Received: 13 January 2017 Revised: 1 November 2019 Accepted: 8 December 2019 DOI: 10.1002/job.2423 RESEARCH ARTICLE Avoiding or embracing social relationships? A conservation of resources perspective of leader narcissism, leader–member exchange differentiation, and follower voice 1Department of Management, Raymond J. Harbert College of Business, Auburn University, Auburn, Alabama, U.S.A. 2Department of Management, College of Business, University of Texas at San Antonio, San Antonio, Texas, U.S.A. 3Department of Management, Culverhouse College of Commerce, University of Alabama, Tuscaloosa, Alabama, U.S.A. Correspondence Lei Huang, Department of Management, Raymond J. Harbert College of Business, Auburn University, Auburn, AL, U.S.A. Email: lhuangmgmt@auburn.edu Summary In this study, we draw from the conservation of resources theory and the narcissism literature to examine why and when narcissistic leaders develop and maintain differentiated social relationships with followers in a group setting, therefore demotivating follower voice. Using data from 457 employees and their 95 supervisors working at a large Chinese consulting company, we tested and found support for our hypotheses that leader narcissism had a negative direct effect on employee voice, as well as a negative indirect effect on voice via group-level leader–member exchange (LMX) differentiation. Our findings further showed that leaders' upward exchange, leader–leader exchange (LLX), with their own supervisor moderated the negative indirect effect of narcissism on voice such that this negative indirect effect was stronger in the presence of low leader LLX but turned nonsignificant in the presence of high leader LLX. Theoretical and practical implications of our research are discussed. Limitations and directions for future research are also offered. 1 KE YWOR DS | INTRODUCTI ON Huang, & Harms, 2018) or evenvoice showing a humble attitude leader-leader exchange, leader narcissism, LMX differentiation, (Owens, Wallace, & Waldman, 2015). Nonetheless, it remains theoretically underdeveloped and empirically underexplored Narcissists, individuals with a heightened sense of selfas to why and when such deviation is more likely to occur. importance, extreme self-confidence, and an excessive need Consequently, it is important to theoretically and empirically for admiration from others (Emmons, 1987; Paulhus & Williams, examine reasons why narcissistic leaders may behave in a way 2002; Raskin, Novacek, & Hogan, 1991), all too often loom large counter to their narcissistic tendencies, and the boundary in terms of their impact on the lives of others. In the conditions under which narcissistic leaders may diverge from organizational research, increasing attention has been paid to their trait-consistent behavioral patterns while interacting with the largely detrimental impact of narcissistic leaders on follower followers. outcomes (Judge, Piccolo, & Kosalka, 2009; Resick, Whitman, In order to reconcile these conflicting accounts of the Weingarden, & Hiller, 2009). Although narcissistic leaders are effects of leader narcissism, we draw from the conservation of some- times perceived as charismatic or even charming resources (COR) theory (Hobfoll, 1989) and the narcissism (Galvin, Waldman, & Balthazard, 2010; Rosenthal & Pittinsky, literature to propose a moderated mediation model 2006), they frequently fail to develop and maintain social delineating why and under what conditions narcissistic leaders relationships with their followers (Ong, Roberts, Arthur, develop highly differentiated social relationships with their Woodman, & Akehurst, 2016). That said, an emergent stream of followers in a group setting and further impact follower organizational scholarship has suggested that there are outcomes. COR theory posits that individuals are motivated to instances when narcissistic leaders deviate from their selfcon- serve their limited resources or to acquire new resources centered tendencies by involving their followers in decisionin order to making (Carnevale, J Organ Behav. 2020; 41:77–92. © 2019 John Wiley & Sons, Ltd. 77 HUANG ET AL. 78 protect themselves against potential threat of resource loss (Hobfoll, 1989). Drawing from COR theory, we argue that due to their over- whelming concern with self-interest, narcissistic leaders may choose to conserve their limited resources (i.e., resource conservation), such as time and energy, by choosing to interact with selected followers instead of engaging all of their followers equally. We also argue that narcissistic leaders may seek to acquire resources (i.e., resource acqui- sition) that are valuable to them by building relationships with those followers who will provide such resources. For instance, they may prefer interacting with the “in-group” followers to hear more praise and feed their inflated ego. This is consistent with writings in the nar- cissism literature such that narcissistic individuals tend to selectively interact with people who can reinforce their grandiose self-image (Campbell & Campbell, 2009). Therefore, we expect that narcissistic leaders are more likely to create a differentiated relational environ- ment that is manifested in higher level of leader–member exchange (LMX) differentiation, a process by which leaders and followers engage in differing types of exchange patterns to form relationships that vary in quality (Liden, Erdogan, Wayne, & Sparrowe, 2006). We further propose that this relational environment, character- ized by higher level of LMX differentiation, will demotivate follower voice, which refers to employees speaking up with improvement- oriented ideas and suggestions (Van Dyne & LePine, 1998). We chose to focus on voice, which is challenging in nature (Morrison, 2011), as a follower outcome because it reflects followers' assessment of the social relational environment created by the leader (Burris, Detert, & Chiaburu, 2008). Previous research has found that leader narcissism negatively influences follower voice via follower's threatened organization-based self-esteem (OBSE; Carnevale et al., 2018). In the present study, we propose LMX differentiation as another important mediating mechanism linking leader narcissism and follower voice. Given that narcissistic leaders have strong egos and demand obedi- ence (Maccoby, 2000), we argue that narcissistic leaders are less likely to be receptive to follower voice and therefore should suppress follower voice directly. We also expect an indirect effect of leader narcissism on follower voice via LMX differentiation, such that in the context of high LMX differentiation, those in high-LMX relationships are more likely to refrain from speaking up but instead defer to their narcissistic leader to feed his/her strong ego, whereas those in low-LMX relationships are less likely to speak up due to the lack of felt obligation to contrib- ute or fear of harmful consequences. Drawing from COR theory (Hobfoll, 1989), we also argue that the salience of the threat of potential resource loss will influence how narcissistic leaders choose to develop relationships with their fol- lowers. Specifically, we propose that the negative indirect effect of leader narcissism on follower voice via LMX differentiation is stronger when narcissistic leaders are threatened by their poor upward exchange relationships with their own superior and thus have to rely more on their relationships with their followers. That is, the way nar- cissistic leaders develop relationships with followers depends on a broader relational context that also involves the upward social exchange between leaders and their superior (i.e., leader– leader exchange, LLX; Tangirala, Green, & Ramanujam, 2007). This upward exchange relationship offers a larger context where leaders are influenced by not only the way they connect with their followers, but also the benefits or losses they acquire in relationships with their own superior (Henderson, Liden, Glibkowski, & Chaudhry, 2009; Herdman, Yang, & Arthur, 2017; Tangirala et al., 2007; Zhou, Wang, Chen, & Shi, 2012). Thus, we argue that when narcissistic leaders have low (vs. high) LLX with their superior and thus are lacking access to resources that come with higher LLX, they are more likely to attempt to compensate for this resource loss by seeking support from followers in the group who can contribute to their leadership success. This is con- sistent with the COR literature, which suggests that individuals are motivated to acquire resources to reinforce their status of resource possession (Halbesleben, Harvey, & Bolino, 2009). Our study makes three contributions to narcissism, leadership, and voice literatures. First, while the mechanisms underlying the impact of leader narcissism on employee outcomes are theoretically underdeveloped and frequently empirically insufficient, we draw from the COR perspective (Hobfoll, 1989) and the narcissism literature to theoretically explore and empirically test a mediating mechanism (i.e., LMX differentiation) linking leader narcissism with follower voice. Second, we answer the call from Morrison (2011) to further examine how leader characteristics, other than the commonly studied leader openness (Detert & Burris, 2007), play a role in influencing employee voice. By so doing, our study also answers the call from Farh and Chen (2014) to examine the “dark side” of leader influence on employee voice, thereby helping shed light on how leaders suppress follower voice, a topic largely overlooked in the existing voice litera- ture. Our study thus further helps to clarify the nomological network of leaderrelated antecedents of voice. Finally, although recent research has suggested that narcissistic leaders may demonstrate behaviors different from their trait-consistent behavioral patterns (Carnevale et al., 2018; Owens et al., 2015), our study is among the first to look into why and when such divergence happens by investi- gating narcissistic leaders' upward social exchange with their own superior as an important boundary condition. We present our research model in Figure 1. 2 | THE NATURE OF LE ADER N ARCI SSI SM Unlike more favorable leader characteristics (e.g., extraversion, agree- ableness), which help foster productive interpersonal relationships, the characteristics associated with narcissistic leaders are typically viewed in a negative light. For example, narcissistic leaders tend to be arrogant with strong feelings of entitlement and frequently react harshly when they perceive that their authority has been challenged (Grijalva & Harms, 2014; Judge et al., 2009; Maccoby, 2000; Raskin et al., 1991; Spain, Harms, & LeBreton, 2014). They also frequently attempt to manipulate their followers' attitudes and behaviors, treating them as pawns or instruments to be used for acquiring resources that could help achieve their success (Morf & Rhodewalt, 2001; O’Reilly, Doerr, Caldwell, & Chatman, 2014). HUANG ET AL. 79 FI G U R E 1 Proposed model of leader narcissism and employee voice in a relational context. Notes. (1) The arrow from leader narcissism to employee voice denotes Path C in the mediation model (i.e., the direct effect of narcissism on voice). This arrow is dotted to distinguish it from Path C0 (i.e., the effect of leader narcissism on employee voice with the mediator, leader–leader exchange differentiation, in the model). (2) Hypothesis 3 predicts mediation. Given that the predicted indirect effect of leader narcissism cannot be captured by a single arrow, we visualize this hypothesis using the dark grey rectangle denoted as H3. (3) Hypothesis 5 predicts moderated mediation that also cannot be captured by a single arrow. Thus, we visualize this hypothesis using the light grey shaded rectangle denoted as H5 In the organizational context, there are mixed findings about how narcissistic leaders interact with their followers. On the one hand, a great deal of research has suggested that narcissistic leaders often have difficulty developing and maintaining interpersonal relationships with followers as they do not care for or trust others (Campbell, Hoff- man, Campbell, & Marchisio, 2011). They are self-centered and arro- gant, which prevents them from realizing key problems in the work unit, foreseeing necessary changes to be taken for the sake of group performance, or accepting followers' suggestions or ideas intended to help the organization (Nevicka, Ten Velden, De Hoogh, & Van Vianen, 2011). On the other hand, recent studies suggest that narcissistic leaders might sometimes deviate from their narcissistic tendencies and instead consult with followers when making work-related decisions (Carnevale et al., 2018) or even show a humble attitude (Owens et al., 2015). These mixed findings indicate that there might also be variations in the way narcissistic leaders choose to interact with their followers. 3 | LE ADER N ARCI SSI SM AND FOLLOWER VOICE To speak up or to remain silent is a decision that often involves fol- lowers' calculative consideration, as their voice might not always be favorably perceived by leaders (Burris, 2012). This is particularly true when leaders have a strong ego and might therefore feel threatened when followers speak up to challenge the status quo (Fast, Burris, & Bartel, 2014). Previous research has found that followers are less likely to speak up to a narcissistic leader due to their threatened OBSE (Carnevale et al., 2018). Building on this prior finding, in this study, we draw from the COR perspective to explain why narcissistic leaders are more likely to create conditions that would demotivate follower voice. According to COR theory, individuals are motivated to engage in resource acquisition or resource conservation activities (Hobfoll, 1989). Resource acquisition involves individuals engaging in activities to increase resources in their possession, whereas resource conservation refers to individuals avoiding potential resource loss by acting to prevent this anticipated threat or withdrawing from situations that may lead to the loss of resources (Halbesleben et al., 2009). In the context of leader narcissism, we contend that narcissistic leaders are motivated to maintain their limited psychological and social resources (e.g., a strong ego and perceptions of superiority) and, by doing so, can hinder follower voice. Specifically, narcissistic leaders resist becoming too invested in social relationships with their followers, but instead emphasize superiority, authority, and control (Maccoby, 2000). Previous research has described narcissistic individuals as having excessive need for dominance over others (Raskin et al., 1991; Raskin & Terry, 1988) and, in many cases, demanding the absolute obedience of others (Bushman & Baumeister, 1998). In work settings, narcissistic leaders tend to HUANG ET AL. 80 demonstrate their superiority over followers by emphasizing superior hierarchical status and position power (Maccoby, 2000). From the COR perspective, these efforts could indicate narcissistic leaders' attempt to maintain their limited resources (e.g., need for social dominance and demand for obedience) while avoiding potential resource loss as they interact with their followers. It is thus likely that when interacting with followers, narcissistic leaders will intentionally create conditions, under which it is harder for followers to express their suggestions or concerns that are often challenging in nature and thus can be hard for their narcissistic leaders to accept (Burris, 2012). That is, avoiding follower voice could be viewed as narcissistic leaders' self- regulatory effort of avoiding potential loss of psychological and social resources (e.g., sense of superiority and power over followers who challenge the status quo). This is also consistent with research findings, indicating that managers who tend to defend their strong ego (e.g., those with excessive selfconfidence) are less likely to solicit voice from their followers (Fast et al., 2014). With the above said, we expect that leader narcissism will have a direct negative impact on fol- lower voice. We thus hypothesize the following: Hypothesis 1. Leader narcissism is negatively related to follower voice. 4 | THE MEDI ATING ROLE OF LMX DI FFE RE NTI ATI ON As we draw from the COR theory to understand why leader narcis- sism could impact follower voice, it is important to note that interac- tions between a leader and a follower occur within, and therefore are influenced by, a larger context of differentiated relationships a leader develops with multiple followers in a group setting (Cashman, Dansereau, Graen, & Haga, 1976). Prior research has suggested that LMX differentiation, which refers to the differing social exchange relationships a leader develops with followers within a group (Liden et al., 2006), speaks to the extent to which the leader differentiates his offer of resources (e.g., self-esteem, emotional and work support, career opportunities, mentoring) in the social exchange processes with differ- ent followers. In the present study, we focus on LMX differentiation as a key mediating mechanism that transmits the negative impact of leader narcissism onto follower voice. 4.1 | Leader narcissism and LMX differentiation Although early narcissistic leadership literature has suggested that narcissistic leaders in general are unable to develop highquality inter- personal relationships with their followers (Maccoby, 2000), recent studies have shown that there are circumstances when narcissistic leaders can diverge from their narcissistic tendencies and instead show care about followers (Owens et al., 2015) or consult with them when making important work-related decisions (Carnevale et al., 2018). Although such positive interpersonal encounters are not widely reported and expected from narcissistic individuals, these recent studies have offered some preliminary evidence that narcissistic leaders may choose to develop high-quality social relationships with their followers if they find it beneficial. In line with such newly emerged empirical evidence, we propose that leader narcissism is positively associated with LMX differentiation. Narcissistic individuals have a strong need to maintain high self- esteem and therefore react positively when others satisfy their need for admiration (Grijalva & Harms, 2014; Raskin et al., 1991). They tend to favor interactions with those people who help reinforce their grandiose self-image and feed their inflated ego (Brown, 1997; Campbell, Rudich, & Sedikides, 2002). This can be portrayed as narcissists' self- enhancing tendency where positive reinforcement from interacting with others elevates their positive self-view (Campbell & Campbell, 2009). In the context of interpersonal relationships, due to their excessive demand for admiration, narcissistic individuals are more likely to engage in interactions that could help reinforce their positive self-image and avoid interactions that are less helpful in this regard (Rogoza, Wyszyńska, Maćkiewicz, & Cieciuch, 2016). Thus, we argue that narcissistic leaders may choose to maintain high-quality LMX relationships with followers who readily praise them or attribute posi- tive outcomes to them. In other words, narcissistic leaders will likely be motivated to develop higher-quality relationships with followers who are deferential to their superior status and can feed their inflated ego, and develop lower-quality relationships with followers from whom they are less likely to receive such pay-offs out of their invest- ment of time and other resources managing such relationships. This is also consistent with the COR perspective suggesting that individuals may strategically invest their limited resources in a way that helps maximize the accumulation of valuable resources without over- consuming their own limited resources such as time, energy, and attention (Hobfoll, 2001). In particular, the COR literature has indi- cated that when developing and maintaining social relationships, indi- viduals are motivated to acquire respect and emotional support as important psychological resources (Hobfoll, Freedy, Lane, & Geller, 1990). Therefore, it is reasonable to argue that narcissistic leaders develop highly differentiated relationships with followers as they stra- tegically select certain followers who can contribute to their resource gains (e.g., feeding their ego), whereas conserving their limited resources developing relationships with the rest. With the above said, we expect that in a group setting, a narcis- sistic leader is likely to develop relationships of differentiated quality with followers, which contributes to stronger within-group variation of LMX. We thus hypothesize the following: Hypothesis 2. Leader narcissism is positively related to LMX differentiation. 4.2 | LMX differentiation as a mediating mechanism We further propose that LMX differentiation in a group setting is likely to serve as a mediating mechanism through which leader HUANG ET AL. narcissism indirectly influences follower voice. Prior LMX differentia- tion literature suggests that variability in LMX relationships within a group affects followers' psychological and behavioral outcomes (Erdogan & Bauer, 2010; Harris, Li, & Kirkman, 2014; Henderson, Wayne, Shore, Bommer, & Tetrick, 2008; Herdman et al., 2017; Liao, Liu, & Loi, 2010; Liden et al., 2006; Sui, Wang, Kirkman, & Li, 2016). Specifically, high LMX differentiation can be considered a dysfunc- tional condition discouraging employees from engaging in their jobs because it violates the norms of equality (Gooty & Yammarino, 2016; Herdman et al., 2017; Li & Liao, 2014). For example, Gooty and Yammarino (2016) found that the positive relationship between LMX and follower performance was only evident when LMX differentiation was low. Using a Chinese sample of 228 employees working in 60 groups, Chen, Yu, & Son, (2014) offered further empirical evidence supporting this negative impact of high leader–member guanxi (i.e., a form of contextspecific social relationship in China) differentiation on outcomes such as job satisfaction, organizational commitment, and coworker helping. In line with these prior findings, we propose that LMX differen- tiation will have a negative effect on follower voice and therefore serve as an important mediating mechanism linking leader narcissism and follower voice. Specifically, we argue that high level of LMX dif- ferentiation (i.e., high variation in the quality of LMX relationships within a group) signals to followers that their narcissistic leader chooses to prioritize relationship development and maintenance within a group. Working in such a highly differentiated relational environment, followers will likely realize that their narcissistic leader strategically allocates his/her limited resources to developing rela- tionships with only selected employees. This will likely discourage both high-LMX followers, those loyal to the leader and concerned about maintaining the privileges associated with high LMX, and low-LMX followers, those less loyal to the leader and not feeling obligated to speak up with suggestions, from expressing their voice. Given that voice could be challenging and uncomfortable for the leader and may lead to leader–follower conflicts or disrupt leader- ship outcomes (Burris, Detert, & Romney, 2013), followers in high- LMX relationships with a narcissistic leader are more likely to be deferential to their leader's superior status and therefore are less likely to speak up to challenge the status quo. This is also consistent with prior findings that employees may choose not to express voice out of fear of losing opportunities to obtain workrelated and career support from their leaders (Detert, Burris, & Harrison, 2010; Morrison & Milliken, 2000). Followers in a lowquality LMX relation- ship with their narcissistic leaders are also unlikely to speak up, because they should not feel obligated to express their ideas or suggestions to help a leader with whom they do not have a close working relationship. They may also choose not to express voice due to fear of the harmful consequences for challenging the status quo (Burris, 2012). LowLMX followers are especially likely to experi- ence such negative consequences due to their inferior standing within the leader's group. In light of the arguments regarding the positive effect of narcis- sism on LMX differentiation coupled with our arguments regarding 81 the negative effect of LMX differentiation on voice, we hypothesize the following: Hypothesis 3. LMX differentiation mediates the negative relationship between leader narcissism and follower voice. 5 | THE M ODE RATI NG ROLE OF LLX Prior research suggests that leader–follower relationships reside within a broader network of upward, lateral, and downward relation- ships among followers, leaders, and leaders' superiors (Cashman et al., 1976; Graen & Uhl-Bien, 1995; Tangirala et al., 2007). This research also suggests that when leaders develop high-quality LLX with their supervisors, they are more likely to develop high LMX with their fol- lowers, thus creating less variation of LMX within their groups (Henderson et al., 2009; Zhou et al., 2012). Extending these prior find- ings regarding the effect of LLX on LMX differentiation, we contend that leaders' LLX with their superior offers a relational context that influences the extent to which narcissistic leaders develop differenti- ated LMX relationships with their followers in a group. Namely, we propose that leaders' LLX with their superior will moderate the positive effect of leader narcissism on LMX differentia- tion such that this effect will be stronger in the presence of low (vs. high) LLX. Prior research shows that leaders in low LLX relationships would develop low LMX relationships with their followers (Henderson et al., 2009; Zhou et al., 2012). However, we argue that the role of LLX in influencing LMX relationships within groups might be more complicated. Specifically, prior research suggests that narcissistic leaders are motivated to maintain control over and demand admira- tion and loyalty from their followers (Maccoby, 2000). Therefore, when narcissistic leaders have low-quality LLX with their own supe- rior, they should be motivated to seek social approval from others and more likely to accept followers' offer of admiration and loyalty. This is consistent with the resource view of social exchange suggesting that when resources valued by leaders (e.g., work-related support, emo- tional support, respect) cannot be obtained in certain social exchange relationships (e.g., LLX), leaders might seek to substitute for these unavailable resources by selectively turning to other relationships (e.g., LMX) that may help compensate for these absent resources (Wilson, Sin, & Conlon, 2010). This is also consistent with prior find- ings in the COR literature such that individuals are more likely to invest in resource acquisition activities when they perceive better opportunities for resource gains (Halbesleben, Neveu, PaustianUnderdahl, & Westman, 2014). Moreover, narcissistic leaders are highly egoistic yet sensitive to ego threats (Maccoby, 2000). Their failure to develop high-quality relationships with superior and gain the benefits associated with those relationships will likely be perceived by themselves as a threat to their positive self-image. It will also likely trigger their desire to compensate for this experienced ego threat by accepting an offer of respect, admiration, and loyalty from those selected followers who are able and willing to offer these resources to the leader. Thus, we HUANG ET AL. 82 expect that when narcissistic leaders experience low LLX with their superior, they will attempt to compensate for the shortage of resources (e.g., support, loyalty, admiration) from their LLX relation- ship by developing higher-quality LMX with those key followers in the group who can and want to offer these valuable resources. Further, low LLX condition should also encourage narcissistic leaders to invest less time into developing LMX relationships with those followers in the group who they believe cannot contribute significantly to their leadership success. On the other hand, when narcissistic leaders have high LLX with their superior, they enjoy the benefits and privileges from these high- quality relationships. From the COR perspective, we argue that such LLX condition is more likely to trigger narcissistic leaders' effort to conserve their limited resources by disengaging followers, because they have access to desirable resources, such as recognition and respect, from their own superior. Therefore, we argue that to maintain a positive selfimage, narcissistic leaders should be more likely to reserve their limited time and effort for interactions with their supe- rior who could offer them valuable resources while avoiding the risk of resource loss by disengaging their followers. Therefore, we expect that when narcissistic leaders experience high LLX, it will lead to less variation in the LMX relationships within a group compared with the circumstance when they experience low LLX. With the above said, we hypothesize the following: Hypothesis 4. Leader's LLX moderates the positive relationship between leader narcissism and LMX differentiation, such that this positive relationship will be stronger in the presence of low (vs. high) LLX. Integrating our arguments above, we further propose a moder- ated mediation model suggesting that narcissistic leaders are likely to develop highly differentiated LMX relationships in their groups and thereby create conditions discouraging followers from expressing their voice. We also expect that this indirect effect of leader narcis- sism on employee voice will be stronger when leaders have low LLX with their superior. We thus hypothesize the following: Hypothesis 5. Leader's LLX moderates the negative indirect effect of leader narcissism on follower voice via LMX differentiation, such that this negative indirect effect will be stronger in the presence of low (vs. high) LLX. 6 6.1 | METHOD were invited to complete a survey containing questions about their narcissism and their LLX relationship with their superior. 96 managers responded (66.2% response rate). At Time 2, which was 6 weeks after Time 1, 921 employees under the supervision of those 96 participat- ing managers were invited to complete a survey with questions about their LMX with the supervising manager. Five hundred eighty-eight employee survey responses were collected (63.8% response rate). At Time 3, which was 6 weeks after Time 2, those 588 employee respon- dents were asked to complete a survey about their voice behavior. Four hundred sixty-two employee survey responses were collected (78.6% response rate). As we found out after the data collection was concluded, one of the 96 participating managers left the company after completing his/her Time 1 survey and a new supervisor was assigned to the employees in that group. As a result, Time 2 and Time 3 survey responses provided by employees from that group were not related to their perceived relationship and behaviors with the supervisor who completed the Time 1 survey. Thus, we deleted this supervisor's and his/her employees' survey responses from our final dataset. This resulted in a final sample of 457 employees nested under 95 supervi- sors. Group size ranged from three to seven employees, with the aver- age of approximately five employees per group. In the final supervisor sample, the average age was 36.84 years and the average organiza- tional tenure was 7.19 years. The supervisor sample was predominantly male (80.3%). In the final employee sample, the average age was 30.27 years and the average organizational tenure was 5.45 years. The employee sample was also predominantly male (60.2%). Average dyadic tenure among supervisors and employees was 3.38 years. 6.2 6.2.1 | Measures | Leader narcissism Leader narcissism was measured using the 16-item Narcissistic Per- sonality Inventory (NPI-16; Ames, Rose, & Anderson, 2006). Following the NPI-16 scoring procedures, we coded narcissismconsistent responses as 1 and narcissism-inconsistent responses as 0. A sample pair item was: “I really like to be the center of attention” (a narcissism- consistent response) and “It makes me uncomfortable to be the center of attention” (a narcissisminconsistent response). Leader narcissism scores were then computed by averaging out the 16 NPI items (α = .72). | Sample and procedures We invited employees and their supervising managers working at a large Chinese consulting company to participate in this study. These employees work in consulting teams that serve the business needs of clients with a primary focus on the strategic side of business planning, operations, and mergers and acquisitions. Data for this study was col- lected at three points in time. At Time 1, 145 supervising managers 6.2.2 | Leaders' LLX The LLX of leaders with their superior was assessed using a 7-item LMX-7 scale (Graen & Uhl-Bien, 1995) that was adapted to capture the quality of a work relationship between supervisors and their supe- rior. A sample item was: “How would you characterize your working relationship with your leader” with a 5-point response scale ranging from 1 (extremely ineffective) to 5 (extremely effective; α = 92). HUANG ET AL. 6.2.3 | LMX differentiation To obtain a measure of LMX differentiation, we asked employees to rate their LMX with supervisors. Employee LMX was measured using a 7-item LMX-7 scale (Graen & Uhl-Bien, 1995). A sample item was: “How would you characterize your working relationship with your leader” with a 5-point response scale ranging from 1 (extremely ineffective) to 5 (extremely effective; α = .91). Variance in employee ratings of LMX was used as an index of LMX differentiation following prior studies on LMX differentiation (Erdogan & Bauer, 2010; Liden et al., 2006). 6.2.4 | Employee voice Employee voice was measured using a 6-item scale (α = .90) devel-oped by Van Dyne and LePine (1998). A sample item was: “I develop and make recommendations concerning issues that affect my work.” 6.2.5 | Control variables We control for employee LMX, group mean LMX, employee demo- graphics (age, gender, organizational tenure, and dyadic tenure with leaders) and manager demographics (age, gender, and organizational tenure) in the model used to test study hypotheses. Also, given that we are interested in narcissism as a voice-suppressing factor, it is important to control for leadership factor that may encourage voice in order to determine whether it is narcissism per se that reduces voice rather than a general positive or negative perception of the leader.1 Given that empowering leader behaviors involve leaders enhancing employee perception that their work is meaningful, encouraging employees to participate in decisionmaking, expressing confidence that employees can achieve high level of performance, and providing employees with autonomy from bureaucratic constraints, it can be expected that empowering leadership can be a powerful leadership factor encouraging employees to express voice (Ahearne, Mathieu, & Rapp, 2005). Thus, we also include empowering leadership as a con- trol variable in our model.2 Empowering leadership was measured using a 12-item scale developed by Ahearne et al. (2005). A sample item was: “My manager allows me to do my job my way.” 6.3 | Analytical strategy Due to the nested nature of the data (multiple employees nested under supervisors), we used multilevel modeling to test the study hypotheses. The outcome variable (voice) demonstrated a moderate amount of between-group variability, 0.13, p < .001; ICC(1) = 0.16, which indicated the presence of a group effect and justified the use of multilevel modeling. Our hypothesized model is, in essence, a multilevel firststage moderated mediation model (cf Edwards & Lamberts, 2007), with a 83 Level-2 predictor, mediator, and moderator, and a Level-1 outcome. This model was tested using a multilevel path analysis in Mplus 7 (Muthén & Muthén, 2012). The unconditional and conditional (i.e., varying across levels of the moderator) effects of narcissism on voice were tested using the Monte Carlo simulation with 20,000 replications (Bauer, Preacher, & Gil, 2006; MacKinnon, Lockwood, & Williams, 2004; Preacher & Selig, 2012). The Monte Carlo approach to constructing confidence intervals (CIs) does not assume the normality of the indirect effect (MacKinnon et al., 2004) and is recommended as a suitable alternative to bootstrapping in complex multilevel models (e.g., Preacher & Selig, 2012). The Monte Carlo (MC) CIs were generated in Rweb. 7 | RESULTS Means, standard deviations, and correlations among the study variables are provided in Table 1. As expected, leader narcissism was positively related to LMX differentiation (r = 0.37, p < .01) while negatively related to employee voice (r = −0.21, p < .01). LMX differentia- tion was also negatively related to employee voice (r = −0.26, p < .01). Hypothesis 1 predicted a negative relationship between leader narcissism and employee voice. As shown in Table 2, leader narcissism was negatively associated with employee voice (γ = −0.84, p < .001), thus supporting Hypothesis 1. Narcissism explained 34.02% of additional group-level variance in voice, above and beyond variance accounted for by other predictors in the model. Hypothesis 2 predicted a positive relationship between leader narcissism and LMX differentiation. As shown in Table 2, the relation- ship between narcissism and LMX differentiation was positive and significant (γ = 0.75, p < .01), thus supporting Hypothesis 2. Narcissism explained 11.74% of additional group-level variance in LMX differenti- ation, above and beyond variance accounted for by other predictors in the model. Hypothesis 3 proposed that the effect of leader narcissism on follower voice would be mediated by LMX differentiation. As shown in Table 2, LMX differentiation was negatively associated with voice (γ = -0.30, p < .01). The unconditional indirect effect of narcissism on LMX differentiation was −.22 and the corresponding 95% MC CI was [−0.49, −0.04]. This mediation was partial, as the effect of narcissism on voice remained negative and significant in the presence of LMX differentiation (γ = −0.62, p < .05). When the mediator was included in the model, narcissism explained 23.33% of additional group-level variance in voice above and beyond other predictors. Thus, Hypothe- sis 3 was supported. Hypothesis 4 predicted that the positive effect of narcissism on LMX differentiation would be moderated by leader LLX with their superior. In line with this prediction, the interactive effect of narcissism and LLX on LMX differentiation was significant (γ = −0.56, p < .001). With the product term included in the model, narcissism explained 6.52% of additional group-level variance in LMX differentia- tion above and beyond other predictors. To probe this interaction, we conducted a simple slope test and constructed a simple slope plot Descriptive statistics and correlations among the study variables Variable 1 2 3 4 5 (.91) 84 T AB L E 1 6 7 8 9 10 11 (.72) 12 13 14 1. Age Emp. 2. Female Emp. 3. Org. tenure Emp. −.04 .75*** .03 .01 4. Dyadic tenure Emp. 5. LMX Emp. −.04 −.04 −.01 −.01 6. Emp. leadership .02 −.11* −.02 −.12* 7. Age Mgr .04 −.02 .05 −.03 .03 .01 8. Female Mgr −.01 .10* −.03 −.01 −.01 −.11* −.20*** 9. Org. tenure Mgr .02 −.04 .05 .08 .04 −.07 .60*** 10. LMX grp. mean .01 −.02 −.04 −.09 11. Narcissism .10* −.09 .13** .15** .01 .08 12. LLX 13. LMX diff. .00 .05 .00 .04 (.94) −.01 .04 .06 −.03 .08 −.07 .03 −.02 .03 −.04 −.15** .01 .13** .02 .08 −.04 −.03 .27*** −.13 (.92) .14** −.18*** −.02 −.01 .07 −.38*** .37*** −.22*** −.21*** .04 −.26*** (.90) *** 14. Voice .04 −.06 .01 −.02 .30 Mean 30.27 0.40 65.42 40.61 3.60 SD 3.26 0.49 31.87 17.49 0.86 −.04 *** *** .01 −.02 .02 .20 3.71 36.84 0.20 86.30 3.60 0.42 3.52 0.72 3.61 0.76 3.32 0.40 26.72 0.41 0.22 0.97 0.50 0.89 .21 Note. At Level 1: n = 457 employees. At Level 2: n = 95 leaders. Emp., employee; Mgr, manager; Org. tenure, organizational tenure; Emp. leadership, empowering leadership; LMX diff., LMX differentiation. Organizational tenure and dyadic tenure are measured in months. Values on the diagonal are internal consistency estimates. *p < .05. **p < .01. ***p < .001. HUANG ET AL. HUANG ET AL. TA BL E 2 85 Effects of narcissism and LLX on LMX differentiation and voice Outcome: LMX differentiation Effects The main effects model γ Intercept 0.74*** Outcome: voice The interactive effect model SE γ 0.05 0.73*** The interactive effect model The main effects models SE γ SE γ SE γ SE 0.05 3.64*** 0.05 3.85*** 0.10 3.84*** 0.10 0.02 Level 1 effects 0.02 0.02 0.02 0.02 0.02 −0.08 0.08 −0.07 0.08 −0.07 0.08 Org. tenure Emp. 0.00 0.00 0.00 0.00 0.00 0.00 Dyadic tenure Emp. −0.00 0.00 −0.00 0.00 −0.00 0.00 LMX Emp. 0.27*** 0.06 0.27*** 0.06 0.27*** 0.06 −0.00 0.02 −0.01 0.02 −0.01 0.02 0.12 Age Emp. Female Emp. Level 2 effects Age Mgr −0.01 0.02 −0.01 Female Mgr −0.07 0.11 −0.10 0.10 0.01 0.12 −0.01 0.12 −0.00 Org. tenure Mgr 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 LMX group mean −0.38** 0.14 −0.37** 0.13 0.34** 0.12 0.23* 0.11 0.24* 0.11 Narcissism 0.75** 0.24 0.54* 0.22 −0.84*** 0.23 −0.62** 0.23 −0.59* 0.24 LLX −0.06 0.05 −0.06 0.05 −0.02 0.05 −0.03 0.05 −0.03 0.05 −0.30** 0.11 −0.29** 0.11 0.11 0.11 0.15 0.11 0.07 0.22 0.02 LMX differentiation Empowering leadership Narcissism*LLX 0.15 −0.56*** 0.16 0.11 Note. At Level 1: n = 457 employees. At Level 2: n = 95 leaders. Emp., employee; Mgr, manager; Org. tenure, organizational tenure. Organizational tenure and dyadic tenure are measured in months. *p < .05. **p < .01. ***p < .001. (Aiken West, & Reno, 1991). As seen in Figure 2, narcissism had a pos- itive and significant effect on LMX differentiation, when LLX was low (simple slope for low LLX: b = 1.08, p < .001) but not when LLX was high (simple slope for high LLX: b = −0.004, p > .05). Although informative, this traditional way of probing interactions is somewhat lim- ited (i.e., the simple slope values are obtained for the selected values of a moderator, e.g., one standard deviation above and below mean). FI G U R E 2: The interactive effect of narcissism and leader–leader exchange (LLX) on leader– member exchange (LMX) Thus, we further explored the moderating effect of LLX by examining simple slopes across the entire range of the moderator values. To do so, we used the online utilities developed by Preacher and colleagues and obtained a plot of simple slopes across all in-range values of LLX (Preacher, Curran, & Bauer, 2006; see Figure 3). As seen in this figure, the simple slope of narcissism on LMX differentiation is positive (i.e., above 0 on the Y-axis) and significant only at the low levels of 86 differentiation HUANG ET AL. HUANG ET AL. 87 FI G U R E 3: Simple slope of narcissism on LMX differentiation across levels of LLX. Note: The vertical line crosses the X-axis at 0.18 units above the mean of grand mean centered LLX. The plot suggests that the effect of narcissism on LMX differentiation is positive (i.e., the simple slope is above 0 on the Yaxis) when LLX is relatively low (i.e., is below 0.18 units above the mean) LLX. The simple slope becomes nonsignificant at LLX of 0.18 units above the mean (see the dashed vertical line on the plot; the simple slope values to the right of this line are not significantly different from zero). This suggests that the moderating effect of LLX is noticeable only at the lower levels of LLX (below 0.18 units above the mean of LLX). Finally, Hypothesis 5 predicted that the indirect effect of narcis- sism on voice would vary across levels of leader LLX. This hypothesis also received support. The index of moderated mediation (Hayes, 2015) was 0.16 with the corresponding 95% MC CI of [0.14, 0.18], suggesting that the indirect effect of leader narcissism on voice via LMX differentiation indeed varied across levels of a moderator (leader LLX) such that it became significant at the lower levels of LLX. Specifi- cally, when LLX was low (i.e., one standard deviation below the mean), this indirect effect was negative and significant (−0.31; 95%MC CI [−0.35, −0.27]). However, when LLX was high (i.e., one standard devi- ation above the mean), this indirect effect was not significantly different from zero (−0.001; 95% MC CI [−0.03, 0.03]). With the product term included in the model, narcissism explained 22.41% of additional group-level variance in voice above and beyond other predictors. Overall, the results of the hypothesis testing suggest that leaders with higher levels of narcissism and lower-quality LLX with their superior tend to create more LMX differentiation among their followers and, as a result, suppress employees' willingness to express their voice. 8 | DI SCUSSION Drawing from the COR theory and the narcissism literature, we proposed and tested a moderated mediation model in which leader narcissism has a negative direct effect on follower voice as well as a negative indirect effect on voice via group-level LMX differentiation. We found that leader LLX moderated the positive relationship between leader narcissism and LMX differentiation such that this pos- itive relationship was significant only in the presence of low leader LLX but turned nonsignificant in the presence of high leader LLX. Moreover, leader LLX moderated the negative indirect effect of narcissism on voice such that this indirect effect was stronger in the presence of low leader LLX but turned nonsignificant in the presence of high leader LLX. Collectively, these results support our hypotheses that narcissistic leaders have a general tendency to create greater LMX differentiation among their followers and, as a result, tend to suppress followers' willingness to express voice. The results also sug- gest that this negative effect of leader narcissism on employee voice is largely neutralized in the condition of higher-quality LLX of leaders with their own superior. Put another way, narcissistic leaders are trig- gered to feel the need to acquire or recapture social and psychological resources by seeking support from favored subordinates only when they themselves feel threatened by a poor-quality relationship with their own supervisor. 8.1 | Theoretical implications Our study makes at least four theoretical contributions. First, although existing research about leader narcissism primarily focuses on its impact on followers' perceptions of leader charisma, leadership effec- tiveness and emergence, and follower work engagement (Galvin et al., 2010; Grijalva, Harms, Newman, Gaddis, & Fraley, 2015; Judge, LePine, & Rich, 2006; Resick et al., 2009), our study extends this line of research by identifying a key mediating mechanism (i.e., LMX differentiation) underlying the relationship between leader narcissism 88 and follower voice. As suggested by prior research, it is imperative for organizational studies to theoretically identify and empirically test mediating mechanisms through which leader narcissism could impact employee outcomes (Carnevale et al., 2018). Drawing from the narcis- sism literature, we argue that narcissistic leaders may selectively develop relationships with followers who can feed their inflated ego in order to acquire self-esteem–related resources (i.e., engage in resource acquisition) and avoid spending time and resources managing relationships with the rest of the group (i.e., engage in resource con- servation). This is consistent with the COR literature such that individ- uals may strategically invest or reserve their resources in order to either maximize potential resource gains or avoid potential threat of resource loss (Hobfoll, 2001). Our findings about a positive relationship between leader narcis- sism and LMX differentiation adds to the narcissism literature by suggesting that narcissistic leaders might not necessarily have a diffi- cult time building relationships with all of their followers (Campbell et al., 2011; Judge et al., 2009). Instead, it is possible that they might instead strategically develop high-quality LMX relationships with some followers in the group who could provide the leader with valu- able resources such as expertise, experience, or high performance. This is consistent with recent advancements in narcissism research about narcissistic leaders' potential deviations from their typical behavioral tendencies as they could appear to be appreciative or car- ing in front of their followers (Carnevale et al., 2018; Owens et al., 2015). Further, narcissists have a strong need for admiration (Raskin et al., 1991). They prefer engaging in interactions where they may receive more favorable evaluations from the other party (Brown, 1997; Campbell et al., 2002; Harms & Spain, 2015; Spain et al., 2014). Thus, narcissistic leaders may choose to strategically develop high- quality LMX relationships with those followers who offer resources valuable to the leader (e.g., praise the leader for his/her successes and accomplishments). This is also consistent with prior narcissism litera- ture suggesting that narcissists might demonstrate selfenhancing ten- dencies to maintain positive self-image (Campbell & Campbell, 2009; Spain et al., 2014). Second, although prior voice research examining leader anteced- ents of employee voice has primarily focused on leader traits (e.g., openness) that are associated with increased voice behaviors (Detert & Burris, 2007), our study is among the first to explore leader charac- teristics that could prevent employees from speaking up. Our study thus extends the nomological network of voice behavior by identify- ing leader narcissism as a leader-related predictor of individual voice and thereby answering the call for a broader and more detailed exami- nation of leader influences on follower voice (Fast et al., 2014; Morri- son, 2011). Our findings showed that when interacting with their followers, narcissistic leaders could signal through their development and maintenance of differentiated LMX relationships their level of receptiveness to follower voice. We contend that followers should be able to recognize such informational cues from their narcissistic leader and use these cues to determine what they should do in such a highly differentiated relational environment. Specifically, followers in high- LMX relationships with their narcissistic leaders might choose not to HUANG ET AL. speak up out of deference to their leaders so as to feed their inflated ego and secure the benefits associated with their highLMX status. Followers in low-LMX relationships might also choose not to speak up because they may not feel obligated to contribute via voice or they may be afraid of potential harmful consequences of speaking up. Therefore, by creating highly differentiated LMX relationships in a work group, narcissistic leaders are likely to hinder followers voice. Our findings thus supplement Farh and Chen (2014) and further shed light on the “dark side” of leader influences on employee voice. Third, by obtaining support for the hypothesized negative effect of LMX differentiation on follower voice, our study adds to a better understanding of the consequences of LMX differentiation. Prior LMX differentiation research has often focused on the moderating effect of LMX differentiation on relationships linking LMX with outcomes such as job performance, organizational commitment, turnover intentions, organizational citizenship behaviors, psychological fulfillment, employee creativity, and coworker helping behavior (Harris et al., 2014; Henderson et al., 2008; Liao et al., 2010; Liden et al., 2006). Attempts to explore the main effects of LMX differentiation on the outcomes, however, have not been always successful. For example, Liden et al. (2006) failed to find a significant main effect of LMX dif- ferentiation on individual performance. Our findings, therefore, add to the LMX differentiation literature by showing that followers are less likely to speak up in groups with higher LMX differentiation. This is an interesting finding that adds to Liden et al. (2006) because it indicates that voice, as an outcome more relevant to the social exchange context (i.e., voice as a way to contribute to the leader–follower relationship), might be a more salient outcome of LMX differentiation than job performance, which concerns more with followers' own success. Finally, our study is among the first to theoretically explain why and when narcissistic individuals may deviate from their typically antisocial behavioral tendencies. Recent studies have shown that nar- cissistic leaders might appear to be humble in front of followers (Owens et al., 2015) or involve followers in decision-making (Carnevale et al., 2018), which therefore helps lessen the detrimental effect of leader narcissism on employee outcomes. Yet, it has remained unclear why and how such deviation from the typical behav- ioral tendencies may happen. Relying on the COR theory (Hobfoll, 1989), our study identifies LLX as an important boundary condition that alters the effect of narcissism on LMX differentiation. Namely, the results of our study demonstrate that narcissistic leaders in lowerquality LLX relationships might more actively seek approval from followers and accept their attempts to build higher-quality LMX relationships as they desire to compensate for lower-quality LLX by gaining recognition, respect, and admiration from followers. This is consistent with findings from the COR literature suggesting that when facing opportunities to acquire valuable resources, individuals become motivated to choose resource acquisition strategy to more actively invest their effort that could result in resource gains (Halbesleben et al., 2014). Our finding that lower-quality LLX magnified, whereas higher- quality LLX diminished, the positive relationship between leader HUANG ET AL. narcissism and LMX differentiation also contribute to the LMX litera- ture regarding the impact of leaders' vertical relationships with supe- rior on their LMX with followers. According to Henderson et al. (2009), leaders in low-quality LLX do not provide followers with resources that could benefit them, which results in lower overall qual- ity of group-level LMX and lower LMX differentiation within a group. Zhou et al. (2012) also found that leaders tend to model LMX with their followers after their LLX with superior. Our findings, however, show that when narcissistic leaders have low-quality LLX with their supervisors, they tend to have LMX relationships with their followers that differ considerably in terms of quality. That is, when threatened by a tenuous relationship with a supervisor, these leaders may defend their egos by embracing followers they deem as helpful while rejecting those who threaten or disappoint them. Consequently, narcissistic leaders in low LLX situations are likely to have higher vari- ation of LMX within a group. However, when narcissistic leaders have high-quality LLX with their superior, their egos or status are strength- ened, therefore creating a situation where they care less about their LMX relationships with followers. This leads to low variation of LMX in a group setting, as indicated by our finding that group mean LMX was significantly and negatively associated with LMX differentiation. 8.2 | Practical implications The results of our study suggest some practical implications for orga- nizations interested in promoting voice, or any discretionary work behavior, among their employees. First, although narcissists are more prone to end up in leadership positions, they are often not well-suited to them (Grijalva et al., 2015; Grijalva & Harms, 2014). This is particu- larly true when those leadership positions require close contact and relationships with subordinates. In the present study, we found that narcissistic leaders tended to differentiate between team members in terms of their relationship quality and resource allocation in a manner that demotivated employees to speak up. When leaders' egos were threatened by low-quality relationship with their superior, this effect was exaggerated. Consequently, one recommendation based on this research would be to actively screen for narcissism when promoting individuals into the positions of leadership. Because the negative effects of narcissism are typically experienced by subordinates and peers rather than superiors, such screening should involve suitability assessments from more than immediate supervisors. Specifically, if well-validated assessment instruments are available, assessments of leadership suitability from subordinates and peers should be utilized. Second, based on the results of the current study and related research on narcissistic aggression (e.g., Baumeister, Bushman, & Campbell, 2000), the negative side of narcissism seemed to be trig- gered when the egos of narcissistic leaders were threatened. In the present study, when leaders felt threatened by poor relations with their own boss, they reacted by creating working contexts where cer- tain subordinates were favored over others with the result that the likelihood of employee voice was diminished. However, these nega- tive effects were not present when the narcissistic leaders felt their 89 relationship with the boss was great. This suggests that even those in upper management roles need to be mindful of the quality of relation- ships they have with their direct reports since low quality relation- ships between those in the managerial hierarchy can potentially result in negative downstream effects beyond their immediate supervision. In some ways, this can be thought of as being reflective of the “cas- cading leadership hypothesis” that suggests leadership behaviors at the upper levels of organizations tend to echo downwards through the organizational hierarchy (Bass, 1990; Bass, Waldman, Avolio, & Webb, 1987; Liu, Liao, & Loi, 2012). In the present case, kind or supportive words from upper managers may have been enough to convince narcissistic leaders that there was no need to believe that their psychological and social resources were threatened and avoid triggering some of the more toxic and divisive behaviors often seen in narcissists. 8.3 | Limitations and directions for future research Our study is subject to several limitations. First, although we collected survey data at three time points, causality cannot be unambiguously established. Although we feel that the relatively stable nature of trait narcissism renders reverse causality unlikely, this research would ben- efit from a follow-up experimental study in which the components of the proposed model are tested with regards to their causal relation- ships. For example, the follower's intent to express voice in response to behaviors of narcissistic leaders can be assessed in an experimental vignette study, where study participants are presented with vignettes depicting leaders with varied levels of narcissism. Also, LLX quality can be manipulated in a lab experiment to see how it may affect indi- viduals with different levels of narcissism and their relationships with others in groups. Second, we focused on just one type of follower contribution to leadership outcomes—follower voice. It is quite possible that other types of follower contributions may be also affected by leader narcis- sism and the resultant variation in LMX. For example, employees may choose to skip their direct supervisor if he/she tends to act in a narcis- sistic manner and speak up to higher-level leaders in the organization, thus potentially creating relationship conflicts across hierarchies (cf Detert & Trevino, 2010). We recommend that future studies investi- gate these other outcomes of leader narcissism and the resultant LMX differentiation. Third, we collected our data from China and thus could not rule out the possibility that our findings might not be generalized to other cultural settings. The LMX measure developed in the western cultural context could suffer from potential cultural bias in one's interpreta- tions of the social exchange phenomenon and therefore be vulnerable to cross-cultural applications. Yet, there is empirical evidence supporting the use of LMX construct (as well as LMX differentiation and LLX constructs) across cultures, including the Chinese context (e.g., Carnevale, Huang, & Paterson, 2019; Huang, Xu, Huang, & Liu, 2018; Mackey, Huang, & He, in press; Sui et al., 2016; Zhou et al., 2012). Another concern is that there could be other contextual (e.g., 90 HUANG ET AL. collectivistic culture) or individual (e.g., traditionality) factors that might influence the way employees and their narcissistic leaders choose to interact in a leader–follower relationship. Future research could investigate the cross-cultural impact on how employees may choose to employ a particular strategy (i.e., resource conservation or resource acquisition) when interacting with a narcissistic leader. Finally, NPI-16 may not fully capture the breadth of the construct of narcissism (Grijalva et al., 2015). Thus, other self-report measures of leader narcissism (e.g., Back et al., 2013) or objective measures of narcissism (e.g., observations of leader behaviors in a lab setting or at the workplace, measuring the size of a leader's picture on a company's website, counting firstperson singular pronouns in leader's speeches, cf Chatterjee & Hambrick, 2007) could potentially be used in future research to better capture the range of behavioral manifestations of narcissism. 9 | CONLUSION Our research has addressed two timely issues in the voice research: why certain leaders are averse to employee voice, and how leader traits and relationship building in a group context could influence employee voice. The findings of the present study demonstrated that leader narcissism could be detrimental to employee voice as narcissis- tic leaders might create socially differentiated relationships with fol- lowers in a group context. Our results further demonstrated that narcissistic leaders' low-quality LLX with their own supervisors would motivate them to focus their limited resources developing highquality relationships with followers who may be more helpful in contributing to leadership outcomes, thus creating a highly differentiated relational environment within a group. ORCID Lei Huang https://orcid.org/0000-0001-9723-7757 REFERENCES Ahearne, M., Mathieu, J., & Rapp, A. (2005). To empower or not to empower your sales force? An empirical examination of the influence of leadership empowerment behavior on customer satisfaction and performance. Journal of Applied Psychology, 90(5), 945–955. https:// doi.org/10.1037/0021-9010.90.5.945 Aiken, L. S., West, S. G., & Reno, R. R. (1991). Multiple regression: Testing and interpreting interactions. Thousand Oaks, CA: Sage. Ames, D. R., Rose, P., & Anderson, C. P. (2006). The NPI-16 as a short measure of narcissism. Journal of Research in Personality, 40(4), 440–450. https://doi.org/10.1016/j.jrp.2005.03.002 Back, M. D., Küfner, A. C., Dufner, M., Gerlach, T. M., Rauthmann, J. F., & Denissen, J. J. (2013). Narcissistic admiration and rivalry: Disentangling the bright and dark sides of narcissism. Journal of Personality and Social Psychology, 105(6), 1013–1037. https://doi.org/10.1037/ a0034431 Bass, B. M. (1990). From transactional to transformational leadership: Learning to share the vision. Organizational Dynamics, 18, 19–31. https://doi.org/10.1016/0090-2616(90)90061-s Bass, B. M., Waldman, D., Avolio, B. J., & Webb, M. (1987). Transforma- tional leadership and the falling dominoes effect. Group and Organization Studies, 12, 73–87. https://doi.org/10.1177/ 105960118701200106 Bauer, D. J., Preacher, K. J., & Gil, K. M. (2006). Conceptualizing and test- ing random indirect effects and moderated mediation in multilevel models: New procedures and recommendations. Psychological Methods, 11(2), 142–163. https://doi.org/10.1037/1082-989x.11. 2.142 Baumeister, R., Bushman, B., & Campbell, W. K. (2000). Self-esteem, nar- cissism, and aggression: Does violence result from low selfesteem or from threatened egotism? Current Directions in Psychological Science, 9, 26–29. https://doi.org/10.1111/14678721.00053 Brown, A. D. (1997). Narcissism, identity, and legitimacy. Academy of Management Review, 22(3), 643–686. https://doi.org/10.5465/amr.1997. 9708210722 Burris, E. R. (2012). The risks and rewards of speaking up: Managerial responses to employee voice. Academy of Management Journal, 55(4), 851–875. https://doi.org/10.5465/amj.2010.0562 Burris, E. R., Detert, J. R., & Chiaburu, D. S. (2008). Quitting before leaving: The mediating effects of psychological attachment and detachment on voice. Journal of Applied Psychology, 93(4), 912–922. https://doi.org/ 10.1037/0021-9010.93.4.912 Burris, E. R., Detert, J. R., & Romney, A. C. (2013). Speaking up vs. being heard: The disagreement around and outcomes of employee voice. Organization Science, 24(1), 22–38. https://doi.org/10.1287/orsc. 1110.0732 Bushman, B. J., & Baumeister, R. F. (1998). Threatened egotism, narcissism, self-esteem, and direct and displaced aggression: Does self-love or self-hate lead to violence? Journal of Personality and Social Psychology, 75(1), 219–229. https://doi.org/10.1037//0022-3514.75.1.219 Campbell, W. K., & Campbell, S. M. (2009). On the self-regulatory dynam- ics created by the peculiar benefits and costs of narcissism: A contex- tual reinforcement model and examination of leadership. Self and Identity, 8(2-3), 214–232. https://doi.org/10.1080/15298860802 505129 Campbell, W. K., Hoffman, B. J., Campbell, S. M., & Marchisio, G. (2011). Narcissism in organizational contexts. Human Resource Management Review, 21(4), 268–284. https://doi.org/10.1016/j.hrmr.2010.10.007 Campbell, W. K., Rudich, E. A., & Sedikides, C. (2002). Narcissism, selfesteem, and the positivity of self-views: Two portraits of self-love. Personality and Social Psychology Bulletin, 28(3), 358–368. https://doi. org/10.1177/0146167202286007 Carnevale, J. B., Huang, L., & Harms, P. D. (2018). Leader consultation mitigates the harmful effects of leader narcissism: A belongingness perspective. Organizational Behavior and Human Decision Processes, 146, 76–84. https://doi.org/10.1016/j.obhdp.2018.04.003 Carnevale, J. B., Huang, L., & Paterson, T. (2019). LMX-differentiation strengthens the prosocial consequences of leader humility: An identifi- cation and social exchange perspective. Journal of Business Research, 96, 287–296. https://doi.org/10.1016/j.jbusres.2018.11.048 Cashman, J., Dansereau, F., Graen, G., & Haga, W. J. (1976). Organizational understructure and leadership: A longitudinal investigation of the managerial role-making process. Organizational Behavior and Human Decision Processes, 15(2), 278–296. https://doi.org/10.1016/00305073(76)90042-8 Chatterjee, A., & Hambrick, D. C. (2007). It's all about me: Narcissistic chief executive officers and their effects on company strategy and perfor- mance. Administrative Science Quarterly, 52(3), 351–386. https://doi. org/10.2189/asqu.52.3.351 Chen, Y., Yu, E., & Son, J. (2014). Beyond leader–member exchange (LMX) differentiation: An indigenous approach to leader– member relation- ship differentiation. The Leadership Quarterly, 25(3), 611–627. https:// doi.org/10.1016/j.leaqua.2013.12.004 Detert, J. R., & Burris, E. R. (2007). Leadership behavior and employee voice: Is the door really open? Academy of Management Journal, 50(4), 869–884. HUANG ET AL. Detert, J. R., Burris, E. R., & Harrison, D. A. (2010). Debunking four myths about employee silence. Harvard Business Review, 88(6), 26–31. Detert, J. R., & Trevino, L. K. (2010). Speaking up to higher-ups: How supervisors and skip-level leaders influence employee voice. Organiza- tion Science, 21(1), 249–270. https://doi.org/10.1287/orsc.1080.0405 Edwards, J. R., & Lambert, L. S. (2007). Methods for integrating moderation and mediation: a general analytical framework using moderated path analysis. Psychological Methods, 12(1), 1–22. https://doi.org/10.1037/ 1082-989X.12.1.1 Emmons, R. A. (1987). Narcissism: Theory and measurement. Journal of Personality and Social Psychology, 52(1), 11–17. https://doi.org/10. 1037//0022-3514.52.1.11 Erdogan, B., & Bauer, T. N. (2010). Differentiated leader-member exchanges: The buffering role of justice climate. Journal of Applied Psychology, 95(6), 1104–1120. https://doi.org/10.1037/a0020578 Farh, C. I., & Chen, Z. (2014). Beyond the individual victim: Multilevel con- sequences of abusive supervision in teams. Journal of Applied Psychology, 99(6), 1074–1095. https://doi.org/10.1037/a0037636 Fast, N. J., Burris, E. R., & Bartel, C. A. (2014). Managing to stay in the dark: Managerial self-efficacy, ego defensiveness, and the aversion to employee voice. Academy of Management Journal, 57(4), 1013–1034. https://doi.org/10.5465/amj.2012.0393 Galvin, B. M., Waldman, D. A., & Balthazard, P. (2010). Visionary communi- cation qualities as mediators of the relationship between narcissism and attributions of leader charisma. Personnel Psychology, 63(3), 509–537. https://doi.org/10.1111/j.1744-6570.2010.01179.x Gooty, J., & Yammarino, F. J. (2016). The leader-member exchange relationship: A multisource, cross-level investigation. Journal of Management, 42(4), 915–935. https://doi.org/10.1177/0149206 313503009 Graen, G. B., & Uhl-Bien, M. (1995). Relationship-based approach to lead- ership: Development of leader-member exchange (LMX) theory of leadership over 25 years—Applying a multi-level multidomain per- spective. The Leadership Quarterly, 6(2), 219–247. https://doi.org/10. 1016/1048-9843(95)90036-5 Grijalva, E., & Harms, P. D. (2014). Narcissism: An integrative synthesis and dominance complementarity model. The Academy of Management Perspectives, 28(2), 108–127. https://doi.org/10.5465/amp.2012. 0048 Grijalva, E., Harms, P. D., Newman, D. A., Gaddis, B. H., & Fraley, R. C. (2015). Narcissism and leadership: A meta-analytic review of linear and nonlinear relationships. Personnel Psychology, 68(1), 1–47. https:// doi.org/10.1111/peps.12072 Halbesleben, J. R., Harvey, J., & Bolino, M. C. (2009). Too engaged? A con- servation of resources view of the relationship between work engage- ment and work interference with family. Journal of Applied Psychology, 94(6), 1452–1465. https://doi.org/10.1037/a0017595 Halbesleben, J. R., Neveu, J. P., Paustian-Underdahl, S. C., & Westman, M. (2014). Getting to the “COR” understanding the role of resources in conservation of resources theory. Journal of Management, 40(5), 1334–1364. https://doi.org/10.1177/0149206314527130 Harms, P. D., & Spain, S. (2015). Beyond the bright side: Dark personality at work. Applied Psychology. An International Review, 64, 15–24. https://doi.org/10.1111/apps.12042 Harris, T. B., Li, N., & Kirkman, B. L. (2014). Leader-member exchange (LMX) in context: How LMX differentiation and LMX relational separa- tion attenuate LMX's influence on OCB and turnover intention. The Leadership Quarterly, 25(2), 314–328. https://doi.org/10.1016/j. leaqua.2013.09.001 Hayes, A. F. (2015). An index and test of linear moderated mediation. Mul- tivariate Behavioral Research, 50(1), 1–22. https://doi.org/10.1080/ 00273171.2014.962683 Henderson, D. J., Liden, R. C., Glibkowski, B. C., & Chaudhry, A. (2009). LMX differentiation: A multilevel review and examination of its 91 antecedents and outcomes. The Leadership Quarterly, 20(4), 517–534. https://doi.org/10.1016/j.leaqua.2009.04.003 Henderson, D. J., Wayne, S. J., Shore, L. M., Bommer, W. H., & Tetrick, L. E. (2008). Leader-member exchange, differentiation, and psychological contract fulfillment: A multilevel examination. Journal of Applied Psy- chology, 93(6), 1208–1219. https://doi.org/10.1037/a0012678 Herdman, A. O., Yang, J., & Arthur, J. B. (2017). How does leadermember exchange disparity affect teamwork behavior and effectiveness in work groups? The moderating role of leaderleader exchange. Journal of Management, 43(5), 1498–1523. https://doi.org/10.1177/ 0149206314556315 Hobfoll, S. E. (1989). Conservation of resources: A new attempt at concep- tualizing stress. American Psychologist, 44(3), 513–524. https://doi. org/10.1037//0003-066x.44.3.513 Hobfoll, S. E. (2001). The influence of culture, community, and the nested self in the stress process: Advancing conservation of resources theory. Applied Psychology. An International Review, 50, 337–370. https://doi. org/10.1111/1464-0597.00062 Hobfoll, S. E., Freedy, J., Lane, C., & Geller, P. (1990). Conservation of social resources: Social support resource theory. Journal of Social and Personal Relationships, 7(4), 465–478. https://doi.org/10.1177/ 0265407590074004 Huang, X., Xu, E., Huang, L., & Liu, W. (2018). Nonlinear consequences of promotive and prohibitive voice for managers’ responses: The roles of voice frequency and LMX. Journal of Applied Psychology, 103(10), 1101–1120. https://doi.org/10.1037/apl0000326 Judge, T. A., LePine, J. A., & Rich, B. L. (2006). Loving yourself abundantly: Relationship of the narcissistic personality to selfand other percep- tions of workplace deviance, leadership, and task and contextual per- formance. Journal of Applied Psychology, 91(4), 762–775. https://doi. org/10.1037/00219010.91.4.762 Judge, T. A., Piccolo, R. F., & Kosalka, T. (2009). The bright and dark sides of leader traits: A review and theoretical extension of the leader trait paradigm. The Leadership Quarterly, 20(6), 855–875. https://doi.org/ 10.1016/j.leaqua.2009.09.004 LeBreton, J. M., & Senter, J. L. (2008). Answers to 20 questions about interrater reliability and interrater agreement. Organizational Research Methods, 11(4), 815–852. https://doi.org/10.1177/1094428 106296642 Li, A. N., & Liao, H. (2014). How do leader-member exchange quality and differentiation affect performance in teams? An integrated multilevel dual process model. Journal of Applied Psychology, 99(5), 847–866. https://doi.org/10.1037/a0037233 Liao, H., Liu, D., & Loi, R. (2010). Looking at both sides of the social exchange coin: A social cognitive perspective on the joint effects of relationship quality and differentiation on creativity. Academy of Man- agement Journal, 53(5), 1090–1109. https://doi.org/10.5465/amj. 2010.54533207 Liden, R. C., Erdogan, B., Wayne, S. J., & Sparrowe, R. T. (2006). Leader- member exchange, differentiation, and task interdependence: Implica- tions for individual and group performance. Journal of Organizational Behavior, 27(6), 723– 746. https://doi.org/10.1002/job.409 Liu, D., Liao, H., & Loi, R. (2012). The dark side of leadership: A threelevel investigation of the cascading effect of abusive supervision on employee creativity. Academy of Management Journal, 55, 1187–1212. https://doi.org/10.5465/amj.2010.0400 Maccoby, M. (2000). Narcissistic leaders: The incredible pros, the inevita- ble cons. Harvard Business Review, 78(1), 68–78. Mackey, J. D., Huang, L., & He, W. (in press). You abuse and I criticize: An ego depletion and leader–member exchange examination of abusive supervision and destructive voice. Journal of Business Ethics, 1–13. https://doi.org/10.1007/s10551018-4024-x MacKinnon, D. P., Lockwood, C. M., & Williams, J. (2004). Confidence limits for the indirect effect: Distribution of the product and 92 resampling methods. Multivariate Behavioral Research, 39(1), 99–128. https://doi.org/10.1207/s15327906mbr3901_4 Morf, C. C., & Rhodewalt, F. (2001). Unraveling the paradoxes of narcis- sism: A dynamic self-regulatory processing model. Psychological Inquiry, 12(4), 177–196. https://doi.org/10.1207/s1532796 5pli1204_1 Morrison, E. W. (2011). Employee voice behavior: Integration and direc- tions for future research. The Academy of Management Annals, 5(1), 373–412. https://doi.org/10.5465/19416520.2011.574506 Morrison, E. W., & Milliken, F. J. (2000). Organizational silence: A barrier to change and development in a pluralistic world. Academy of Manage- ment Review, 25(4), 706–725. https://doi.org/10.5465/amr.2000. 3707697 Muthén, L. K., & Muthén, B. O. (2012). Mplus 7. Los Angeles, CA: Muthén & Muthén. Nevicka, B., Ten Velden, F. S., De Hoogh, A. H., & Van Vianen, A. E. (2011). Reality at odds with perceptions: Narcissistic leaders and group performance. Psychological Science, 22(10), 1259–1264. https://doi. org/10.1177/0956797611417259 O’Reilly, C. A., Doerr, B., Caldwell, D. F., & Chatman, J. A. (2014). Narcissis- tic CEOs and executive compensation. The Leadership Quarterly, 25(2), 218–231. https://doi.org/10.1016/j.leaqua.2013.08.002 Ong, C. W., Roberts, R., Arthur, C. A., Woodman, T., & Akehurst, S. (2016). The leadership is sinking: A temporal investigation of narcissistic lead- ership. Journal of Personality, 84(2), 237–247. https://doi.org/10. 1111/jopy.12155 Owens, B. P., Wallace, A. S., & Waldman, D. A. (2015). Leader narcissism and follower outcomes: The counterbalancing effect of leader humil- ity. Journal of Applied Psychology, 100(4), 1203– 1213. https://doi.org/ 10.1037/a0038698 Paulhus, D. L., & Williams, K. M. (2002). The dark triad of personality: Nar- cissism, Machiavellianism, and psychopathy. Journal of Research in Personality, 36(6), 556–563. https://doi.org/10.1016/s0092-6566(02) 00505-6 Preacher, K. J., Curran, P. J., & Bauer, D. J. (2006). Computational tools for probing interactions in multiple linear regression, multilevel modeling, and latent curve analysis. Journal of Educational and Behavioral Statistics, 31(4), 437–448. https://doi.org/10.3102/ 10769986031004437 Preacher, K. J., & Selig, J. P. (2012). Advantages of Monte Carlo confidence intervals for indirect effects. Communication Methods and Measures, 6(2), 77–98. https://doi.org/10.1080/19312458.2012.679848 Raskin, R., Novacek, J., & Hogan, R. (1991). Narcissism, self-esteem, and defensive self-enhancement. Journal of Personality, 59(1), 19–38. https://doi.org/10.1111/j.1467-6494.1991.tb00766.x Raskin, R., & Terry, H. (1988). A principal-components analysis of the Nar- cissistic Personality Inventory and further evidence of its construct validity. Journal of Personality and Social Psychology, 54(5), 890–902. https://doi.org/10.1037//0022-3514.54.5.890 Resick, C. J., Whitman, D. S., Weingarden, S. M., & Hiller, N. J. (2009). The bright-side and the dark-side of CEO personality: Examining core self- evaluations, narcissism, transformational leadership, and strategic influence. Journal of Applied Psychology, 94(6), 1365–1381. https://doi. org/10.1037/a0016238 Rogoza, R., Wyszyńska, P., Maćkiewicz, M., & Cieciuch, J. (2016). Differen- tiation of the two narcissistic faces in their relations to personality traits and basic values. Personality and Individual Differences, 95, 85–88. https://doi.org/10.1016/j.paid.2016.02.038 Rosenthal, S. A., & Pittinsky, T. L. (2006). Narcissistic leadership. The Leadership Quarterly, 17(6), 617–633. https://doi.org/10.1016/j.leaqua. 2006.10.005 Spain, S. M., Harms, P., & LeBreton, J. M. (2014). The dark side of person- ality at work. Journal of Organizational Behavior, 35(S1), S41–S60. https://doi.org/10.1002/job.1894 HUANG ET AL. Sui, Y., Wang, H., Kirkman, B. L., & Li, N. (2016). Understanding the curvi- linear relationships between LMX differentiation and team coordina- tion and performance. Personnel Psychology, 69(3), 559–597. https:// doi.org/10.1111/peps.12115 Tangirala, S., Green, S. G., & Ramanujam, R. (2007). In the shadow of the boss's boss: Effects of supervisors' upward exchange relationships on employees. Journal of Applied Psychology, 92(2), 309–320. https://doi. org/10.1037/0021-9010.92.2.309 Van Dyne, L., & LePine, J. A. (1998). Helping and voice extra-role behav- iors: Evidence of construct and predictive validity. Academy of Man- agement Journal, 41(1), 108–119. https://doi.org/10.2307/256902 Wilson, K. S., Sin, H. P., & Conlon, D. E. (2010). What about the leader in leader–member exchange? The impact of resource exchanges and substitutability on the leader. Academy of Management Review, 35(3), 358–372. https://doi.org/10.5465/amr.2010.51141654 Zhou, L., Wang, M., Chen, G., & Shi, J. (2012). Supervisors' upward exchange relationships and subordinate outcomes: Testing the multi- level mediation role of empowerment. Journal of Applied Psychology, 97(3), 668–680. https://doi.org/10.1037/a0026305 FOOTNOTE 1. We are thankful to an anonymous reviewer for this suggestion. 2. Before computing group-level LMX and empowerment to control for, we examined levels of interrater agreement for these two vari- ables. To do so, we computed rWG(J) indices for each group using a uniform null distribution to account for lack of agreement in the absence of rating biases, and three skewed distributions (slightly, moderately, and highly skewed) to account for potential leniency bias in subordinate ratings of LMX and empowerment as positive leader- ship phenomena. The resultant rWG(J)s varied across groups, with some groups having strong agreement that can be used to justify aggregation (>.70) and some groups failing to reach that level of agreement. Specifically, 40% (38) and 33% (31) of groups showed strong (>.70) and moderate (.51–.70) agreement, respectively, on LMX under at least one of the null distributions. Further, 61% (58) and 32% (30) of groups showed strong (>.70) and moderate (.51–.70) agreement, respectively, on empowerment under at least one of the null distributions. When groups reach different levels of agreement, a few options are available: deleting data from groups with low agreement (unde- sirable as it results in data loss), not aggregating, aggregating scores for all groups as the risk of diluting results with data from low- agreement groups, and including a dummy variable to denote high agreement as a moderator of the effect of the aggregated variable (LeBreton & Senter, 2008). We tested study hypotheses under each of these options, except the undesirable option of deleting data, and the pattern of results did not change. Given that (1) the result of hypothesis testing remains the same across options, (2) aggregated variables are not of substantive interest (i.e., are not part of the hypotheses) but are included as control variables, and (3) most groups in the sample demonstrated moderate to strong levels of agreement (73% on LMX and 93% on empowerment), in this paper, we report results with grouplevel LMX and empowerment. Results of hypothesis testing under other options are available from the authors upon request. HUANG ET AL. 93 ORCID Lei Huang https://orcid.org/0000-0001-9723-7757 AUTHOR BIOGRAPHIES Lei Huang is an assistant professor of Management at Auburn University. He received his PhD in Management from the Univer- sity of Nebraska–Lincoln. His current research interests include leadership, employee voice and creativity, and workplace mistreatment. Dina V. Krasikova is an associate professor of Management at the University of Texas at San Antonio. She received her PhD in Industrial/Organizational Psychology from Purdue University. Her research interests include leadership, employee well-being, and statistical methods. Peter D. Harms is an associate professor of Management at the University of Alabama. He received his PhD in Psychology from the University of Illinois at Urbana– Champaign. His current research interests include the assessment and development of personality, psychological well-being, and leadership. How to cite this article: Huang L, Krasikova DV, Harms PD. Avoiding or embracing social relationships? A conservation of resources perspective of leader narcissism, leader– member exchange differentiation, and follower voice. J Organ Behav. 2020;41:77–92. https://doi.org/10.1002/job.2423