When Core Self-Evaluations Influence Employees’ Deviant Reactions to Abusive Supervision: The Moderating Role of Cognitive Ability

Viewing workplace deviance within a victim precipitation framework, we explore how abusive supervisors target subordinates low in core self-evaluations (CSE) to explain when such employees respond by engaging in workplace deviance. We theorize that employees who are lower in CSE receive more abusive supervision, which generates subsequent harmful reactions toward supervisors, peers, and the organization. This occurs primarily when employees lack sufficient cognitive resources in dealing with supervisor abuse. We test, replicate, and extend our theoretical model in three empirical studies. Results demonstrate that lower employee CSE drew more abusive supervision and led low-CSE employees to exhibit workplace deviance. This abusive supervision mediation effect was stronger for employees with comparatively lower cognitive ability levels. The findings are discussed with regard to theoretical and ethical issues in confronting employee abuse.

actions that violate organization norms and threaten the well-being of its members (Robinson and Bennett 1995) and comprises counterproductive behaviors that may be directed at coworkers, supervisors, and the organization. In general, researchers have focused more on contextual elements (e.g., work climate, supervision) surrounding employees in attempts to predict workplace deviance than on characteristics of the employees themselves (Wang et al. 2015). This is surprising in that some theories (e.g., victim precipitation theory; Aquino and Lamertz 2004) suggest employees can become enmeshed in relational dynamics leading to workplace deviance. Interestingly, recent research supports a connection between CSE and deviant behavior suggesting that individuals lower on CSE exhibit undesirable behavior at work (Chang et al. 2012). Fundamental to self-identity, core self-evaluations are "… basic or bottom-line evaluations that individuals hold about themselves" (Judge and Bono 2001, p. 80). For these reasons, we believe CSE could be a key antecedent of workplace deviance. However, evidence of an association between CSE and deviance is relatively unexplored, and thus little is known about how and when CSE leads to workplace deviance (Chen 2012).
Researchers have often framed employee deviance as a "hot" affective response to negative work experiences. According to victim precipitation theory (Aquino 2000;Aquino and Lamertz 2004), especially vulnerable subordinates would be those who act as "submissive victims," harboring negative views of themselves and their situations (Olweus 1978(Olweus , 1993. Such a profile would be typical of individuals with low self-assessments of their worthiness and competence, that is, individuals possessing low core self-evaluations (CSE; Judge et al. 2003). Thus, one focus of our research effort was to examine whether CSE would be a "vulnerability marker" (cf. Aquino and Thau 2009) such that subordinates' with lower CSE would more likely be targeted for mistreatment, in turn resulting in workplace deviance. Abusive supervision is one such form of mistreatment associated with organizational ethics and morality (Wang et al. 2015). It includes destructive behaviors of supervisors toward their subordinates, including irritable outbursts, public ridicule, and scapegoating (Tepper 2007). Interactions with leaders erode when they hold negative views of subordinates, establishing abusive supervision as potential precursor to their counterproductive work behavior (Tepper et al. 2009).
Researchers have acknowledged that deviance can stem from the dynamics present within tense dyadic relationships (Aquino and Thau 2009;Hershcovis and Barling 2007). To date, however, researchers have not examined abusive supervisors as the key relational link potentially explaining why low-CSE subordinates might exhibit more deviance than their high-CSE counterparts. From a victim precipitation theory perspective, an abusive supervisor could be viewed as an interpersonal catalyst for subordinates' deviant reactions. Examining connections between CSE, abusive supervision, and employee deviance simultaneously could contribute toward a more integrated understanding of antecedents and consequences of abusive supervision. We contend lower-CSE subordinates will experience supervisory abuse, leading to greater workplace deviance (cf. Wei and Si 2013). Therefore, a key focus in the present paper is on abusive supervision as a mediating mechanism through which CSE leads to workplace deviance.
It is well known that abused employees tend to react negatively (Mackey et al. 2017). However, a line of inquiry has emerged that retaliatory deviant behavior can also be influenced by cognition, planning, and professional strategies (e.g., Fox and Spector 2010). Aggrieved employees may weigh both the abuser's intent and the viability of retaliation before taking any action (Anderson and Bushman 2002). Thus, even when low-CSE subordinates experience abusive supervision and consider countering with deviant behavior, thoughtful consideration might militate against such action. We suggest that individuals with greater cognitive ability resources are more likely to recognize the risks of intemperate reactions to abusive supervision. Cognitive ability entails attentional and mental skills like insight, anticipation, problem-solving, and inhibition (cf. Kane et al. 2007;Miyake and Friedman 2012) that should influence how employees process abusive supervision. Hence, higher cognitive ability (CA) may afford subordinates greater prescience about responding to supervisor abuse and might attenuate deviant workplace reactions.
The present research makes two primary contributions to the literature. First, the interactional dynamics inherent in victim precipitation theory suggests that abusive supervision could very well be a critical link in examining the connection between CSE and deviant behavior (Judge and Kammeyer-Mueller 2011). Unfortunately, the approach used to study relations between CSE, abusive supervision, and employee deviance can be characterized as piecemeal. That is, researchers have examined associations between CSE and abusive supervision (rarely), CSE and deviance (rarely), and abusive supervision and deviance (often). However, no studies of which we are aware have linked these three constructs together. Victim precipitation theory recognizes that victims (i.e., low-CSE individuals) and perpetrators (i.e., abusive supervisors) are intertwined parts of victimization. By expanding the theoretical scope of our investigation to linkages between all three constructs, we sought an integrated understanding of how CSE may influence workplace deviance. As such, we add to emerging literatures investigating antecedents and outcomes of abusive supervision (Kiewitz et al. 2012).
A second contribution of the present research is our consideration of the moderating role of CA on relations between abusive supervision (precipitated by low CSE) and workplace deviance. Deviant reactions to abusive supervision by subordinates are risky (Thau and Mitchell 2010). Those with greater CA may be more likely to think through potential negative repercussions of such reactions. Thus, our work stands to enhance our understanding of when abusive supervision will manifest as workplace deviance. Again, no studies of which we are aware have explored cognitive ability as a boundary condition through which abusive supervision leads to workplace deviance. Figure 1 displays the theoretical model to be investigated by our research efforts.

CSE, Abusive Supervision, and Workplace Deviance: A Victim Precipitation Perspective
A key claim of victim precipitation theory (Aquino and Lamertz 2004) is that dyadic relationships between a dominating perpetrator and a submissive victim will lead to high levels of institutionalized victimization. The intuitive combination of dominating perpetrator and submissive victim likely leads to institutionalized victimization because the roles are mutually reinforcing (Aquino and Lamertz 2004). Institutionalized victimization is a perpetrator-victim interaction pattern where both parties have become committed to and make sense of a relationship through a shared history of reciprocal victim-perpetrator enactment (Ashforth and Humphrey 1997). A dominating perpetrator desires to control or exploit others and uses coercive supervision, authority and status differences, arbitrary actions, and punitive deterrence of subordinates' initiative and dissent (Ashforth 1997). Such behavior is closely tied to abusive supervision.
Submissive employees are people who are low in selfesteem and assertiveness and high in social anxiety, nervousness, and insecurity, and are more likely to be taken advantage of or treated disrespectfully (Aquino 2000). They are insecure and less responsive when interacting with others, potentially indicating a lack of self-confidence and detachment from peers. Such employees can be susceptible to harassment from others (Olweus 1978(Olweus , 1993). Those with a negative self-image and high in neuroticism are bullying targets because a self-deprecating identity is associated with low social position in the peer group (Aquino and Lamertz 2004). This low position signals to the domineering perpetrator that the person is highly vulnerable to attack or exploitation.
CSE is a trait that fits well with the submissive victim role. It is a higher-order construct comprising four core traits: self-esteem (i.e., sense of self-worth), locus of control (i.e., perceived control of events in one's life), emotional stability (i.e., feeling calm and secure), and generalized selfefficacy (i.e., self-ascribed capacity to meet personal challenges). CSE describes fundamental premises that individuals hold about themselves and their functioning in the world (Judge and Bono 2001). Whereas higher CSE individuals see themselves as capable and in control, lower-CSE individuals are prone to worry and feel less capable of solving problems or controlling what happens to them. Individuals with higher CSE levels exhibit positive work attitudes (e.g., satisfaction) and behaviors (e.g., task performance- Bono and Judge 2003), whereas lower-CSE individuals are less likely to do so. CSE has relational implications in work contexts that have been overlooked, in part because its effects are manifested less directly. For example, it has been suggested that lower-CSE individuals are less protective of their social reputations  and therefore should be less prone to interact in ways that maintain positive impressions others have of them. As Judge and Kammeyer-Mueller (2011, p. 338) note, "… individuals with positive [negative] core self-evaluations should be more [less] effective in an interpersonal sense." In addition, CSEs can influence job satisfaction and performance (Judge and Bono 2001) are observed by others (Scott and Judge 2009), and influence situational specific appraisals and related behaviors (Rode et al. 2012).
We believe those with a negative self-view associated with low CSE are more likely to elicit abuse from their supervisors. Low-CSE followers might become targets of supervisor abuse if they require constant reassurance and guidance, or hesitate to communicate their needs (see Egan and Perry 1998;Tepper et al. 2011). Other researchers have observed that nervous, apprehensive behavior may be viewed as bothersome and elicit supervisory abuse in turn (Kamer and Annen 2010;Milam et al. 2009). Although low-CSE employees may be undeserving of abuse, their behavior can cause supervisors to direct abuse toward them (cf. Henle and Gross 2014). All in all, within leader-follower relationships, abusive supervisors are more likely to lash out at submissive low-CSE followers (Aquino and Thau 2009).
Repeated exposure to victimization can trigger a pattern of retaliatory responses from the victim culminating in workplace deviance (Andersson and Pearson 1999). Victim precipitation theory suggests when individuals experience supervisor abuse, they may reciprocate in degree if not in kind (Cropanzano and Mitchell 2005). Simply put, poor treatment from supervisors creates an imbalance followers redress through negative or unethical actions ). Further, employees who are routinely victimized by others experience high levels of fear and anxiety that negatively impact performance and motivation (Janoff-Bulman 1992). As those low in CSE are more reactive to stressful stimuli , workplace deviance serves as a dysfunctional coping mechanism germane to those low in CSE. As such, low-CSE individuals who experience abuse are likely to respond by engaging in workplace deviance. When individuals experience harmful treatment, they may reciprocate with harmful actions not only toward supervisors, but other organizational members or the organization itself (e.g., Penney and Spector 2005). Displaced aggression theory (i.e., the frustration-aggression hypothesis; Dollard et al. 1939;Marcus-Newhall et al. 2000) has been used to explain behavioral outcomes resulting from supervisory mistreatment (e.g., Mitchell and Ambrose 2012). This theory proposes that employees engaging in deviant acts may not aggress only against the source of their frustration (e.g., supervisors). Because of social norms or fear of retaliation, they may aggress against convenient targets (e.g., coworkers and the organization) that are less likely to retaliate than supervisors (Mackey et al. 2016(Mackey et al. , 2017. Thus, we argue employees are likely to engage in negative behaviors toward individuals, the organization, as well as toward their supervisor . The positive association between abusive supervision and subsequent employee deviance has been documented (e.g., Hershcovis and Barling 2010;Martinko et al. 2013). Thus, we expect individuals with low levels of CSE who experience more abusive supervision to engage in a range of deviant behaviors.
Hypothesis 1 Core self-evaluation will indirectly affect workplace deviance through abusive supervision.

The Moderating Effect of Follower Cognitive Ability
We believe employees' CA may play a role in mitigating their harmful behavioral responses to abusive supervision suggested by victim precipitation theory. CA reflects general cognitive resources and encompasses several specific abilities (e.g., verbal, quantitative, spatial) critical to regular social functioning. It is associated with employee wellbeing (Judge et al. 2010) and job performance (Schmidt and Hunter 2004). Given its relevance for expected and desired employee behaviors, we offer two complementary reasons that differences in followers' CA are likely to moderate the effect of abusive supervision on deviance.
First, Dilchert et al. (2007) suggest CA tempers the expression of deviant work behaviors in that individuals with greater CA are more likely to think through potential negative repercussions. Deviant reactions to abusive supervision are risky and potentially costly for individuals (Thau and Mitchell 2010). Followers with more cognitive resources should more carefully weigh the possible consequences of retaliatory behavior (Morgan and Lilienfeld 2000) in response to abusive supervision. In contrast, individuals with lower CA may demonstrate less foresight about the consequences of retaliating to abusive supervision with deviant actions. As such, those lower in CA might not fully weigh the potential repercussions associated with deviant workplace responses to abusive supervision (e.g., termination; Dilchert et al. 2007). This renders them more likely to respond with deviance when subjected to supervisor abuse in comparison to colleagues with higher CA levels.
A second reason CA should attenuate deviant reactions to abusive supervision stems from research suggesting effortful cognitive functioning helps resolve incompatible reactions to experienced hostility. This might occur when abused employees balance conflicting desires to retaliate with desires to maintain normal reporting relationships and employment (Botvinick et al. 2001). Greater CA enables individuals to manage retaliatory excesses in thoughts and behaviors . As it also facilitates processing behavioral expectations (Thau and Mitchell 2010), individuals with greater CA should better understand normative workplace demands and the impropriety of vengeful thinking . As such, we extend victim precipitation theory by proposing that the effect of abusive supervision on workplace deviance would be weaker among individuals possessing greater CA and stronger among those with lower CA.
Hypothesis 2 Cognitive ability will moderate the relationship between abusive supervision and workplace deviance such that the relationship is stronger for employees with lower cognitive ability levels.
Taken together, we argue that the indirect effect of CSE to workplace deviance through abusive supervision depends on subordinates' CA level. As shown in Fig. 1, our expectation corresponds to a form of moderated mediation (Edwards and Lambert 2007) in which a mediated effect varies over levels of a moderator operating at the second stage of the mediated relationship. It is important to examine CSE's indirect effect in such a manner because our theoretical framework proposes a specific form of moderation with respect to linkages between all CSE, abusive supervision, and employee deviance. Piecemeal tests of relations among our focal constructs have been more typical in past research. With our integrative approach, a better understanding how and when CSE may influence workplace deviance may be possible.

Hypothesis 3
Cognitive ability will moderate the indirect effect of CSE on workplace deviance via abusive supervision such that the mediated effect will be stronger for employees with lower cognitive ability levels.

Overview of the Studies
We tested our hypotheses across three studies with varied design features. Study 1 established the indirect effects of CSE on deviance through abusive supervision, using multiple data sources (self and coworker) for CSE to minimize common method variance concerns associated with CSE self-reports. Although facets of employees' CSE (e.g., external locus of control, low generalized self-efficacy) have been connected with problematic supervisory relations indicative of abusiveness, and abusive supervision has been associated with workplace deviance outcomes (e.g., see Hershcovis and Barling 2010), sequential links among the three constructs in this substantive chain have not been examined simultaneously in prior research. Thus, this study provides an incremental contribution to the literature by examining the theoretical linkages proposed in Hypothesis 1.
Study 2 tested all three hypotheses in a US sample using a cross-lagged design. Although personality constructs within the abusive supervision and workplace deviance literatures have been considered, CA's potential as a moderator of this chain of events is novel. We examined whether elevated levels of this cognitive employee resource may attenuate negative reactions associated with supervisory abuse, influencing how this mediated sequential chain unfolds.
With Study 3, we conducted a constructive replication (Lykken 1968) of Study 2 using multisource cross-lagged data and two distinct convergent triangulation approaches (Turner et al. 2015). First, although there are widely established measures for CSE and abusive supervision, there are several commonly used measures of cognitive ability and workplace deviance. As such, we used different measures of CA and workplace deviance to demonstrate our proposed relationships are not restricted to particular measures of these constructs. Second, we engaged in convergent triangulation in a different cultural setting by collecting data in the country of Romania. On Hofstede's (2001) five cultural factors, Romania scores lower than the USA in individualism and masculinity but substantially higher in power distance, uncertainty avoidance, and long-term orientation. As such, Study 3 also examined whether our theoretical model may generalize to non-US work settings. Finally, Study 3 extended Study 2 in two ways: by using full-time employees, and by including supervisor-directed deviance.

Mediation (indirect effects)
To test Hypothesis 1, we used the recommended bootstrap approach in testing indirect effects (MacKinnon et al. 2007;Shrout and Bolger 2002). We first estimated the mediated effects using a nonlinear regression module to estimate the associated coefficients from 10,000 bootstrap samples, and we examined bias-corrected and accelerated (BCa) confidence intervals to determine the indirect effects.
Moderation (interaction effects) To test Hypothesis 2, we performed stepwise hierarchical regression analyses. Control variables were entered first, and main effect variables were entered second, followed by the interaction term in the last step. Following Aiken and West (1991), we plotted significant interaction effects, using plus or minus one standard deviation on the independent and moderator variables as benchmarks. To illustrate the results, we plotted the simple effects at high (+ 1 SD) and low (− 1 SD) levels of the hypothesized moderator variable and conducted simple slope analyses based on procedures described by Edwards and Lambert (2007).
Moderated mediation (conditional indirect effects) To test Hypothesis 3, we followed Edwards and Lambert's (2007) procedures to compute conditional indirect effects with reduced form equations. We used a constrained nonlinear regression module to estimate the associated coefficients from 10,000 bootstrap samples, and we examined bias-corrected and accelerated (BCa) confidence intervals to determine whether the indirect effects differed at select values of the hypothesized moderator variable.

Study 1: US Multisource Sample
The few studies that link CSE and abusive supervision have relied exclusively on self-report employee data (e.g., Neves 2014; Wu and Hu 2009). With such data, it is difficult to determine whether results are impacted by common method bias or whether those low in CSE actually experience more abusive supervision or simply perceive more due to their overall negative outlook. To offer clearer support for our arguments that low-CSE individuals are actually subject to greater supervisory abuse, we assessed their CSE and abusive supervision using data collected from another source. Fortunately, research shows that other reports can be employed to reliably assess manifestations of personality (Connelly and Hülsheger 2012;Milam et al. 2009) and workplace mistreatment (Carpenter et al. 2017) such as abusive supervision.
Self-and other ratings of CSE and abusive supervision may not capture precisely the same substance, but they are complementary sources that provide valid coinciding information (e.g., Funder et al. 1995). There is a precedent for using other reports of CSE (e.g., Scott and Judge 2009) and workplace mistreatment (Kluemper et al. 2015). Indeed, much of the organizational behavior literature rests on the notion that the perceived environment influences behavior more strongly than do its objective characteristics (Lewin 1951). We are hopeful this study provides insight into the process through which CSE leads to deviance through abusive supervision, as well as to minimize potential concerns regarding the alternative explanations detailed above.

Participants and Procedures
The sample consisted of full-time working adults. Students from a large university in the southeastern USA were given extra course credit for providing contact information for independent employee-coworker dyads that expressed a willingness to participate in our study. We first emailed employee-coworker dyads a link to an online survey, with an accompanying message containing the name of the employee to be rated (self or coworker). They were assured verbally and in writing of the confidentiality of their responses, and informed their participation was voluntary and could be stopped at any time without penalty. Within each dyad, ratings of the employee's CSE, experiences of abusive supervision, and deviant behavior were collected from both the employee and the coworker. A total of 197 employees (75% response rate) and 158 of their coworkers (60% response rate) responded to our surveys. After matching employees and coworkers with usable data, our subsequent analyses were based on 121 employee-coworker pairs. Participants were working in a wide range of organizations and industries (e.g., retail, engineering, telecommunications, healthcare, financial institutions) and spanned a wide range of job types (e.g., accountant, secretary, sales rep, customer service rep, manager). Sixty percent of employees were female. They averaged 33 years of age (SD = 12.6), worked on average 38 h per week (SD = 11.3), and had spent 3 years (SD = 4.2) working with their supervisor. Fifty-six percent of coworkers were female. They averaged 33 years of age (SD = 11.9), worked on average 38 h per week (SD = 9.6), and had 5 years of prior work experience (SD = 5.3).

Measures
Core self-evaluation CSE was gauged using Judge et al.'s (2003) 12-item measure. Respondents indicated the extent of agreement (1 = strongly disagree; 5 = strongly agree) with each item. Sample items include "I am confident I get the success I deserve in life" and "Sometimes when I fail I feel worthless" (reverse-scored). Subordinates participating in the study self-reported CSE, while a coworker also rated the CSE of these subordinates (e.g., "They are confident they get the success they deserve in life").
Abusive supervision We used Tepper's (2000) 15-item measure to assess the degree that employees experienced abuse from their supervisors. Consistent with previous studies (e.g., Thau and Mitchell 2010), respondents indicated the extent of agreement with each statement (1 = strongly disagree; 5 = strongly agree). Sample items include "The boss ridicules them" for coworker ratings and "My boss tells me my thoughts or feelings are stupid" for self-ratings.
Subordinate deviance Interpersonal and organization deviance were assessed with Bennett and Robinson's (2000) full measures. Supervisor-directed deviance was measured with three items from Skarlicki and Folger's (1997) measure (see Tepper et al. 2009). Subordinates and coworkers reported the frequency (1 = never; 5 = every day) with which subordinates had engaged in each behavior. Sample items include "Made fun of someone at work" (interpersonal deviance), "Came in late to work without permission" (organization deviance), and "Talked back to the boss" (supervisor-directed deviance). We measured subordinate deviance using an aggregate of subordinate and coworker ratings.
Controls We controlled for demographic variables (gender, age, and dyad tenure) that have been shown to influence the likelihood of mistreatment (e.g., Aryee et al. 2007). In addition, certain personality variables, specifically conscientiousness, agreeableness, and extraversion, may also affect employees' perceptions of and reactions to abuse (Coyne et al. 2000) and their tendencies to manifest deviance (Berry et al. 2007). Therefore, we measured these personality traits using the mini-IPIP (Donnellan et al. 2006). Respondents reported their agreement (1 = strongly disagree; 5 = strongly agree) with four items for each trait.

Results
Descriptive statistics and intercorrelations for Study 1 variables are shown in Table 1. We first examined certain intercorrelations to check on the possibility that low-CSE employees may simply perceive abusive supervision due to their negative outlook rather than actually experience greater abuse. We did this by including coworker ratings of the abuse an employee received from a supervisor. Coworker ratings were correlated .54 with self-ratings of abusive supervision. As shown in Table 1, coworker ratings of abusive supervision correlate significantly with self (− .22, p < .05) ratings of CSE, as well as self-ratings of interpersonal (.38, p < .01), organizational (.41, p < .01), and supervisor-directed (.47, p < .01) deviance. Together, these results suggest that employees who are low in CSE are likely to actually experience more supervisor abuse than their counterparts, and those victims of supervisor abuse (rather than simply perceiving themselves as being abused) likewise engage in more workplace deviance.
We conducted confirmatory factor analysis to evaluate the discriminant validity of the Study 1 variables (see Table 2). We created three parcels using the single-factor method for self-and coworker-rated CSE and abusive supervision and used the content method for interpersonal, organization, and supervisor-directed deviance (for a review of parceling approaches, see Landis et al. 2000). The full model demonstrated acceptable fit to the data (χ 2 = 223, df = 120, Table 1 Study 1: descriptive statistics and intercorrelations N = 121. Gender is coded 0 = male; 1 = female. Variables 1-5 and 11-13 were self-rated by employees (SR); variables 6-10 were rated by coworkers (CR) *p < .05; **p < .01 Variable  .70 .73 ---1 3 CFI = .95, SRMR = .05, RMSEA = .08) and a significantly better fit (p < .05) than alternative models in which the focal study constructs were variously combined. All factor loadings were statistically significant (p < .05), and all study variables had average variance-extracted estimates above .50. We also determined that the average variance shared between each construct and its indicators was greater than the shared variance between that construct and each other construct. These results demonstrate the measures exhibit adequate discriminant validity. Boot-strapped regression results reported in Table 3

Participants and Procedures
The Study 2 sample consisted of job incumbents who were enrolled as undergraduate students at a large public university located in the southeastern USA. The sample was .14 62% male, 85% Caucasian, with an average age of 22 years (SD = 3.1). Study participants were employed in a variety of for-profit and not-for-profit organizations (e.g., pharmacy technician, accountant, and cashier), averaging 23 h (SD = 10.6) per week and 5.3 years (SD = 3.6) of work experience. They were assured verbally and in writing of the confidentiality of their responses, and informed their participation was voluntary and could be stopped at any time without penalty. As an incentive to participate, potential sample members were informed that those with complete paired data from their supervisor would be included in random drawings for a chance to win a $25 gift card. Data were obtained from multiple sources and at different points in time to reduce common method variance concerns (Podsakoff et al. 2003(Podsakoff et al. , 2012. Specifically, we temporally separated participants' ratings of CSE (time 1), their CA measure (time 2), experience of abusive supervision (time 3), and supervisors' ratings of employee deviance (time 4) in four data collection waves separated by about 2 weeks each. The CA measure was completed during class time, whereas the other measures were administered through online surveys. To match participant responses with their supervisors' assessment of deviance, we asked employees to provide the name and contact information for their direct supervisor.
We first emailed supervisors a link to an online rating form, with an accompanying message containing the name of the employee to be rated. Follow-up reminders were emailed about 2 weeks later. Non-respondents were mailed a printed copy of the rating form and a postage-paid return envelope. If they had not responded after 2 weeks, follow-up reminders were made by telephone. We obtained completed employee surveys and supervisor data from 189 pairs. Because 26 participants provided survey data without completing our CA measure, the final sample consisted of 163 leader-follower pairs, yielding an 86% response rate.

Measures
Core self-evaluation CSE was measured with the same scale used in Study 1.
Abusive supervision Abusive supervision was measured in the same manner as in Study 1.
Cognitive ability CA was assessed with the Wonderlic Personnel Test (WPT; Wonderlic Inc 2002), which is a 50-item general intelligence test in which questions become progressively more difficult and reflect a variety of problem types (e.g., word comparisons, number series, logic solutions). This process captures a variety of CA  (Carroll 1993). Because no measure is all inclusive, we felt using a global measure would be a conservative means of tapping cognitive abilities important in shaping responses to abuse. The 2002 WPT manual reports test-retest reliabilities ranging from .82 to .94 and inter-form reliabilities ranging from .73 to .95. Employee deviance Supervisors assessed employees' interpersonal and organization deviance using shortened versions (see Kluemper et al. 2013) of Bennett and Robinson's (2000) original measures. A five-point scale was used (1 = never; 5 = daily) to assess the frequency with which employees had engaged in each of the deviant behaviors in the past year. Sample items include "Made fun of someone at work" (interpersonal deviance; 5 items) and "Came in late to work without permission" (organization deviance; 5 items).
Control variables We controlled for gender, age, organizational tenure, conscientiousness, agreeableness, and extraversion. Personality traits were measured with 30 items from the IPIP (Goldberg et al. 2006) using a fivepoint scale (1 = strongly disagree; 5 = strongly agree). Sample items include "I am always prepared," "I make people feel welcome," and "I feel comfortable around people."

Results
Descriptive statistics and correlations coefficients for all study variables are shown in Table 4. In Table 2, we report results from several confirmatory factor analyses evaluating the discriminant validity of the study variables. The results of a four-variable model (Model 1: CSE, abusive supervision, interpersonal deviance, organization deviance) demonstrated acceptable fit to the data (χ 2 = 1321, df = 623, CFI = .93, SRMR = .07, RMSEA = .08) and a significantly better fit (p < .05) than alternative models in which the focal study constructs were variously combined (Models 2 through 7). We also assessed standardized factor loadings, composite reliability coefficients, and variance-extracted estimates. For Model 1 all factor loadings were statistically significant (p < .05), and all study variables had average variance-extracted estimates above .50 (Fornell and Larcker 1981). Finally, we assessed discriminant validity by ensuring that the average variance shared between each construct and its indicators was greater than the shared variance between that construct and each other construct. Results from these analyses indicated there was adequate discriminant validity for.
Boot-strapped regression results reported in Table 3 demonstrate the indirect effects of CSE on interpersonal Results from these analyses would indicate that individuals lower in CSE are associated with their perceived experience of supervisory abuse and, in turn, their subsequent deviant behaviors. These results provide support for Hypothesis 1. As shown in Table 5, a significant moderation effect for CA was found in predicting interpersonal (ΔR 2 = .02, p < .05) and organizational deviance (ΔR 2 = .03, p < .01). As Fig. 2 shows, the relationship between abusive supervision and workplace deviance is stronger when CA is low. We then calculated the significance of the simple slopes using one standard deviation above and below the mean. The slope of the line between abusive supervision and workplace deviance when CA is low is statistically significant, but the slope of the line representing high CA is not significant. Thus, Hypothesis 2 is supported.
Hypothesis 3 proposed that CA would moderate the indirect (mediated) relationship of CSE on workplace deviance through abusive supervision. As shown in Table 5, the indirect effects were significant at low (− 1 SD) but not high (+ 1 SD) CA levels. In particular, the indirect effects of CSE (through abusive supervision) on interpersonal and organization deviance were significant (p < .01) among individuals with low but not high levels of CA. Our results support the thesis that those lower in CSE and CA are more likely to manifest workplace deviance after experiencing abusive supervision. As such, Hypothesis 3 is supported.

Participants and Procedures
The Study 3 sample consisted of Romanian fulltime working adults. Participants were independent subordinate-supervisor dyads employed in a range of organizations (e.g., retail, telecommunications, consumer goods, financial institutions). No monetary incentives were provided to study participants. Subordinates' responses were collected by trained professional interviewers during off-work hours at the participants' homes. In following this protocol, we sought to improve the response rate and reduce disruption to participants' work. Surveying subordinates away from the presence of supervisors and colleagues may reduce underreporting of unethical behaviors or sensitive information (cf. Tourangeau and Yan 2007). Demographic information and responses to personality and CA measures were gathered from respondents during the initial wave of data collection. Their perceptions of abusive supervision and supervisor-directed deviance were collected in a second survey administered 4 weeks later. As in other studies where abusive supervision and supervisordirected deviance data were collected in the same survey (e.g., Mitchell and Ambrose 2007;Thau and Mitchell 2010), methodological separation was used to minimize potential bias in relations between these two variables (Podsakoff et al. 2012). The two scales were physically separated in the survey, and the deviance scale preceded the abusive supervision measure to reduce chances of the proposed mediator priming responses to the dependent variable. Two weeks after the second employee survey was administered, supervisors provided ratings of interpersonal and organization deviance.
We obtained usable matching data from 281 subordinate-supervisor pairs (70% response rate). The sample was 69% female, with an average age of 34 years (SD = 11.7). Ninety-six percent had a college degree, worked an average 39 h (SD = 9.7) per week, and spent 4 (SD = 3.4) years working with their supervisor. Approximately one-third of subordinates held blue-collar positions while two-thirds were white-collar professionals.

Measures
We followed procedures recommended by Hulin (1987) to translate Study 3 measures into the Romanian language. To verify semantic equivalence between original and translated measures, each item was independently back-translated by two of the authors, who then reviewed the translations to reach consensus. CSE and abusive supervision were assessed with the same scales used in the first two studies. The CSE, cognitive ability, and deviance measures have been previously validated and used with Romanian samples (e.g., Ilie et al. 2012;Ispas et al. 2010).
Cognitive ability CA was assessed with the General Ability Measure for Adults (GAMA; Naglieri and Bardos 1997), a 66-item general intelligence test which, like the WPT, reflects a variety of problem types (e.g., matching, analogies, sequences, construction). Test-retest reliabilities range from .67 to .84, and the GAMA demonstrates convergent validity with the WPT (i.e., rs around .70; Iliescu and Livinţi 2008;Naglieri and Bardos 1997).
Subordinate deviance Supervisors assessed subordinates' interpersonal and organization deviance with the short version of Spector et al.'s (2006) measure. A five-point scale was used (1 = never; 5 = every day) to assess the frequency with which employees had engaged in each of 32 behaviors. Sample items include "Insulted or made fun of someone at work" (interpersonal deviance; 18 items) and "Came to work late without permission" (organization deviance; 14 items). Subordinates reported the extent of their supervisordirected deviance on a five-point frequency scale (1 = never; 5 = everyday) with the three items from Skarlicki and Folger's (1997) measure that capture retaliation against one's immediate supervisor (see Tepper et al. 2009). A sample item is "I talked back to my boss." Control variables We controlled for gender, age, dyad tenure, conscientiousness, agreeableness, and extraversion. Personality traits were assessed with the Romanian version (Ispas et al. 2014) of the NEO-FFI (Costa and McCrae 1992), which has been validated with over 4000 individuals across six samples following International Test Commission guidelines (2017). Respondents indicated their agreement (1 = strongly disagree; 5 = strongly agree) with 12 items for each trait.

Results
Descriptive statistics and intercorrelations for all Study 3 variables are given in Table 4. We again conducted confirmatory factor analysis to evaluate the discriminant validity of the Study 3 variables (see Table 2). We used the single-factor method to create three parcels for each construct (Landis et al. 2000). The results of a five-variable model (Model 1: CSE, abusive supervision, interpersonal deviance, organization deviance, supervisor-directed deviance) demonstrated acceptable fit to the data (χ 2 = 280, df = 80, CFI = .94, SRMR = .076, RMSEA = .095) and a significantly better fit (p < .05) than alternative models in which the focal study constructs were variously combined (Models 2 through 12). We also assessed standardized factor loadings, composite reliability coefficients, and variance-extracted estimates. All factor loadings were statistically significant (p < .05), and all study variables had average variance-extracted estimates above .50. Finally, we assessed discriminant validity by ensuring that the average variance shared between each construct and its indicators was greater than the shared variance between that construct and each other construct. Results indicated there was adequate discriminant validity.
As shown in Table 6, a significant moderation effect for CA was found in predicting interpersonal (ΔR 2 = .04, p < .01), organizational deviance (ΔR 2 = .01, p < .05), and supervisor deviance (ΔR 2 = .08, p < .01) from abusive supervision. As Fig. 3 shows, the relationship between abusive supervision and workplace deviance is stronger when CA is low. We then calculated the significance of the simple slopes using one standard deviation above and below the mean. The slope of the line between CSE and workplace deviance when CA is low is statistically significant, but the slope of the line representing high CA is not significant. These results provide support for Hypothesis 2.
Hypothesis 3 proposed that CA would moderate the indirect (mediated) relationship of CSE on workplace deviance through abusive supervision. As shown in Table 6, the indirect effects did not differ at low versus high CA levels. As such, Hypothesis 3 is not supported. When considering factors that might explain the lack of replication for this hypothesis in Study 3, we discovered that the Study 3 sample differs from Study 2 in that the Romanian subjects were, on the average, working far more hours per week. (Study 2 subjects worked 24 h per week on average while Study 3 subjects worked 39 h per week on average.) We suspected this demographic difference may have influenced our test of Hypothesis 3. Neves (2014) found that with higher levels of uncertainty, stronger relations between CSE and abusive supervision surfaced. Elsewhere, Zhang and Liao (2015) suggested subordinates working fewer hours may experience greater ambiguity in relationships with their supervisors and found time spent with supervisors moderated abusive supervision effects.
In light of these findings, we examined whether hours worked per week acted as a moderator in our model. We conducted post hoc analyses with hours worked per week as a first-stage moderator in our model. These post hoc analyses reveal that the indirect effects are indeed significant at low (− 1 SD) but not high (+ 1 SD) CA levels. In particular, the indirect effects of CSE on interpersonal, organization, and supervisor-directed deviance (as transmitted by abusive supervision) were significant (p < .05) among individuals with low but not high levels of CA. For all three variables, the second stage of the mediated effects varied significantly across levels of the moderator.

Discussion
In this paper, we examined how CSE is related to interpersonal deviance through abusive supervision and when CA moderated this relationship. Our theoretical model and empirical results are consistent with tenets of victim precipitation theory. We found general support for the idea that subordinates with lower-CSE levels are subject to more abusive supervision, and in turn react with greater workplace deviance. In this respect, we advance the literature by shedding light on how those with low CSE become victimized by supervisors, and then respond by directing counterproductive behaviors toward coworkers, the organization, and their supervisor. To our knowledge, this the first set of studies that have examined theoretical linkages emanating between CSE, abusive supervision, and employee deviance. Findings uncovered in Studies 1 through 3 suggest that examining only subordinate or supervisor antecedents of employee deviance might miss subtle connections between these parties that result in undesirable employee behavior. The integrated approach employed in these studies may encourage future fine-grained examinations of interactions between abused subordinates and abusive supervisors.
We also extend victim precipitation theory by testing whether certain characteristics of victims restrain them from reacting in ways that might exacerbate later interactions with their supervisors. More specifically, employees' cognitive capacities were found to influence their deviant behavioral reactions to supervisory abuse. Results across both studies testing this effect demonstrated that abusive supervision had a stronger impact on deviance for individuals with lower rather than higher CA levels. Much research suggests employees' emotion management abilities can help them control reactions stemming from abuse (e.g., Sears and Holmvall 2010) and minimize workplace deviant behaviors (Kluemper et al. 2013). However, there is a dearth of research examining cognitive processes in connection with deviant work behavior (Dilchert et al. 2007). Related, researchers have distinguished between "hot" acts of deviance motivated primarily by affective states and "cold" acts primarily driven by cognitive judgments of anticipated costs and benefits (Fox and Spector 2010). Little empirical research has addressed the latter issue. Our finding that higher levels of CA attenuate deviant reactions to abusive supervision suggests cognitive resources are used to inhibit emotionally laden reactions. This process should enhance emotion management (e.g., McRae 2016). Nonetheless, those higher in CA may still react, but not in a deviant manner (e.g., filing legitimate grievances, informing upper-level management, or engaging in acts against the abuser that are not easily detected). Power differentials between supervisors and subordinates, as well as the potential costs of retaliation, may discourage abused employees from lashing out against abusive supervisors and encourage them to make more calculated responses (Tepper et al. 2009;Thau and Mitchell 2010). Achieving a better understanding of the role of CA may yield novel insights into when subordinates' instrumental responses to abusive supervision may trump their urge to vent pent up feelings. Our results should encourage researchers to examine circumstances when cognition-driven strategies are employed to counter abuse rather than more affectdriven ones.

Practical Implications
Organizations would be remiss to focus only on the subordinate side of the deviance equation given that follower CSE may be linked indirectly with deviance through abusive supervision. The effects of abusive supervision are especially telling for low-CSE employees. When problematic supervisor-subordinate dyads are identified, organizations should encourage supervisors to participate in leadership development programs aimed at mitigating abusive supervision. Because predicting specific occurrences of abuse is difficult due to the complexity of the variables involved, organizations might be better served by first attempting to create more ethical work climates in which mistreatment is much less likely. Uncivil work climates can adversely affect employees beyond the impact of personal mistreatment experiences (Glomb and Liao 2003;Lim et al. 2008). Mossholder et al. (2011) proposed that climates pertinent for employee relationships may be affected by human resource systems as well. We offer that broadly promoting positive and ethical relational climates would help organizations to diminish the occurrence of harmful workplace deviance behaviors.
When specific instances of abuse have been identified, organizations might provide specialized training to supervisors who might be prone to target certain types of employees (e.g., low CSE). Such training could focus directly on behavior (e.g., conflict resolution; Pearson and Porath 2005) or be more diffuse in focus (e.g., mindfulness training; Brown and Ryan 2003). Supervisors might also be assigned coaches or mentors to help them improve interpersonal competencies and manage stress levels (Joo et al. 2010). Mitchell and Ambrose (2012) recently suggested educating employees in problem-solving techniques as constructive responses to abusive supervision. We believe this palliative would also work well with supervisors because it would require dealing with the "facts" underlying negative interactions with subordinates as well as interpersonal characteristics of subordinates that highlight tendencies for them to become abused victims.
If such proactive approaches in addressing abusive supervision are ineffective, organizations should consider sanctions on offending supervisors. Because supervisors enjoy higher social status than rank-and-file employees by virtue of their position, they might not initially be subject to repercussions for such unethical behavior bordering on the abusive (Shapiro et al. 2011). Politically expedient processing of supervisor abuse that results in very few repercussions would send the wrong signals and reinforce employee silence. Followers who are reticent to speak up (viz., those with a negative self-concept; LePine and Van Dyne 1998) may inadvertently convey acceptance of supervisory misconduct and, as such, encourage continued abuse (Pearson and Porath 2005).
Increasing workloads, communication opportunities, and workplace interdependencies have increased the potential for interpersonal mistreatment and deviance in the workplace (Porath and Pearson 2010). We have framed the issue of workplace deviance as being affected by leader actions, implying that organizations could attempt to reduce workplace deviance through actions directed at either party (i.e., followers or leaders). In regard to followers, an emphasis should initially be placed on selecting applicants who are higher in CSE. As Judge and Kammeyer-Mueller (2011) noted, it is unrealistic to rely only on CSE selection to immunize the workplace against deviance because of difficulties in selecting employees on characteristics subject to positive impression management biases. They suggested there may be ways of bolstering individuals' CSE through positive performance feedback (e.g., Schinkel et al. 2004) and workspecific contingencies designed to raise employees' selfevaluations (e.g., Ferris et al. 2009). Extrapolating Olweus's (1993) recommendations to organizational settings, we would emphasize to managers the importance of developing employees' self-confidence and interpersonal skills in working with colleagues and standing up for themselves.

Limitations and Future Research
Using supervisor ratings of deviance may have affected our results, in that employees with lower levels of CA may have been rated less favorably in general. This negative bias against such employees could have spilled over to supervisors' interpersonal and organization deviance ratings. However, the correlations between CA and workplace deviance in Studies 2 and 3 are not significant, thereby minimizing this concern. Further, Kluemper et al. (2015) showed that correlations between personality traits and deviance are similar in magnitude across self and supervisor ratings of deviance.
Although our results support our conceptual framework, alternative causal relations among study variables cannot be ruled out by our study design. However, we argue that the causal ordering we proposed among the constructs is substantively logical. First, regarding the primacy of CSE, its core traits are partly hereditary and stable temporally (Judge and Kammeyer-Mueller 2011). Second, researchers have argued (e.g., Mitchell and Ambrose 2007;Thau et al. 2009) and demonstrated empirically, in both field and laboratory settings (e.g., Lian et al. 2012a, b;Mitchell and Ambrose 2012), that abusive supervision affects subsequent employee deviance. Further, we addressed this issue in part in Study 2 and Study 3 by temporally separating the assessment of our four constructs in the theorized causal order. Thus, we maintain that reverse causality is unlikely. We tested this possibility via empirically examining other causal orderings with our data. Employee deviance failed to mediate the relationship between CSE and abusive supervision, and abusive supervision failed to mediate the relationship between employee deviance and CSE. Further, CA did not moderate either of these alternative temporal indirect effects.
Although the mediation and moderation hypotheses were supported in Study 3, the moderated mediation proposed in Hypothesis 3 was not. However, through post hoc reliance on theoretical arguments advanced by Neves (2014), we employed number of hours worked per week as a stage 1 moderator (Edwards and Lambert 2007) of relations between CSE and abusive supervision. We were unable to test this conditional effect in Study 2 due to of the low average hours worked per week in the sample. As such, it appears that across both studies our proposed model holds when hours per week are low, but perhaps not (in Study 3) when hours per week are high. This is intriguing, as working fewer hours per week may be a proxy for the types of low-CSE employees that may be abused. Specifically, part-time employees may be less influential, embedded, or valued by supervisors, leading to more abuse. Future research could examine the interaction between hours worked (and other theoretically related constructs) and CSE on abusive supervision. Although future research should examine this moderated mediation model in other cultures, the successful constructive replication in Romania suggests our model is not culture bound. Our model appears to have implications for work settings in other countries.
Another avenue for future research is to expand the outcomes of victim precipitation via abusive supervision beyond acts of workplace deviance. It is possible, for example, that high CA subordinates might exhibit "challenging" types of organizational citizenship behavior (e.g., speaking up/voice-Grant 2013; taking charge- Morrison and Phelps 1999) in reaction to supervisory abuse. Now that we have found a significant effect for the reactivity of low CA subordinates, it seems reasonable for future research to pursue a wider range of reactions to abusive supervision.
Our findings shed new light on a type of employees likely to be abused by supervisors. In addition, whereas low-CSE individuals may experience supervisory abuse, our results suggest that those with higher CA are less likely to engage in retaliatory, deviant behaviors. Our theoretical framework considers how employee characteristics influence their interpersonal relations with supervisors and how they respond to interactions perceived as abusive. It also suggests supervisors serve as an interpersonal linchpin connecting employees' CSE to their deviant behavior. In sum, the reported research contributes to our understanding of the potential positive consequences of high CSE, high CA, and low abusive supervision, along with their complex interplay when found at less optimal levels to produce workplace deviance.