Abusive Supervisory Reactions to Coworker Relationship Conflict

The Leadership Quarterly 22 (2011) 10101023 Contents lists available at ScienceDirect The Leadership Quarterly j o u r n a l h o m e p a g e w w w. e l s ev i e r. c o m / l o c a t e / l e a q u a inglorious executive programy programy reactions to co failer family kind inpatient? ict Kenneth J. Harris a,? , Paul Harvey b, K. Michele Kacmar cIndiana University Southeast, School of Business, 4201 Grant Line Road, New Albany, IN 47150, USA cargon Department, Whitte much School of Business and E scamomics, University of New Hampshire, USA Department of Man senescement and Marketing, Culverhouse College of Commerce and Business Administration, 143 Alston H every(prenominal), Box 870225, The University of Alabama, Tuscaloosa, Alabama 35487-0225, USA b c a a r t i c l e i n f o a b s t r a c t This study extends seek on shameful lapse by exploring how supervisor reports of determineflict with their co travelers argon link up to opprobrious doingss and resulting outcomes.We hold research on displaced onslaught, abductflict, and loss leadermember exchange (LMX) theory to formulate our hypotheses. Results from two samples of 121 and 134 equaliseed supervisor suppress dyads support the idea that supervisors experiencing co practiseer kindred nobbleflict atomic number 18 promising to engage in inglorious behaviors directed toward their subordinates and that LMX graphic symbol moderates this relationship. Additionally, ignominious checktrol was associated with decreased scat trial and organizational citizenship behaviors (OCB).Results in any case indicate that in two samples disgraceful memorizetrol mediates the relationships mingled with supervisor reports of co puzzle outer relationship filchflict and OCB, and in iodine sample mediates the association between supervisor-reported co wrencher relationship ascertainflict and work stew. 2011 Elsevier Inc. All rights reserved. Available online 10 August 2011 Keywords opprobrious supervision Coworker relationship fiddle? ict Multi-level 1. Introduction disgraceful supervision, or the prolonged hostile treatment of subordinates, has been recognized as a signi? ant threat to employee well being and productivity in both the popular press (e. g. , Elmer, 2006) and in organizational research (e. g. , Duffy, Ganster, & Pagon, 2002 Harris, Kacmar, & Zivnuska, 2007 Harvey, Stoner, Hochwarter, & Kacmar, 2007 Hoobler & Brass, 2006 Mitchell & Ambrose, 2007 Tepper, 2000, 2007 Tepper, Duffy, & Shaw, 2001 Zellars, Tepper, & Duffy, 2002). Behaviors that draw under the umbrella of disgraceful supervision, such as sabotaging, yelling at, or ignoring subordinates, have been linked to an array of ostracise convictsequences (see Tepper, 2007 for an over vista).Research to a fault educes that these forms of tread are alarmingly special K in modern organizations (Namie & Namie, 2000 Tepper, 2007). The purpose of this study is to develop and test a conceptual example that expands our knowledge of antecedents, moderators, and consequences of disgraceful supervision. We also build on past research showing that supervisors relationship con? icts quite a little trickle down to subordinates in the form of abusive behaviors (Aryee, Chen, Sun, & Debrah, 2007). Speci? cally, we test the ruling that supervisors who experience relationship con? ct, de? ned as interpersonal tension, animosity, and annoyance (Jehn, 1995, p. 258), with their coworkers reply by abusing subordinates. The proposed relationship between supervisor-level coworker relationship con? ict and abusive supervision is rooted in the notion of displaced aggression, which occurs when the reaction to an unpleasant outcome or behavior from one source is redirected to a second source (Miller, Pedersen, Earlywine, & Pollock, 2003 Tedeschi & Norman, 1985).Consistent with Tepper (2007), we argue that the relatively weak retaliatory post of subordinates, as compared to coworkers, increases t he likelihood that relationship con? ict-driven frustration give be vented at subordinates. We qualify this assumption, however, by arguing that supervisors who experience coworker relationship con? ict will not behave abusively toward all of their subordinates. We explore ? Corresponding author. E-mail addresses emailprotected edu (K. J. Harris), Paul. emailprotected edu (P. Harvey), emailprotected ua. edu (K. M. Kacmar). 1048-9843/$ see front matter 2011 Elsevier Inc.All rights reserved. doi10. 1016/j. leaqua. 2011. 07. 020 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 1011 this idea by examining leadermember relationship (LMX) bore as a moderator of the relationship between supervisors levels of coworker relationship con? ict and abusive supervision. Finally, we advance the living research by investigating two supervisorrated employee outcomes (work ride, and organizational citizenship behaviors (OCB)), one of which has not previously been examined in th e context of abusive supervision.These outcomes were chosen as they extend the publications and we were interested in actual behaviors directed toward the seam/task (work effort and task-foc utilize OCB). We examine these relationships, shown in Fig. 1, in two separate samples of matched supervisorsubordinate dyads. Thus, the period study makes several contri exactlyions to the literature. First, we examine the in? uence of con? ict between supervisors on subordinate reports of abusive supervision. Examining this relationship is fundamental because although coworker relationship con? cts have negative outcomes, studies have yet to investigate how supervisors experiencing these con? icts treat their subordinates. Second, we investigate LMX quality as a relationship multivariate that changes how supervisor reports of coworker relationship con? ict and abusive supervision are related. Third, we extend the nomological network of abusive supervision by examining the outcomes of work effort and OCB. Finally, we investigate the potential for abusive supervision to mediate the associations between supervisor reports of coworker relationship con? ict and distal consequences.Thus, this study takes a ? rst musical note toward explaining how (through the intermediary mechanism of abusive supervision) supervisors experiences of coworker relationship con? ict ultimately impact important job outcomes. 2. Abuse as a displaced response to coworker relationship con? ict Abusive supervision is de? ned as prolonged hostile treatment toward subordinates, excluding physical fierceness (Tepper, 2000). Research indicates that supervisors who perceive that they are victims of fundamental interactional or procedural injustice, both of which may be associated with coworker relationship con? ct (Fox, Spector, & Miles, 2001), are relatively more than than probably than new(prenominal)s to evil their subordinates (Aryee, Chen, Sun, & Debrah, 2007 Tepper, Duffy, Henle, & Lambe rt, 2006). Tepper, Duffy, Henle, and Lambert (2006) argued that this trickle-down power, in which supervisors frustrations are channeled into abusive behaviors targeted at subordinates, may occur because subordinates are a relatively reliable target toward which supervisors roll in the hay vent their frustrations (Tepper, Duffy, Henle, & Lambert, 2006).This argument suggests abusive supervision may be a response to frustrating workplace events such as coworker relationship con? ict. Coworker con? ict has been linked to undesirable emotional states and nooky negatively impact interpersonal relationships (e. g. , Bergmann & Volkema, 1994 Deutch, 1969). Emotion research suggests that the anger and frustration associated with interpersonal con? ict notify conjure verbal (e. g. , shouting) and behavioral (e. g. , theft, sabotage, violence) aggression toward those who stimulate the con? ct (e. g. , Ambrose, Seabright, & Schminke, 2002 Dollard, Doob, Miller, Mowrer, & Sears, 1939 Fox & Spector, 1999 Greenberg, 1990 Spector, 1975). Many of these behaviors, with the exception of physical violence, would fall under Teppers (2000) de? nition of abusive supervision if aimed at subordinates. Drawing on ? ndings from research on displaced aggression we argue that, due to the relative power of supervisors coworkers, these relationship con? ict-driven behaviors index, in concomitant, be targeted at subordinates.Displaced aggression occurs when individuals experience mistreatment from one party and respond by mistreating a second party (Hoobler & Brass, 2006, Miller, Pedersen, Earlywine & Pollock, 2003, Twenge & Campbell, 2003). Several triggers of displaced aggression have been identi? ed, including social rejection (Twenge & Campbell, 2003) and negative feedback (Bushman & Baumeister, 1998). Hoobler and Brass (2006) also showed that abusive supervision at work finish promote displaced aggression toward family members at home. We examine abusive supervision as a form of displaced aggression ather than a predictor, although both conceptualisations are logical. Displaced aggression is often triggered by unpleasant workplace events (e. g. , Miller, Pedersen, Earlywine & Pollock, 2003) and abusive supervision ? ts this criteria. We argue that abusive supervision also can ? t the criteria of displaced aggression if it is triggered by events beyond the control of subordinates, such as the twistrs coworker relationship con? ict. Thus, abusive supervision can probably be both a cause of displaced aggression and a type of displaced aggression. cable Dashed lines re give way hypothesized talk terms linkages supervisor-Rated promoter Work Effort Supervisor-Rated Coworker Conflict Abusive Supervision Supervisor-Rated Subordinate TaskFocused OCB Moderator Leader-Member Exchange Fig. 1. Hypothesized model. 1012 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 As Tepper, Duffy, Henle and Lambert (2006) argued, abusive supervision can be used as a agent for run frustration because subordinates have relatively low levels of retaliatory power and, thitherfore, serve as a lower-risk target for vent behaviors than do employees in positions of greater hierarchal power.Victim precipitation research also supports this logic, indicating that displaced aggression is often targeted at those who are uneffective or unwilling to defend themselves, as is likely the case among subordinates who can be disciplined and terminated by their supervisors (e. g. , Aquino, 2000). This desire to vent frustration at individuals who are unassociated with the initial con? ict, similar to the anecdotal notion of recoil the dog after a bad day at work, can be understood in the context of displaced aggression. Coworker relationship con? ct is a potent source of stress and frustration (Thomas, 1976, 1992) and, because these are unpleasant, individuals are motivated to engage in coping behaviors that will diminish their presence (Kemper, 196 6). These emotion-driven coping behaviors can often take the form of hostile behaviors such as sabotage (Ambrose, Seabright & Schminke, 2002) and verbal assaults (Douglas & Martinko, 2001). Thus, coworker relationship con? ict may trigger aggressive behaviors (e. g. , yelling at others) that serve a coping function. Thomas (1976) noted, however, that the relative power of the parties to a con? ct in? uences the manner in which both parties will respond. When authorized power levels are equal, as in the case of coworkers, hostile responses are likely to be met with retaliation although it is possible that the target of retaliation will respond with additional hostility, creating an escalating cycle of con? ict. Subordinates, on the other hand, are often reluctant to respond in kind to hostile supervisor behaviors for fear of losing their jobs. The fact that subordinates are not the cause of the supervisors frustration, that is, the frustration is caused by supervisors con? ct with t heir coworkers, may have little impact on the behavioral response if the behavior is largely motivated by emotion as opposed to logic. That is, the desire to vent anger over coworker relationship con? ict using a galosh target may override concerns that subordinates are not the logical targets for retaliation, given that they are not the cause of the con? ict. Based on these arguments, we predict dead reckoning 1. Supervisors reports of coworker relationship con? ict are positively associated with abusive supervisory behaviors, as rated by subordinates. 2. 1. The moderating in? ence of LMX relationship quality Thomas (1976, 1992) argued that a conceptualization process occurs between the con? ict experience and the behavioral outcome in which information is processed and behavioral options are evaluated. Although this cognitive process is likely to re limn a across-the-board range of information, we argue that an evaluation of relationships with subordinates is particularly rele vant when behaviors toward these individuals are concerned. LMX theory suggests that the quality of leadermember relationships varies from high to low (Dienesch & Liden, 1986 Graen & Uhl-Bien, 1995).Subordinates in high quality exchanges are seen more favorably and satisfy advantages from their supervisors that their low quality LMX counterparts do not (e. g. , Liden, Sparrowe, & Wayne, 1997). As such, members in high quality exchanges receive preferential treatment from supervisors who are motivated to maintain these productive relationships. We expect that supervisors who experience high levels of coworker relationship con? ict may become abusive toward subordinates, but will be selective in choosing which subordinates to target. Abusive supervisory behaviors generally have a negative effect on ictims levels of want and attitudes toward their jobs (e. g. , Duffy, Ganster & Pagon, 2002 Schat, Desmarais, & Kelloway, 2006). Although it can be argued that effective managers would no t want to risk these consequences with any employees, LMX theory would suggest that supervisors are especially motivated to maintain effective relationships with their high quality LMX subordinates. We argue, therefore, that supervisors who are frustrated by coworker relationship con? ict and who choose to react in an abusive manner will generally choose low quality LMX subordinates as their targets.Put differently, we expect that when con? ict-driven abuse occurs, members in low quality exchanges will experience it more lovesomely and frequently than members in high quality exchanges. legal expert and victim precipitation theories provide additional support for this argument (e. g. , Aquino, 2000 Bies & Moag, 1986). From a justice perspective, instead of perceiving members of low quality LMX relationships as less risky targets for abuse, it can also be argued that supervisors ? nd it easier to justify abuse toward these employees. Members of low quality exchanges are often charac terized by relatively low performance levels (e. . , Deluga & Perry, 1994 Liden, Wayne, & Stilwell, 1993), and it capacity be argued that supervisors who use abusive behaviors to cope with relationship con? ict-driven frustration will witness about justi? ed in reduceing on these employees. That is, supervisors might hack the abuse by convincing themselves that relatively lowperforming subordinates in low quality LMX relationships deserve the abusive behavior. Victim precipitation research also suggests that several characteristics common among low quality LMX subordinates make them likely targets of abuse.Although provocative and threatening behaviors have been linked to retaliatory aggression (e. g. , Aquino & Byron, 2002 Tepper, 2007), more salient to our focus on leadermember relationships is the precipitation research indicating that abusive individuals often target those who are seen as weak or defenseless. Individuals who are hesitant to defend themselves or linear pers pective themselves or their situations negatively appear to draw the attention of aggressive individuals (Aquino, 2000 Olweus, 1978 Rahim, 1983 Tepper, 2007).As discussed above, the hierarchical nature of their relationship likely promotes the former tendency among subordinates, making them relatively safe targets for abuse. Members in low quality exchanges, in particular, might be unwilling to further jeopardize their relationship with their supervisors by retaliating against abuse and might also impute their undesirable status, promoting the negative perceptions of their workplace competence and situation (e. g. , Ferris, Brown, & Heller, 2009) that can provoke victimization. akin(predicate) to our arguments concerning displaced abuse of subordinates, victim precipitation research suggests that these aggressors might wish to engage in abusive behavior as a means to K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 1013 preserve their social standing and bolster pe rceptions of their control over a situation (e. g. , Baumeister, Smart, & Boden, 1996 Felson, 1978). As such, this line of research reinforces the notion that subordinates might be targeted for displaced abuse and suggests that low quality LMX subordinates are especially likely to be viewed as vulnerable, and therefore relatively safe, targets.Based on these arguments, we predict Hypothesis 2. The relationship between supervisor-reported coworker relationship con? ict and member-reported abusive supervision is moderated by LMX, such that the positive relationship is steadyer when LMX relationship quality is lower. 2. 2. Outcomes of abusive supervision The outcome portion of our conceptual model, shown in Fig. 1, examines the effects of abusive supervisory responses to coworker relationship con? ict on work effort and OCB. While we do not posit that abusive supervision is the unless figure mediating the relationships between supervisors coworker relationship con? ct and these outc omes, we argue that abuse can serve as an explanatory mechanism and explain a relevant amount of variance in each consequence. Abusive supervision is a negative workplace event that, like con? ict, can have negative attitudinal and behavioral consequences (Tepper, 2007 Tepper, Henle, Lambert, Giacalone, & Duffy, 2008 Tepper, Moss, Lockhart, & Carr, 2007). It has been argued that these outcomes are caused by the stress and emotional strain associated with abuse from individuals in a position of power (e. g. Duffy, Ganster & Pagon, 2002 Harvey, Stoner, Hochwarter & Kacmar, 2007 Tepper, 2000). Further, Duffy, Ganster and Pagon (2002) be evidence suggesting that abuse promotes diminished self-ef? cacy. As we discuss in the following sections, each of these consequences of abusive supervision can be logically linked to the outcomes depicted in Fig. 1. 2. 2. 1. Work effort Because abusive supervision can diminish victims con? dence in their abilities (Duffy, Ganster & Pagon, 2002), it fo llows that motivation to exert high levels of effort at work will likely decrease in response to abuse.Abusive supervisors, who by de? nition are consistent in their abuse (Tepper, 2000), might eventually wear employees down with a steady onslaught of aggressive behavior (e. g. , yelling, criticizing), reducing their con? dence and motivation. Similarly, it may be that over period abusive supervision promotes emotional exhaustion (Harvey, Stoner, Hochwarter & Kacmar, 2007 Tepper, 2000), a condition characterized by diminished emotional and physical coping abilities and closely associated with job burnout (Brewer & Shapard, 2004 Cropanzano, Rupp, & Byrne, 2003).Harvey, Stoner, Hochwarter and Kacmar (2007) argued that this relationship was likely due to the persistent assault on employees feelings and ef? cacy perceptions (Savicki & Cooley, 1983) associated with abusive supervision. When emotional exhaustion occurs, individuals demonstrate diminished motivation and a reduced ability to handle disagreeable work events, promoting a reduction in work effort (Brewer & Shapard, 2004 Kahill, 1988 Leiter & Maslach, 1988).Using a different lens to view the abusework effort association, employees might also view abusive supervision as a form of psychological contract breach, as subordinates generally do not expect to be step by those given the authority to supervise them (Tepper, 2000). When employees perceive that a breach has taken place, they often feel less compelled to ful? ll their obligation to exert high levels of work effort (Harris, Kacmar & Zivnuska, 2007). 2. 2. 2. Citizenship behaviors The ? nal outcome depicted in Fig. 1 concerns the negative in? ence of coworker relationship con? ict-driven abuse and subordinates propensity to engage in OCB. This predicted relationship is found on research indicating that abusive supervision is associated with factors, including decreased organizational commitment, poor work-related attitudes, and injustice perceptions (Aryee, Chen, Sun & Debrah, 2007 Duffy, Ganster & Pagon, 2002 Schat, Desmarais, & Kelloway, 2006 Zellars, Tepper & Duffy, 2002), that can inhibit citizenship behaviors (Ambrose, Seabright & Schminke, 2002 Zellars, Tepper & Duffy, 2002).Victims of abusive supervision often feel that they have been treated unjustly (Tepper, 2000), a perception that is associated with reduced levels of OCB (Moorman, 1991). As Judge, Scott, and Ilies (2006) argued, unjust treatment is likely to qualify as a negative affective event and can therefore provoke a retaliatory behavioral response. One such response could logically be the withholding of citizenship behaviors, which are not a requirement of the job and could run counter to the goal of retaliation by making the supervisors job easier (e. g. , Zellars, Tepper & Duffy, 2002).In support of this reasoning, additional research indicates that abusive supervision motivates retaliatory behaviors such as workplace deviance and aggression that run contr ary to the notion of citizenship behavior (Dupre, Inness, Connelly, Barling, & Hoption, 2006 Schaubhut, Adams, & Jex, 2004). Based on these arguments, we predict Hypothesis 3. Abusive supervision is negatively related to supervisor reports of subordinate work effort and organizational citizenship behaviors. 2. 3. The mediating role of abusive supervision We have argued that relationship con? ct between supervisors and their coworkers is associated with abusive supervisory behaviors, and that such behaviors have negative implications for victims levels of work effort and OCB. Implicit in this line of reasoning is the notion that coworker relationship con? ict at the supervisor level is ultimately associated with decreased levels of 1014 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 effort and OCB at the subordinate level, and that abusive supervision acts a intermediator between these variables. More speci? ally, the negative effects of supervisors relationship c on? ict with their coworkers are predicted to manifest themselves in the form of abusive behaviors that negatively affect employees attitudes and behaviors, promoting negative subordinate outcomes. Thus, while a relationship between a supervisors level of coworker relationship con? ict and subordinates levels of effort and OCB may seem somewhat abstract, we suggest that coworker relationship con? ict-driven abusive supervision provides an intermediary link between these variables.Based on these arguments, we predict Hypothesis 4. Abusive supervision mediates the negative relationships between supervisor-rated coworker relationship con? ict and work effort and organizational citizenship behaviors. 3. Method 3. 1. precedents and procedures The samples utilized in this study were from two different divisions of a state government. The division in pattern 1 was prudent for handling disease related issues (e. g. , STDs, immunizations, tuberculosis), whereas the division in taste 2 ha ndled environmental health related issues (e. g. , radiation, clean water).To begin the data collection efforts, the theatre director of each division sent an email to all employees in their branch. The email in make the potential respondents of the studys purpose, that participation was voluntary, and that the results would be con? dential. After this email, the researchers sent a personalized message again explaining the goal of the survey, the con? dentiality of responses, and a web link to the survey. Respondents were asked to complete the survey during the next month. Respondents were required to provide their supervisors name to match supervisorsubordinate responses.At the same epoch, supervisors were asked to provide ratings on each of their direct reports. In Sample 1, eliminating responses with missing data or those that were unable to be matched (i. e. , we true a subordinate response, but not a matching supervisor response) resulted in a sample size of 121 (58% respons e rate). Subordinates were 68% female, the mean(a) age was 41. 68 years, the fair job advance was 3. 38 years, and their average organizational promote was 5. 22 years. In total, 28 supervisors provided ratings, resulting in an average of 4. 32 ratings per supervisor.For the supervisors, the demographic breakdown was 57% female, the average age was 47. 91 years, the average job tenure was 4. 79 years, and their average organizational tenure was 7. 73 years. After the voidance of unusable responses in Sample 2, our usable sample size was 134 (64% response rate). Participants in Sample 2 were 60% male, had an average age of 46. 04 years, average job tenure of 7. 04 years, and average organizational tenure of 11. 51 years. Forty-four supervisors provided ratings, which resulted in an average of 3. 05 ratings per supervisor.The demographic breakdown for the supervisors was 75% male, an average age of 49. 29 years, average job tenure of 9. 64 years, and average organizational tenure of 16. 26 years. 3. 2. Measures Unless otherwise noted, a 5-point Likert home (anchors strongly disagree (1) to strongly agree (5)) was used for all survey items. Scales were coded with high values representing high levels of the constructs. 3. 3. Subordinate greenbacks 3. 3. 1. Abusive supervision In both samples abusive supervision was measured with six items from Teppers (2000) measure.We were unable to use the full 15-item measure due to management concerns about the surveys overall length. Thus, we had experts in the area look at the content of each of the items, and we chose 6 items that outflank captured the full range of abusive supervisory behaviors. The items we chose were My supervisor makes negative comments about me to others, My supervisor gives me the silent treatment, My supervisor expresses anger at me when he/she is mad for another(prenominal) reason, My supervisor is rude to me, My supervisor breaks promises he/she makes, and My supervisor puts me down in fron t of others. In an effort to establish the validity of our shortened subdue, we compared our reduced photographic plate to the full measure using the data from the Tepper (2000) article. 1 We found that the full 15-item scale was correlated with our 6-item scale at . 96. The Cronbach alpha for the scale was . 90 for Sample 1 and . 92 for Sample 2. 3. 3. 2. Leadermember exchange We used Liden and Maslyns (1998) 12-item leadermember exchange multidimensional scale to measure exchange quality in both samples. A sample item included My supervisor would defend me to others in the organization if I made an honest mistake. The Cronbach alpha for the scale was . 94 for Sample 1 and . 92 for Sample 2. 1 We thank Ben Tepper for catering us to use his original data for this correlation. K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 1015 3. 4. Supervisor measures 3. 4. 1. Coworker relationship con? ict In both samples supervisors rated their relationship con? icts with th eir coworkers using the 4-item Jehn (1995) scale. A sample item included Is there tension among your coworkers? These questions were included in a section of the survey here the supervisors were answering questions about their attitudes, behaviors, and relationships with their coworkers. This section was separate from the section where supervisors commented on their subordinates, thus making it clear that these relationship con? ict questions were focused on coworkers at their level in the organization (e. g. , managers relationship con? icts with other managers). The response scale for this construct was Not at all (1) to To a very great extent (5). The Cronbach alpha for the scale was . 95 for Sample 1 and . 94 for Sample 2. 3. 4. 2.Work effort In both samples supervisors rated subordinates work effort using Brown and Leighs (1996) 5-item scale. A sample item was When theres a job to be done, this subordinate devotes all his/her energy to getting it done. The Cronbach alpha for the scale was . 93 for Sample 1 and . 94 for Sample 2. 3. 4. 3. Organizational citizenship behaviors Supervisors responded to Settoon and Mossholders (2002) 6-item scale to measure subordinate task-focused OCB in both samples. A sample item was This subordinate assists coworkers with heavy work loads even though it is not part of the job. The Cronbach alpha for the scale was . 84 for Sample 1 and . 81 for Sample 2. 3. 5. condition variables We controlled for four variables, all measured from the subordinate, in an effort to minimize potentially spurious relationships. The variables we controlled for were age (measured in years), job tenure (measured in months), organizational tenure (measured in months), and supervisorsubordinate relationship tenure (measured in months). 3. 6. Analytical approach In both samples in this study, supervisors coworker relationship con? ict responses were used as predictors of subordinate outcomes (i. . , cross-level main effect). Thus, a single supervi sor coworker relationship con? ict rating was used as the predictor variable for multiple subordinates. As a result, for these variables there was no within-supervisor variance and all of the variance was between supervisors (i. e. , ICCs were 1. 00). Additionally, supervisors provided ratings on certain scales (e. g. , work effort and OCB) for multiple subordinates, thus resulting in a supervisor effect (e. g. , ICC1s for OCB of . 11 in sample 1 and . 13 and sample 2, and ICC2s of . 48 in sample 1 and . 51 in sample 2).To study for the supervisor-level effect in our data, hierarchical elongate modeling (HLM Raudenbush, Bryk, Cheong, & Congdon, 2004) with grand-mean centering was used to carry out our analyses. In the HLM analyses involving supervisor-rated coworker relationship con? ict, this variable was included as a take 2 variable (Raudenbush, Bryk, Cheong & Congdon, 2004). To test Hypotheses 12, there were four steps. In the ? rst step, we entered the four control variables . In the second step we entered the Level 2 variable of supervisor-rated coworker relationship con? ict, and it was here that we tested Hypothesis 1.In the third step, we entered the Level 1 moderator variable, LMX. In the fourth step, we entered the cross-level interaction term formed between supervisor-rated coworker relationship con? ict and LMX. It was in this step that we tested Hypothesis 2. To test the abusive supervision-outcome and mediation hypotheses (3 and 4), we conducted Baron and Kennys (1986) threestep procedure. The HLM equations are available from the ? rst author request. 4. Results The means, modular deviations, and correlation matrix for the variables in this study are provided in hold over 1 for Sample 1 and Table 2 for Sample 2.In both samples abusive supervision was signi? cantly correlated with supervisor reports of coworker relationship con? ict, as well as our dependent variables. Given that a few of the correlations between our focal variables were hi gh, we take to run a series of con? rmatory factor analyses (CFA) on the scales used in our study to ensure that they were independent and that the items produced the expected factor structures. These analyses were run on both samples separately. To conduct our CFAs, we used LISREL 8. 80, a covariance matrix as input, and a maximum-likelihood estimation.We elected to conduct our CFA analyses using composite indicators rather than items due to the large scrap of items and our moderate sample sizes. To create our composite indicators, we assigned items based on factor loadings from an exploratory factor analysis (Bagozzi & Heatherton, 1994 Eddleston, Viega, & Powell, 2006). Speci? cally, for our four-item scales we combined the two items with the highest and lowest factor loadings to the ? rst indicator and the remaining two items to the second indicator. For the ? ve-item scales we created the ? st indicator as described above and included the remaining three items on the second in dicator. For our six-item scale we mated the highest and lowest loading item to create the ? rst indicator and then repeated this process for the remaining two indicators. Finally, for the LMX scale we used the four subscales (loyalty, contribution, professional respect, and affect) as composite indicators. Our approach resulted in 15 indicators for our 6 scales. 1016 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 Table 1 smasheds, standard deviations, and intercorrelations among study variables in Sample 1.Variable 1. Abusive supervision 2. Sup. coworker con? ict 3. Leadermember exchange (LMX) 4. Work effort 5. OCB 6. LMX affect 7. LMX contribution 8. LMX loyalty 9. LMX professional respect 10. Age 11. Job tenure 12. Organizational tenure 13. Relationship tenure Mean 1. 31 3. 03 3. 92 4. 03 3. 87 3. 86 4. 10 3. 69 4. 03 41. 68 3. 38 5. 22 1. 99 SD . 57 1. 02 . 77 . 79 . 72 . 97 . 68 . 84 1. 09 11. 1 3. 88 5. 23 2. 02 1 . 77 . 21? ? . 67 ? . 27 ? . 29 .60 .36 .69 .62 .10 . 10 . 05 . 25 2 . 95 ? .11 ? .20? ? . 18? ? . 05 . 04 . 19? ? . 14 . 01 . 23? .01 . 17 3 4 5 6 7 8 9 10 11 12 .76 . 3 .35 .91 .77 .83 .90 ? . 00 . 05 . 08 ? .00 .86 . 40 .28 .22? .35 .28 .03 ? .00 . 10 . 00 .65 . 27 .22? .33 .35 .01 ? .03 . 05 . 12 .92 . 62 .68 .79 ? . 02 . 11 . 11 . 04 .75 . 56 .58 .11 . 05 . 11 . 04 .74 . 64 ? . 04 ? .01 . 05 ? .11 .94 ? .03 . 02 . 01 . 02 . 35 .39 .26 . 69 .48 . 49 Note value in italics on the diagonal are the square root of the average variance explained which moldiness be larger than all zero-order correlations in the row and tower in which they appear to demonstrate discriminant validity (Fornell & Larcker, 1981).N = 121. ? p b . 05. p b . 01. We began by estimating a six-factor solution, with each factor representing a scale in our study. Fit indices, shown in Table 3, indicate that the six-factor model ? t the data. To verify that the six-factor structure was the outdo representation of our data, we estimated three alternative models and compared them to our baseline model via chi-square difference tests. The alternative models estimated included two ? ve-factor models and a unidimensional model. The alternative models were created by combining scales that had strong correlations to form a larger factor.The ? rst alternative model combined abusive supervision and LMX into one factor while the second combined OCB and work effort. A description of each alternative model and the CFA results are offered in Table 3. As shown in Table 3, the chi-square difference test results support the six-factor structure as originally designed. To further explore the discriminant validity of our scales we followed the procedure outlined by Fornell and Larcker (1981) and calculated the square root of the average variance explained for each of the scales in our study.This value, which we present on the diagonal in Tables 1 and 2, represents the variance accounted for by the items that compose the scale. To demonstrate discri minant validity, this value mustiness exceed the corresponding latent variable correlations in the same row and column. If this condition is met, then we have evidence that the variance shared between any two constructs is less than the average variance explained by the items that compose the scale (i. e. , discriminant validity). As shown in Tables 1 and 2, this condition is met for all of the scales used in our study.The HLM results predicting abusive supervision are shown in Tables 4 (for Sample 1) and 5 (for Sample 2) and the HLM results investigating abusive supervision as a mediator and/or predictor are provided in Tables 6 and 7. First describing our interaction results in Table 4, step 1 reveals that relationship tenure (? = . 08, p b . 05) was the only control variable signi? cantly associated with abusive supervision. footstep 2 shows that supervisor reports of coworker relationship con? ict are positively and signi? cantly related to abusive supervision (? = . 09, p b . 05).This result provides support for Hypothesis 1 in Sample 1. pure tone 3 in this analysis shows that LMX was negatively associated with abusive supervision (? = ?. 48, p b . 01). Finally, step 4 shows that the interaction term between supervisor reports of coworker relationship con? ict and LMX was negatively and signi? cantly related to abusive Table 2 Means, standard deviations, and intercorrelations among study variables in Sample 2. Variable 1. Abusive supervision 2. Sup. coworker con? ict 3. LMX12 (overall) 4. Work effort 5. OCB 6. LMX affect 7. LMX contribution 8. LMX loyalty 9.LMX professional respect 10. Age 11. Job tenure 12. Organizational tenure 13. Relationship tenure Mean 1. 32 2. 42 4. 04 4. 31 4. 31 4. 04 4. 15 3. 78 4. 19 45. 86 6. 55 11. 16 6. 08 SD . 58 . 76 . 60 . 73 . 67 . 78 . 56 . 78 . 95 6. 89 2. 66 4. 37 2. 12 1 . 92 . 15? ? . 55 ? . 26 ? . 21? ? . 53 .05 ? .52 ? . 57 .04 . 02 . 01 ? .01 2 . 94 ? .04 ? .03 ? .19? ? . 03 ? .06 ? .02 ? .02 ? .15 ? .09 ? .07 . 00 3 4 5 6 7 8 9 10 11 12 .92 . 09 . 05 . 84 .53 .83 .86 ? . 07 . 08 . 05 . 07 .87 . 72 ? . 01 ? .03 . 18? .11 ? .03 ? .00 . 03 ? .02 .85 . 01 ? .13 . 09 . 13 ? .13 . 1 ? .05 . 07 .88 . 28 .56 .69 ? . 10 . 05 ? .03 . 00 .71 . 38 .22? .08 . 16* . 18? .15 .84 . 59 ? . 08 . 03 . 03 . 01 .95 ? .06 . 04 . 01 . 08 . 14 . 23 .18? . 61 .27 . 26 Note Values in italics on the diagonal are the square root of the average variance explained which must be larger than all zero-order correlations in the row and column in which they appear to demonstrate discriminant validity (Fornell & Larcker, 1981). N = 134. ? p b . 05. p b . 01. K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 Table 3 substitute model test results.Model Sample 1 (N = 121) Baseline 6-factor model 5-factor combining abuse and LMX 5-factor combining work effort and OCB 1-factor Sample 2 (N = 134) Baseline 6-factor model 5-factor combining abuse and LMX 5-factor combining work effort and OCB 1-factor X2 102 196 127 706 df 75 80 80 90 X2diff dfdiff CFI . 98 . 95 . 97 . 59 NFI . 95 . 91 . 94 . 57 1017 RMSEA . 048 . 093 . 059 . 200 94 25 604 5 5 15 112 276 224 1177 75 80 80 90 164 112 1065 5 5 15 .98 . 93 . 93 . 47 .94 . 89 . 89 . 46 .056 . 125 . 107 . 280 Note Abuse = abusive supervision, LMX = leadermember exchange, OCB = organizational citizenship behaviors. p b . 001. supervision (? = ?. 12, p b . 01). Overall, the results in Table 5 (Sample 2) are similar. In step 1 none of the control variables were signi? cantly associated with the outcome, but in step 2, supervisor reports of coworker relationship con? ict were positively and signi? cantly related to abusive supervision (? = . 11, p b . 05), again supporting Hypothesis 1. Step 3 in Table 5 shows that LMX was negatively associated with abusive supervision (? = ?. 54, p b . 01). In the ? nal step, the supervisor reported coworker relationship con? ict ? LMX interaction term was negatively and signi? antly related to abusive supervi sion (? = ? .29, p b . 05). To determine support for our interaction hypothesis, we graphed the two signi? cant moderating effects. We did so by plotting two slopes, one at one standard deviation below and one at one standard deviation above the mean (Stone & Hollenbeck, 1989). Figs. 2 (for Sample 1) and 3 (for Sample 2) illustrate the signi? cant interactions and show that the positive relationships between supervisor reports of coworker relationship con? ict and abusive supervision were stronger when LMX relationship quality was lower.Additionally, we calculated open slopes for each of our interactions. In sample 1, we found that the slope of the low LMX line was signi? cant (t = 2. 00, p b . 05), whereas the slope of the high LMX line was not signi? cant. Similar to sample 1, in sample 2 the slope of the low LMX was signi? cant (t = 2. 11, p b . 05), but the slope of the high LMX line was not signi? cant. In total, these results provide support for Hypothesis 2 in both samples. Tables 6 and 7 provide the results of our mediation analyses. First discussing the results from Sample 1 shown in Table 6, supervisor-reported coworker relationship con? ct was signi? cantly related to abusive supervision (? = . 09, p b . 05) (which ful? lls one of Baron and Kennys (1986) mediation requirements) and to OCB (? = ? .08, p b . 10) and work effort (? = ?. 14, p b . 05) (ful? lling another mediation requirement). Steps 2c and 3c show that when both supervisor reports of coworker relationship con? ict and abusive supervision are entered into the equation, the coworker relationship con? ict variable is no longer signi? cant. In particular, the gammas for supervisor-reported coworker relationship con? ict predicting OCB dropped from ?. 08 to ?. 6 and for predicting work effort dropped from ?. 14 to ? .11. However, abusive supervision is signi? cantly and positively related to OCB (? = ?. 37, p b . 01) and signi? cantly and negatively related to work effort (? = ?. 27, p b . 05). Thus, Hypothesis 3 is supported in Sample 1. In terms of the mediation results, the results from Baron and Kennys (1986) three-step procedure show that abusive supervision fully mediated the relationship between supervisor-rated coworker relationship con? ict and OCB and partially mediated the relationship with work effort. Thus, Hypothesis 4 was supported in Sample 1.Table 4 Hierarchical linear modeling results predicting abusive supervision in Sample 1. Step 1 Control variables Age Job tenure Organizational tenure Relationship tenure Independent variable Sup-rated coworker con? ict (A) Moderator LMX (B) interaction term A? B ? R2 . 00 . 00 ? .01 . 08? Step 2 . 00 ? .00 ? .01 . 07 . 09? Step 3 . 00 . 00 ? .00 . 07? .05? ? . 48 Step 4 . 00 ? .00 ? .00 . 06? .05 ? .46 ? . 12 .02 .02 .02 .45 Note Sup-rated coworker con? ict = supervisor-rated coworker relationship con? ict, LMX = leadermember exchange. N = 121. ? p b . 05. p b . 01. 018 K. J. Harris et al. / The Leadership Qua rterly 22 (2011) 10101023 Table 5 Hierarchical linear modeling results predicting abusive supervision in Sample 2. Step 1 Control variables Age Job tenure Organizational tenure Relationship tenure Independent variable Sup-rated coworker con? ict (A) Moderator LMX (B) Interaction term A? B ? R2 . 00 . 00 ? .00 ? .00 Step 2 . 01 .00 ? .00 ? .00 . 11? Step 3 ? .00 . 00 ? .00 . 00 . 09? ? . 54 Step 4 . 00 . 00 ? .00 . 00 . 13? ? . 55 ? . 29 .05 .01 .01 .35 Note Sup-rated coworker con? ict = supervisor-rated coworker relationship con? ct, LMX = leadermember exchange. N = 134. ? p b . 05. p b . 01. Next we turn to the HLM results presented for Sample 2 in Table 7. This table shows that supervisor-reported coworker relationship con? ict was signi? cantly related to abusive supervision in step 1b (which passes Baron and Kennys (1986) ? rst step) and OCB (in step 2b), but not work effort (in step 3b). These results pass the ? rst two steps for mediation for OCB, but not work effort. Table 7 also reveals that abusive supervision is negatively and signi? cantly related to OCB (? = ?. 26, p b . 05) in step 2c, and signi? antly and negatively related to work effort (? = ?. 39, p b . 01) in step 3c. Thus, Hypothesis 3, which was supported in Sample 1, is also supported in Sample 2. Step 2c shows that when both supervisor reports of coworker relationship con? ict and abusive supervision are entered into the equation, the coworker relationship con? ict variable is no longer a signi? cant predictor of OCB. In terms of the mediation results, the results from Baron and Kennys (1986) three-step procedure show that abusive supervision mediated the relationship between supervisor-rated coworker relationship con? ct and OCB, but not work effort. Thus, Hypothesis 4, which was supported for both dependent variables in Sample 1, was only supported for OCB in Sample 2. 5. Discussion The purpose of this study was to further our knowledge of the predictors and outcomes of abusive supervi sion. We pursued this goal by examining supervisor reports of relationship con? ict with their coworkers as a predictor of subordinate-rated abusive supervision, and LMX quality as a situational variable in? uencing this relationship. Additionally, we examined the outcomes of supervisor-rated OCB nd work effort and found that abusive supervision fully mediated the relationships between supervisor reports of coworker relationship con? ict and OCB in both samples and the outcomes of work effort in one sample. Returning to our theoretical arguments, we found that displaced aggression and LMX theories provide useful lenses for discussing predictors and outcomes of abusive supervision. Coworker relationship con? ict at any level is a potent source of stress and frustration as it impedes the achievement of goals and the attainment of desired outcomes (e. g. , Thomas, 1976).Like past abusive supervision research (Tepper, Duffy, Henle & Lambert, 2006), our results suggest that some supervis ors will resort to abusive behaviors against their employees as a means of coping with these consequences. This study advances existing research by explicitly examining situations where subordinates are not the logical target of retaliation (i. e. , they are not the source of the con? ict). Because subordinates are an easy and accessible target, however, having less power and less of an ability to retaliate, they make relatively safe candidates for abuse from frustrated supervisors.Table 6 Hierarchical linear modeling mediation results in Sample 1. DV = abusive supervision Step 1a Age Job tenure Organizational tenure Relationship tenure Supervisor-rated coworker relationship con? ict Abusive supervision Note OCB = organizational citizenship behaviors. N = 121. ? p b . 05. p b . 01. .00 . 00 ? .01 . 08? Step 1b . 00 ? .00 ? .01 . 07 . 09? Step 2a . 00 ? .02 . 00 . 05 DV = OCB DV = work effort Step 2b . 00 ? .01 ? .00 . 05 ? .08+ Step 2c . 00 ? .01 ? .00 . 07 ? .06 ? .27? Step 3a ? . 00 ? .02 . 02 . 00 Step 2b ? .00 ? .01 . 02 . 01 ? .14? Step 3c . 0 ? .01 . 01 . 04 . 11 ? .37 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 Table 7 Hierarchical linear modeling mediation results in Sample 2. DV = abusive supervision Step 1a Age Job tenure Organizational tenure Relationship tenure Supervisor-rated coworker relationship con? ict Abusive supervision Note OCB = organizational citizenship behaviors. N = 134. ? p b . 05. p b . 01. .00 . 00 ? .00 ? .00 Step 1b . 01 . 00 ? .00 ? .00 . 11? Step 2a ? .01 ? .00 . 00 . 00 DV = OCB DV = work effort 1019 Step 2b ? .01 ? .00 . 00 . 00 ? .13? Step 2c ? .01 . 0 ? .00 . 00 ? .09 ? .26? Step 3a ? .00 ? .00 . 00 ? .00 Step 3b ? .00 ? .00 . 00 ? .00 ? .03 Step 3c . 00 ? .00 . 00 ? .00 . 02 ? .39 Additionally, when supervisors experience coworker relationship con? ict, our results indicate that they are most likely to abuse subordinates with whom they have low quality LMX relationships. This ? nding appears to suppo rt our argument that supervisors will focus their abusive behaviors on those employees in low quality exchanges in order to shield their high quality relationships from the deleterious effects of abusive supervision.In this way, supervisors may reason that abusive behaviors allow them to vent frustration while minimizing the negative in? uence of this coping behavior on their most valued employees. Naturally, there are ? aws in this method of coping, most notably that the performance levels of abused employees will likely suffer, causing added strain and frustration for other employees and the supervisors themselves. Among supervisors who make the worryatic choice to cope through abuse, however, it appears that employees in low-quality relationships are the most likely targets.We also extended abusive supervision research with our ? ndings indicating that this variable is related to the outcomes of OCB and work effort. These ? ndings are noteworthy as they extend the nomological n etwork of outcomes related to abusive supervision, and because both outcomes were supervisor-rated, which helps to minimize common source bias concerns (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). Additionally, in sample 1 we found that abusive supervision served as an intermediary mechanism explaining the relationships between supervisor reports of coworker relationship con? ct and both consequences examined, and that there was also mediation on the outcome of OCB in sample 2. These results are important as they begin to answer the questions related to how situational supervisor variables, such as coworker relationship con? ict, ultimately are translated into subordinate outcomes. Surprisingly, we did not ? nd support for the work effort mediation hypothesis in Sample 2. A post hoc explanation for these insigni? cant ? ndings may relate to the demographic composition of the samples. Sample 2 was different from Sample 1 for both subordinates and supervisors.It was primarily male, the average age was higher, and average job and organizational tenure were both more than double (except for supervisor job tenure) those in the ? rst sample. Although it is possible to deduce explanations as to how these differences might have in? uenced our results, such atheoretical logic would be overly speculative. Thus, as we suggest below, we encourage replicative research in additional samples that would allow for a more systematic assessment of these, or other, sample-speci? c characteristics. 5. 1. Contributions These ? dings make several contributions to the extant research on abusive supervision and LMX relationships. First, they build support for the notion of displaced abusive supervision and undermine a potential alternative explanation. In Teppers (2007) review of abusive supervision literature, he concluded that supervisors perceptions of organization-level factors, such as Fig. 2. Moderating effect of LMX on the relationship between supervisor-rated coworker relat ionship con? ict and abusive supervision in Sample 1. 1020 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023Fig. 3. Moderating effect of LMX on the relationship between supervisor-rated coworker relationship con? ict and abusive supervision in Sample 2. injustice and contract violation, can trigger abuse toward individual targets (i. e. , subordinates). He argued that this phenomenon might be explained by displaced aggression logic, in that subordinates serve as safe abuse targets even if the abuse is unlikely to resolve the perceptions triggering the desire to be abusive. An alternative, although somewhat tenuous, explanation is that these negative perceptions in? ence animosity toward the overall organization and that supervisors justify the abuse of subordinates who are seen as complicit in the perceived negative aspects of the organization. Our ? ndings suggest that this alternative basis of justi? cation would not adequately explain displaced abusive supervision . Looking beyond generalized organizational perceptions, we found that even frustration stemming from speci? c, identi? able non-subordinate sources (i. e. , supervisors coworkers) might translate into abuse toward subordinates.This suggests that abusive supervision may serve as a self-defeating coping mechanism (e. g. , Baumeister & Scher, 1988), akin to mechanisms such as difficulty drinking and procrastination, in that it seeks short-term stress-reduction (e. g. , through emotional venting) in a harmful way that does not address the true source of the underlying problem (e. g. , con? ict with peers). We also expand on Teppers conclusion, again stemming from his 2007 review of abusive supervision research, that subordinate characteristics in? uence the likelihood that they will experience abuse.As in the present study, Tepper (2007) cited victimization research to argue that subordinates who appear overly provocative or passive put themselves at a heightened risk for abuse. Expan ding on the latter idea, we argued and spy that employees in low quality LMX relationships, who we expect demonstrate relatively high levels of passivity and vulnerability, report higher levels of abuse. This suggests that instead of identifying each of the potential subordinate characteristics that can incite abuse, a more parsimonious approach might be to look at broad relationship variables such as LMX that can be viewed as re? cting the aggregate impact of these individual characteristics. This conclusion also adds to LMX research by revealing an additional consequence of low-quality LMX relationships. In addition to the wide body of research showing that low-quality LMX subordinates experience outcomes such as fewer rewards, lower resource levels, and reduced job satisfaction (e. g. , Liden, Sparrowe & Wayne, 1997), this study suggests a more serious potential consequence in the form of victimization by abusive supervisors.Additionally, our results, and the fact that most were replicated across the two samples, demonstrate the utility of multi-level models for predicting employee consequences of abusive supervision. Abusive supervision is an inherently multi-level phenomenon and this study shows that insights into some causes of abuse, such as con? ict levels between supervisors, exist that cannot be assessed from subordinate self-reports. Similarly, it identi? es supervisor-rated subordinate outcomes of abusive supervision (effort levels and OCB) that are dif? cult to assess with self-reports due to social desirability and common source bias concerns.Further, these supervisor-rated effects provide some indication that abusive supervisors are at least indirectly aware of the selfdefeating consequences of abuse. Our data do not tell us whether supervisors consciously related their abuse to lower levels of employee effort and citizenship behavior. Their awareness of lower levels among the abused subordinates, however, suggests that a degree of denial would be needed for the supervisors to overlook these causeeffect relationships. Although existing research has not, to our knowledge, explicitly verbalize that supervisors are unaware of the consequences of abusive behavior, this ? ding suggests that in store(predicate) research on preventing abuse might bene? t from focusing not on why supervisors view the behavior as acceptable, but why they engage in it despite an apparent awareness of these consequences. 5. 2. Limitations In addition to the aforementioned strengths and contributions, there are limitations that we must acknowledge to properly interpret the studys results. First we acknowledge that the theoretical framework we have developed is not the only logical explanation for the hypothesized and observed relationships.For example, it is glib that the link between supervisors coworker relationship con? ict and abusive supervision is less cognitive than we have argued. Instead of selectively choosing subordinates as a low-risk target for venting frustration, it might be that some supervisors simply possess traits that predispose K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 1021 them toward con? ict and abusive behaviors (with higher levels of abuse directed at low quality members). Examples of such traits might include negative affectivity or hostile attribution styles (Douglas & Martinko, 2001).An investigation of these possibilities would be useful in forming a more comprehensive understanding of the empirical relationships observed in the present study. In terms of methodological limitations, survey length constraints required us to use a reduced version of the abusive supervision scale. Even though we chose items that tapped into the full set of behaviors and found an extremely high correlation between our shortened measure and the full scale, this may still be viewed as a limitation. other limitation is that we were unable to measure causality.Thus, there is the potential that our relationships actually have turnaround time causality or that variables predict each other in a recursive manner. This is particularly true regarding the association between LMX perceptions and abusive supervision. Our results suggest that supervisors are more abusive toward some employees than others and that this difference is associated with variations in subordinates LMX scores. It can be argued, and is indeed very likely, that an abused employee would report lower LMX scores because of the abuse.The ? nding that supervisors are selective in their abuse targets suggests that some criterion is evaluated before targets are chosen and we have argued that preexisting LMX relationship qualities could serve as this criterion. Our design does not allow us to make this claim de? nitively, however. Similarly, it may be that abusive supervision is not the predictor of work effort, but that insuf? cient effort by subordinates promotes higher levels of abusive supervision or that both variables in? uence each other in a cyclical manner.We are particularly sensitive to the argument that there may be a feedback loop between abusive supervision and the outcome variables, such that abuse reduces subordinates effort and citizenship levels, and this reduction provokes further abuse, although the design of the study did not allow us to test this possibility. Along a similar line, it could be that abusive supervision toward subordinates is actually the cause of the supervisors con? ict among peers. We hope that future studies will be designed to better answer these causality questions.There are also limitations associated with the sampling of public, white-collar organizations. Different organizations (e. g. , private, military, blue-collar) have different rules and norms governing behavior and it is likely that the abusive supervisory behaviors studied would be more or less permissible, and therefore more or less common, in different organizational settings. 5. 3. Direc tions for future research This studys ? ndings suggest a number of directions for future research. First, we hope future researchers will examine our hypotheses in other, more diverse samples.Although we examined two separate organizations, it is necessary to examine additional samples to better establish the generalizability or boundary conditions of our relationships. A second suggestion is to examine the relationships in this study with a longitudinal research design. The extant research on abusive supervision, including this study, has primarily relied on cross-sectional designs. Although telling, these studies leave out situations and behaviors that impact subordinates over time. In the case of both supervisor reports of coworker relationship con? ct and abusive supervision, it may be that supervisors and subordinates learn to cope with these situations, and become accustomed to them. Conversely, it could be that these situations and behaviors become worse as they store over t ime (Harris, Kacmar, & Witt, 2005) as argued by Tepper (2000) and as noted in our discussion of cyclical relationships between abuse and behavioral outcomes in the previous section. Another avenue for future research is to conduct additional multi-level investigations to determine how supervisor experiences and situations impact their subordinates.In this study we examined supervisor reports of coworker relationship con? ict, but it also would be interesting to investigate the effect of supervisors supervisor relationship con? ict, abusive supervision, LMX, team member exchange, and perceived organizational support (Erdogan & Enders, 2007 Tangirala, Green, & Ramanujam, 2007) as these variables are likely to have trickle-down effects on employee outcomes. Additionally, the aforementioned implication that supervisors might be aware of the consequences of abusive supervision suggests that a multilevel, or at least supervisor-level, focus on understanding the justi? ation process might provide insight into interventions for preventing such behavior. It would also be interesting to investigate personality characteristics, such as Machiavellianism, entitlement, and narcissism, of supervisors and subordinates and how these variables are related to abuse (Harvey & Harris, 2010 Kiazad, Restubog, Zagenczyk, Kiewitz, & Tang, 2010). Finally, we examined LMX from the perspective of the member, but it would be insightful to investigate leader reports of the LMX quality with their subordinates and how this rating interacts with supervisor coworker con? ict. 5. 4. Practical implications Before discussing speci? practicable implications from this study, it should be noted that the overarching implication from this and most of the existing body of research on abusive supervision is that abusive supervision is detrimental to all parties. It is trying for victims and hurts organizational performance and a supervisors effectiveness by negatively affecting desirable outcomes (see Tepper, 2007) such as increased levels of effort and OCB. Employees may feel affright and afraid to report the behavior of abusive supervisors, however, making it dif? cult for organizational leaders to identify and eliminate these abusive managers.Because of the dif? culty in reducing existing levels of abuse, handicap techniques for reducing the likelihood of abusive supervision are advisable. The results of this study suggest that one such technique is for organizational leaders to observe and mediate con? icts between supervisory employees, thereby removing an antecedent of abusive behaviors. Additionally, because the supervisors in our study were more likely to abuse employees with whom they shared low-quality relationships, an organization-wide focus on the development of strong leadermember relationships might foster a climate where there are few 022 K. J. Harris et al. / The Leadership Quarterly 22 (2011) 10101023 desirable targets for abuse. We acknowledge that neither o f these suggestions (i. e. , mediating supervisor con? icts and promoting strong leadermember relationships) are simple tasks. We suggest, however, that a continuous focus on these goals would consume far less time and energy than dealing with the consequences of abusive supervision. 6