 3.5, SD = 0.61, min = 2.71, max = 4.75). Censoring at the high and low ends of the group-level LMX distribution causes differences between LMXD measured as CV and standard deviation to be minimized and correlations between these measures to be high. In contrast, for group-level LMX mean distributions that are either normally distributed or skewed to the right (higher average LMX scores with a few low scores), which appear to be more common in previous research using U.S. samples, we would expect weaker correlations between standard deviation and CV. When empirical differences between CV and standard deviation scores are present, it is both theoretically and empirically most appropriate to measure LMXD as CV in order to support the conceptual arguments regarding LMXD as disparity that we presented earlier. In our case, however, we find that the use of either standard deviation or CV as a measure of LMXD leads to identical empirical results and, thus, are empirically interchangeable. Because we are unable to empirically support using CV (a ratio measure) in our analysis, we chose to operationalize group-level LMXD using standard deviation.
Teamwork behavior
Teamwork behavior was assessed using a six-item scale developed by Glaser et al. (1987) to measure perceived group-level coworker communication, cooperation, and helping behavior. This measure is conceptually consistent with teamwork measures used in recent studies (e.g., Eby, Meade, Parisi, & Douthitt, 1999; Glaser et al.). Sample items included, “People I work with are cooperative,” “People I work with resolve disagreements well,” and “People in my department help each other with problems at work.” For each item, subordinates were asked to indicate the degree to which they agreed with the statement for the individuals they work with directly; responses ranged from strongly disagree (1) to strongly agree (5). Cronbach’s alpha for this measure was .84. We assessed agreement levels for this measure using recommended agreement indices—Rwg and intraclass correlations ICC(1) and ICC(2) (Bliese, 1998, 2000). The mean Rwg was .86, and ICC(1) and ICC(2) were .38 and .83, respectively, showing strong within-group agreement.
Team effectiveness
Four items were adapted from the scale developed by Campion, Papper, and Medsker (1996) to measure team effectiveness. We decided to use only these four items to measure team effectiveness because the original measure includes several items not appropriate for our sample. Managers’ responses were used to assess the level of perceived effectiveness of the work group they supervise. Sample items included, “Please indicate the quality of work done by your team,” “Please indicate how quickly your team responds to problems and opportunities,” and “Please indicate the overall performance level of your team.” Responses ranged from strongly disagree (1) to strongly agree (5), and the Cronbach’s alpha for this measure was .85.
Control variables
In order to rule out alternative explanations for our findings and assess the degree to which LMXD, LLX, and the LMX × LLX interaction account for variance in our outcome measures beyond the variance accounted for by other potential explanatory variables, we included several control variables in our analyses. First, we included the average age of team members because it may be associated with their experience in working within team-based work structures and, therefore, related to teamwork and team effectiveness. The average educational level of team members was included as a control because it may be associated with their understanding of and ability to effectively apply strategies for team functioning and coordination. The average tenure of the team members was included as a control because the amount of time team members have spent interacting with their supervisor and one another may be related to teamwork and team effectiveness (Liden et al., 2006). Group size was included as a control because it may affect a leader’s need to differentiate as well as the level of complexity of task coordination within the team. Because group-level LMX mean has been found to be an important predictor of group effectiveness in previous studies (Boies & Howell, 2006; Liden et al.; Le Blanc & González-Romá, 2012), we included group-level LMX mean as a control variable in this study. Including the group mean as a control variable is also consistent with previous recommendations that have been provided for using standard deviation as a measure of disparity in empirical analyses (Harrison & Klien, 2007).
Results
Zero-order correlations, means, and standard deviations for each of the study variables are provided in Table 1.
Correlations and Descriptive Statistics
Note: LMX = leader-member exchange; LMXD = leader-member exchange disparity; LLX = leader-leader exchange; T1 = measured in Time 1; T2 = measured in Time 2. N = 74.
These variables are measured in ordinal scale.
p < .05 (two-tailed).
p < .01 (two-tailed).
To test the hypothesized moderating effect of LLX, we used hierarchical linear regression analysis (Aiken & West, 1991). In Step 1, control variables were entered into the prediction model (Model 1). In Step 2, LMXD and LLX were entered into the model (Model 2). These variables produced an R-squared of .37 for teamwork behavior. The significant portion of variance accounted for by these variables was attributable mostly to the robust relationship of LMX mean (β = 0.39, p < .01) and LMXD (β = 0.68, p < .01) with teamwork behavior. In Step 3, the interaction term (LMXD × LLX) was entered into the model. The introduction of the interaction term produced a statistically significant change in R-squared of .08 (p < .01), providing support for Hypothesis 1.
In order to understand the nature of these moderating effects, we plotted the slopes of the prediction line for the hypothesized relationship at 1 SD above and below the mean for LLX (Aiken & West, 1991). As shown in Figure 1, consistent with our expectations, the relationship between LMXD and teamwork behavior was more strongly positive in the low-LLX condition. It is also worth noting that while no a priori expectation was offered regarding the main effect of LMX differentiation on teamwork, LMXD was positively and significantly (p < .01) related to teamwork behavior at the mean level of LLX. In the high-LLX condition, the slope of this line was not found to be significantly different from zero (p > .05).

Leader-Leader Exchange as a Moderator of the Relationship Between Leader-Member Exchange Disparity and Teamwork Behavior
Models 4 and 5 in Table 2 show the results of testing Hypothesis 2. This hypothesis predicted a positive relationship between teamwork behavior and team effectiveness. In Model 4, we entered all control variables into the model, producing an R-squared of .14, which was statistically insignificant (p > .05). In Model 5, teamwork was a statistically significant predictor of team effectiveness (β = 0.44, p < .05). Adding this variable to Model 3 explained an additional 7% of the variance. Thus, Hypothesis 2 was supported.
Hierarchical Regression Analysis Testing Hypotheses 1 and 2
Note: LMX = leader-member exchange; LMXD = leader-member exchange disparity; LLX = leader-leader exchange. N = 74.
These variables are measured in ordinal scale.
p < .05 (two-tailed).
p < .01 (two-tailed).
p < .001 (two-tailed).
Hypothesis 3 suggested that the indirect effect of LMXD on team effectiveness through teamwork behavior is moderated by LLX. To test the hypothesis, we ran a moderated-mediation analysis (Edwards & Lambert, 2007; Muller, Judd, & Yzerbyt, 2005; Preacher, Rucker, & Hayes, 2007), which provides the advantage of testing moderation and mediation in a single model. This methodology allows us to test for a conditional mediated effect in which the effect of a mediating variable between two other variables is dependent on levels of a fourth variable (e.g., Baron & Kenny, 1986; Edwards & Lambert; Hayes, 2013; James & Brett, 1984). Additionally, moderated-mediation analysis can provide more accurate estimations when used with bootstrap analysis. Studies comparing multiple methods that can be used to test mediation analysis suggested that bootstrap analysis can adjust for nonnormal distribution of mediated effects (MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002) and provides the most effective and logical way to obtain confidence limits of indirect effects for most conditions (Preacher & Hayes, 2008). In the present study, the bootstrap analysis involved the generation of 5,000 samples with replacement to produce confidence intervals for three indirect effects.
Table 3 shows the results of the test of Hypothesis 3. The model includes all control variables used in previous models. Indirect effects of LMXD were calculated at three different levels of LLX (low: 1 SD below the mean; medium: mean; high: 1 SD above the mean). Confidence intervals (95%) of the three indirect effects were estimated with bootstrap analysis (using 5,000 samples with replacement). The results showed statistically significant indirect effects at low (0.67) and medium (0.35) conditions of LLX. The statistical significance of these results is confirmed by the fact that the 95% confidence intervals for these effects calculated using the bootstrap analysis do not include zero. In contrast, the 95% confidence interval for the mediated effect in the high-LLX condition does include zero, indicating an insignificant indirect effect under this condition. Thus, our results in Table 3 support Hypothesis 3 by showing that strength of the mediated relationship between LMXD and team effectiveness through aggregated teamwork behavior varies on the basis of the level of LLX. Specifically, a significant positive mediated effect is found in the low-LLX condition but not in the high-LLX condition.
Bootstrap Analysis Result for Conditional Indirect Effect of LMXD on Team Effectiveness at Different Levels of LLX
Note: LMXD = leader-member exchange disparity; LLX = leader-leader exchange. CI = 95% confidence interval (two-tailed). Bootstrap sample size = 5,000; N = 74.
In order to empirically determine the effect that including control variables had on our results, we reran the analyses without control variables. We found that excluding the control variables did not substantively change our reported results. However, because the purpose of including control variables in our study was to determine whether LMX dispersion, LLX, and the LMX × LLX interaction provided incremental prediction beyond that explained by our control variables, we judged that including control variables as the first step in our hierarchical regression models and calculating the incremental change in R-squared associated with adding our independent variables was appropriate in this case (Carlson & Wu, 2012).
Discussion
Consistent with the predictions from relative deprivation theory, results showed that LLX moderated the relationship between LMXD and group-level teamwork behavior. Specifically, these results suggest increasing levels of disparity are associated with more cooperative teamwork behavior and greater team effectiveness when LLX is low. Because their leaders are themselves deprived of resources under low LLX, disparity is more easily justified and group members are less likely to experience a sense of deprivation and disenfranchisement. As a result, efficiencies from coordination among team members theorized to emanate from LMXD are more likely to be realized under low LLX (Dansereau et al., 1975; Liden et al., 1997). In contrast, we find that the mediated relationship between LMXD and team effectiveness through teamwork behavior is slightly negative (and not statistically significant) under the high-LLX condition.
While no expectation regarding main effect between LMXD and teamwork behavior was hypothesized, we observed a positive main effect in this sample. This finding is consistent with past research examining the impact of LMX differentiation on group-level helping behavior (Erdogan & Bauer, 2010). Consistent with theoretic logic offered in our own study is Erdogan and Bauer’s argument that perceptions of justice climate would moderate the relationship between LMX differentiation and group outcomes. In contrast to their attitudinal outcomes in which a significant negative relationship was found between LMX differentiation and group attitudinal outcomes in the low justice condition, they find no significant relationship in the low justice climate condition for their teamwork outcome. Moreover, they report a significant positive main effect between LMX differentiation and group helping behavior and that this positive effect was strengthened in the high justice climate condition (our low-LLX condition).
Relative deprivation theory provides a possible explanation for why researchers might find a positive main effect between LMXD and teamwork as well as a negative relationship between LMXD and aggregate employee attitudes towards the group leader and/or organization. According to relative deprivation theory, one consequence of increasing the proportion of group members who experience relative deprivation in a group context is that deprived members see themselves as sharing a common fate as “have-nots” while a relatively small number of elites control a large proportion of resources (Dansereau et al., 1975; Graen, 1976). Social psychologists have found that that perceived common fate among out-group members leads to increased cohesion within this subgroup (Crosby, 1976; Martin, 1981). As LMXD increases (and the deprived minority becomes larger), employees are expected to experience a sense of connection with fellow group members and to engage in behavior supportive of the team or group.
In contrast to teamwork behavior, which is based on employee relationships with peers, employee attitudes, such as job satisfaction or organizational commitment, are expected to be more reflective of employees’ perceptions of their job and/or the organization. It is logical to expect that the average levels of employees’ attitudes toward the leader or the organization would decrease as LMXD increases and the number of employees who find themselves in a deprived condition (i.e., low LMX) increases. Thus, relative deprivation theory helps to explain why group-level LMXD (measured as CV) may have a positive main effect on the aggregate teamwork behavior but may also have a negative relationship with average levels of employees’ attitudes as found in Erdogan and Bauer (2010).
Theoretical and Practical Implications
This research has several important implications on future LMX theory development. First, this study is only the third in the past several decades to examine LLX as an important contextual variable. Both previous studies showed that LLX is important to the cognitive appraisal of LMX quality such that higher levels of LLX strengthen the main effects of LMX on individual attitudinal outcomes (Tangirala et al., 2007; Zhou et al., 2012). We add to these previous studies by demonstrating that LLX is also an important contextual variable in shaping the group-level dynamics and outcomes. Together, these findings support the idea that a leader’s LLX status is a salient attribute of the work environment (Duchon et al., 1986; Venkataramani et al., 2010). We believe that theoretical progress would be well served by continued attention to the role LLX quality has in shaping the interpretation and subsequent effects of LMX and LMXD.
Second, this research contributes to our understanding of the effects of LMXD by establishing LLX as an important boundary condition. These findings suggest that aggregate affective and behavioral responses to LMXD differ in high- and low-LLX conditions. Because these effects were argued to operate, at least in part, through an employee’s appraisal of the relative fairness of the differentiation decision, this work extends findings that justice climate is an important moderating condition (Erdogan & Bauer, 2010) and identifies LLX as a potential determinant of these collective justice perceptions. Future research directed towards identifying other characteristics of the work group and their environments that affect the appraisal of fairness is necessary.
This research, and the use of relative deprivation theory, also moves beyond dyadic LMX theory and provides important insight into how the group-level comparative processes, employee preferences, and coworker dynamics might evolve in circumstances of high and low LLX. For example, given the diminished value associated with a high-quality LMX relationship with a low-LLX boss, our results are consistent with expectations derived from relative deprivation theory that (1) there will be a weaker adverse response to LMXD within work groups under a low-LLX condition and (2) there will be a shift in orientation of employees under this condition away from developing a high-quality relationship with the leader and more towards developing high-quality relationships with coworkers. Therefore, further exploring the interplay between LMX quality and the quality of team member exchange (Seers, 1989) relationships represents an important future research pursuit. We also contend that in the context of high disparity, group leaders will be more selective in their choice of surrogates—likely identifying those who are both capable and, especially in a low-LLX context, hold a strong preference for the benefits and privileges associated with their minority status. In order to test this contention, future research could directly assess the characteristics and attitudes of individuals who report a high-LMX relationship with a low-LLX leader in the context of a high-disparity group.
Although not central to this investigation, we found that LLX level was unrelated to LMXD in our sample. That is, in this sample, a leader’s LLX status was not associated with a tendency for a leader to engage in more or less LMX differentiation. Future research that seeks to understand the characteristics of the leader and the organization that influence his or her decision to differentiate represents an important line of inquiry. For example, certain disposition characteristics may be important determinants in differentiation tendencies. Furthermore, LLX relationships are embedded within a network of other LLX relationships (Sparrowe & Liden, 1997). Within this network of LLX relationships, the extent that leaders differentiate among their subordinates may be influenced by the level of LLX disparity they experience as well as their LLX status relative to their peers. Furthermore, there may be systematic differences in the capacity for low- and high-LLX leaders to develop high-quality LMX with more competent employees as a result of employee perceptions of diminished, or even negative, value associated with a relationship with a resource deprived superior. In short, examining the rich network of upward exchange relationships within an organization represents fertile ground for future research (Goodwin et al., 2009; Sparrowe & Liden).
Turning to the practical implications for organizations, this study contributes to an emerging body of research examining the effects of LMX differentiation and supports the notion that its effects are shaped by a number of important moderating conditions. Thus, the search for a simple response to the normative question “Should leaders differentiate?” seems to be an increasingly futile pursuit. The findings reported here suggest that as organizations endeavor to enhance team effectiveness, attention to a leader’s relationship to his or her own leader is warranted. A clear implication of this research is that leaders do not operate in a vacuum but are enabled and constrained by situational characteristics. To help promote cooperative and productive team functioning, low-LLX leaders are best served by high-LMX differentiation among subordinates. Conversely, low-LLX leaders who fail to differentiate may compromise cooperative team interactions and reduce team effectiveness.
Limitations
Data from the study were collected from work groups in a diverse set of organizations in mainland China. Chinese culture is known as highly collective because of its historical context, which is rooted in socialism and Confucianism (Yang, 2012). Socialism emphasizes the value of common wealth and discourages unequal material incentives, and Confucianism stresses social harmony and establishment of benevolent policy favoring the collective. In the context of these social and organizational values, a supervisor’s high relational differentiation across subordinates might increase the feelings of deprivation of subordinates when resources are abundant, eventually leading to the development of more unified and cohesive groups. Thus, it is possible that the cultural characteristics of our Chinese sample may limit the generalizability of findings outside of the Chinese context.
A second limitation of the study is the fact that our measure of LMX is endogenous. For this reason, we are unable to rule out the possibility that our results may be attributable to some unmeasured common cause and, therefore, reflect biased estimates of the effects of LMX differentiation (Antonakis, Bendahan, Jacquart, & Lalive, 2010). Therefore, it is critical that future research that seeks to replicate and extend our findings attend to the development of prediction models that avoid issues of endogeneity through development of objectively derived instrumental variables that might be used to validate measures of LMX (Antonakis et al.; Shaver, 1998). For example, collecting actual performance data of subordinates would be one way to mitigate issues around endogeneity. Arguably, the distribution of performance levels would be related to the LMXD measure because LMX literature suggested that subordinates’ job performance is a covariate of LMX (Dulebohn et al., 2012; Gerstner & Day, 1997). However, we believe that there are not a priori reasons to believe that distributions of performance levels could relate to teamwork behaviors within a work group. Alternatively, collecting information about individual traits of supervisors could be another way to mitigate endogeneity issues. Stable individual difference, such as personality or cognitive ability, could be used as an instrumental variable (Antonakis, Day, & Schyns, 2012).
In this study, we offered that disparity (CV) provides the most theoretically appropriate measure of differentiation because it allows us to directly assess the circumstance we sought to understand—that is, a work group composed of a relative few beneficiaries of high-quality LMX, to the exclusion of the majority. However, as a ratio variable, CV imposes specific requirements on the development of regression models (e.g., inclusion of both standard deviation and 1/mean as controls; Bradshaw & Radbill, 1987; Firebaugh & Gibbs, 1985; Liermann et al., 2004; Liu & Schutz, 2003; Schuessler, 1974). In our data, CV and standard deviation were nearly perfectly correlated as a result of a relative few instances of high disparity (i.e., one high and all others low LMX), making the detection of variance attributable to CV over and above standard deviation impossible. Therefore, we were unable to use CV in our test of hypotheses. Because they were found to be interchangeable in our sample, we believe the use of standard deviation was justified and represents the within-group LMX distribution important to testing the predictions of relative deprivation theory. Nevertheless, we believe that disparity (assessed using CV) is especially appropriate to LMX differentiation research and theory at the group level and future research would be well served, as provided here, by applying theoretical and empirical rigor in operationalizing and modeling the effects of differentiation. Doing so is critical to guaranteeing confidence in finding and ensuring the development of a cumulative understanding across studies.
Another limitation of the study is the fact that no measures of justice or fairness perception were collected. Because the predictions of relative deprivation theory are grounded in an individual’s appraisal of fairness, our ability to explicitly test predictions from this theory is limited. Thus, while we believe that the data support the theoretic logic of our conclusions, it is not possible to definitively isolate fairness perceptions as the causal mechanism. Future research designs should collect measures of fairness to further validate these findings.
Finally, the measure of LMX/LLX employed in this study was unidimensional and intended to capture the entirety of the exchange relationship. Research has demonstrated that LMX is multidimensional, with affective, loyalty, contribution, and professional respect dimensions (Greguras & Ford, 2006; Liden & Maslyn, 1998). While the use of a single unidimensional or composite measure of LMX is consistent with past LMX differentiation research, future research should seek to understand whether more narrowly defined LMX dimensions might affect the results reported here (Greguras & Ford).
Conclusion
This study responds to calls for LMX research and theory to account for the influence of complex social forces and environmental constraints at the group level (Graen & Uhl-Bien, 1995; Henderson et al., 2009; House & Aditya, 1997; Liden et al., 1997; Scandura, 1999). Because differentiated LMX relationships have been found to be the rule rather than the exception (Liden & Graen, 1980), a central question confronting the LMX literature is how within-group LMX differentiation affects important group-level outcomes (Henderson et al.). Our use of relative deprivation theory in this study contributes to this understanding by introducing LLX as an important moderating influence on these effects. Our findings offer new insight into the role of the LLX relationship as an important boundary condition for predicting the consequences of LMX differentiation. In doing so, we further efforts to develop a more complete theory of LMX that extends beyond the leader-member dyad.
Footnotes
Acknowledgements
This article was accepted under the editorship of Deborah E. Rupp. The authors would like to acknowledge the many valuable insights and suggestions offered by the reviewers to improve the manuscript. The second author is now at the Hankuk University of Foreign Studies, Korea.
