Introduction

Leadership research has long recognized the close link between leaders’ emotions and leadership (George 2000; Gooty et al. 2010; Harms and Credé 2010; Kaiser et al. 2015; Rajah et al. 2011; Walter et al. 2011). Research considers leadership as an emotion-laden interpersonal process, where supervisors express emotions they experience and affect subordinates’ emotions, attitudes, and behaviors (Dasborough and Ashkanasy 2002; George 2000). More specifically, supervisors may interact with their subordinates by engaging in certain leadership behaviors to express their feelings; and it is such expression that connects supervisors’ feelings and subordinates’ work-related outcomes.

Although the notion that supervisors’ emotional experiences will influence the types of leadership behaviors they display has been widely discussed, only a few studies were conducted to empirically test this idea. Research found that when supervisors had higher positive mood, they showed more transformational leadership behaviors (e.g., Chi et al. 2011; Jin et al. 2016). However, research on how supervisors’ negative affect at work affects their interaction with subordinates is limited. According to the cognitive-neoassociationistic model of aggression (Berkowitz 1989, 1990), when one has negative feelings, his/her hostile intention will be automatically triggered due to the activation of an associative network of negative concepts in one’s memory, and therefore s/he is more likely to engage in aggressive behaviors. Given that abusive supervision is defined as prolonged hostile behaviors toward subordinates, excluding physical violence (Tepper 2000), supervisors’ negative affect may manifest itself in abusive supervision, which involves putting subordinates down in front of others or telling subordinates that their thoughts or feelings are stupid.

However, supervisors do not lash out at anyone when they are unhappy. The cognitive-neoassociationistic model of aggression (Berkowitz 1989, 1990) further suggests that, it is not always the case that one will turn his/her negative feelings into aggression because people’s higher cognitive processes would “suppress or enhance the action tendencies associated with these feelings” (Anderson and Bushman 2002, p. 31). Specifically, individuals will carefully consider whom they could express their aggressive impulse to. Hence, supervisors may be selective in choosing which subordinate to abuse. This is in line with some arguments in the existing leadership literature that destructive leader behaviors could be goal directed and planned and therefore personalized (Krasikova et al. 2013; Tepper et al. 2012). Existing studies have focused on subordinates’ traits and performance to explain which subordinates are more likely to be the victims of abusive supervision when supervisors encounter negative experiences (Tepper et al. 2006, 2011). We extend this line of research by considering leader–member exchange quality (LMX, Graen and Uhl-Bien 1995) as a moderator influencing the relationship between supervisors’ negative affect and abusive supervision, to clarify why some subordinates are “safe targets” of abusive supervision while others are not. We choose LMX because it describes the social interaction between supervisors and subordinates, and “studying emotionality in the domain of leadership is not complete without taking into account leader-member relationships” (Rajah et al. 2011, p. 1113).

Lastly, by integrating research on the consequences of abusive supervision (e.g., Tepper 2000; Zellars et al. 2002) and examining three subordinate outcomes, including negative affect at work, job satisfaction, and personal initiative, we are able to depict a comprehensive picture of how supervisors’ negative affect influences subordinates from various perspectives and when the negative cascading effects occur. Figure 1 shows the theoretical model of this study.

Fig. 1
figure 1

Theoretical model

Full size image

This study aims to contribute to the leadership literature from three perspectives. First, we contribute to research on abusive supervision by introducing the cognitive-neoassociationistic model of aggression (Berkowitz 1990) to the literature and integrating it with the power/resource interdependence perspective and victim precipitation theory. Simultaneously considering a supervisor-related antecedent (i.e., supervisors’ negative affect at work), a relationship-related antecedent (i.e., LMX), and their interaction allows us to identify which subordinates are more likely to be affected than others. Second, we contribute to the line of research on how leaders’ emotional experience trickles down to influence subordinates’ well-being and behavior by incorporating abusive supervision as a behavioral mechanism and LMX as a boundary condition. Third, we tested our theoretical model in China, a high power distance culture (Hofstede 2001), where abusive supervision is more likely to occur in the workplace (Hu et al. 2011). Given the authoritarian nature and the norm of compliance in the Chinese society, it is imperative for researchers to examine the profound effects of abusive supervision on subordinates’ well-being (i.e., negative affect and job satisfaction). Additionally, Chinese subordinates are less likely to directly retaliate against abusive supervisors; instead, they may react to such mistreatments in a conservative and passive-aggressive manner (i.e., withholding personal initiative).

Theoretical Background and Hypothesis Development

Supervisors’ Negative Affect and Abusive Supervision

The cognitive-neoassociationistic model of aggression (Berkowitz 1989, 1990) helps us to understand the link between supervisors’ negative affect and their reactions caused by negative affect. This model highlights that one’s feelings, thoughts, and expressive motor reactions are stored in emotional networks in memory, with one concept connecting to others. For example, negative feelings, negative thoughts, and negative memories are linked together in the same negative emotional network. When individuals experience negative affect, aggressive inclinations will be automatically activated regardless of its cause because they are all stored in a negative emotional network in memory. Therefore, aggressive thoughts are more salient to those who experience higher negative affect and such individuals are more likely to enact aggressive behaviors than those who experience lower negative affect. Based on this model, it is straightforward to argue that, when supervisors experience higher negative affect at work, their aggressive inclinations will be automatically triggered, and therefore they are more likely to behave aggressively compared to others experiencing lower negative affect.

In the workplace, displaying aggressive behaviors might lead to punishment on the perpetrators and even retaliation from the victims (Aquino and Thau 2009). Supervisors, who have higher hierarchical status, are more likely to vent their negative feelings to subordinates, instead of higher-level managers or clients, because they have greater legitimate power over their subordinates (Aquino 2000). This may lead to behaviors indicative of abusive supervision. Although the traditional literature on abusive supervision defined it as supervisors’ sustained behavioral pattern perceived by subordinates (Tepper 2000), recent research showed that abusive supervision was not as stable as it was originally conceptualized and that supervisors’ display of abusive behaviors fluctuated on a monthly and even daily basis (Barnes et al. 2015; Simon et al. 2015). Therefore, it is plausible that supervisors’ experiences in a specific period of time would affect their display of abusive supervisory behaviors in the same period of time.

Research on abusive supervision has found various aversive events/experiences as antecedents of abusive supervision, such as supervisors’ perceived organizational injustice and interpersonal conflicts (Aryee et al. 2007; Burton et al. 2012; Tepper et al. 2011). To note that, such aversive experiences will elicit a variety of negative affective experiences at work (Bruk-Lee and Spector 2006). Therefore, supervisors’ negative affect can be regarded as a more proximal antecedent of abusive supervision. Indeed, the link between negative affect and abusive supervision has been supported in previous research. Engaging in abusive supervision was considered as a means of coping with negative affect (Mawritz et al. 2014) or a way to restore balance at work (Hoobler and Hu 2013). Emphasizing the role of negative affect also echoes Berkowitz’s sentiment that “it is not the exact nature of the aversive incident that is important but how intense the resulting negative affect is” (1989, p. 68). Based on the cognitive-neoassociationistic model of aggression (Berkowitz 1989, 1990), we propose,

Hypothesis 1

Supervisors’ negative affect at work is positively related to abusive supervision.

The Moderating Role of LMX

In the workplace, we do not always see unhappy supervisors venting their negative feelings to every subordinate. Tepper argued that “abusive supervision may be characterized as displaced aggression against safe target” (2007, p. 269). This is in line with the second argument embedded in the cognitive-neoassociationistic model (Berkowitz 1990), that despite the heightened aggressive tendency triggered by negative affect, individuals’ higher-order cognitive processes will suppress or enhance the transition from negative feelings to aggressive behaviors toward specific targets. Specifically, one will analyze the consequences of being aggressive toward specific others before enacting aggressive behaviors. That is, higher cognitive processes will guide supervisors to make choices based on the costs and benefits of showing hostile nonphysical behaviors to express their negative feelings, and to determine which subordinate, among others, is the safe target before they direct their aggressive drive to a certain subordinate. Thus, factors influencing supervisors’ choices will moderate the relationship between supervisors’ negative affect and abusive supervision.

Although diverse factors will influence higher cognitive processes, we argue that relationship quality with subordinates (i.e., LMX) is one of the most relevant moderators to be considered because supervisors often interact with subordinates in different manners based on the relationship quality with individual subordinate (Henderson et al. 2009). In low-LMX supervisor–subordinate dyads, the exchange between supervisors and subordinates is “contractual,” with subordinates assuming responsibilities assigned to them based on contracts. In high-LMX supervisor–subordinate dyads, supervisors and subordinates share similar goals, and show care and support to each other. Such high-LMX dyads are characterized with high mutual trust, respect, and obligations (Graen and Uhl-Bien 1995). Based on research on power/resource interdependence, we propose that low-LMX subordinates are often identified as safe targets.

Despite the emphasis of supervisors’ power over subordinates in the prevailing leadership literature (e.g., Aquino et al. 2006), power and resources between supervisors and subordinates are interdependent (Tjosvold 1989). Supervisors have a variety of positional resources for exchange, such as control of payroll, promotion opportunities, and delegation. Although subordinates cannot exchange such resources in kind, they often repay the favors in other forms, such as lateral information, commitment and loyalty toward their supervisors, and higher performance (Oc and Bashshur 2013; Wilson et al. 2010). Therefore, supervisors are dependent on subordinates to achieve personal benefits (e.g., career advancement) by acquiring valuable resources (Oc and Bashshur 2013; Wilson et al. 2010).

Supervisors will anticipate and evaluate potential consequences and strategically engage in abusive behaviors (Tepper et al. 2012). Research has shown that, when being abused, subordinates may reduce their work effort (Mackey et al. 2013) or withhold extra resources (Ouyang et al. 2015; Zellars et al. 2002). Supervisors may anticipate different impacts of resource-withholding behaviors from different subordinates. In lower-LMX dyads, both supervisors and subordinates have fewer valuable resources to exchange or they are less motivated to exchange (Graen and Uhl-Bien 1995), therefore, abusing low-LMX subordinates will generate less of a threat of resource loss to supervisors. In contrast, potential resource loss caused by abusing high-LMX subordinates will be higher because such subordinates possess abundant and valuable resources for exchange (Wilson et al. 2010), and they are more inclined to exchange by demonstrating higher work performance and enacting more extra-role behaviors (Gerstner and Day 1997; Ilies et al. 2007). Therefore, when supervisors experience negative affect at work, it is more strategic to abuse low-LMX subordinates than to abuse high-LMX subordinates.

Additionally, victim precipitation theory (Aquino et al. 1999; Elias 1986; Olweus 1978), which has been widely used in the abusive supervision literature (e.g., Henle and Gross 2013; Martinko et al. 2011; Tepper et al. 2006, 2011; Wang et al. 2014), provides support for why supervisors inclined to hostility are more likely to abuse low-LMX subordinates. Victim precipitation theory suggests that individuals who are weak or defenseless may provoke hostility because such subordinates are perceived to be submissive and are less likely to fight back (Henle and Gross 2013; Wang et al. 2014). This is especially the case for subordinates who are less likable or have poorer relationships with their supervisors (Harris et al. 2011; Tepper et al. 2006). For example, Harris et al. (2011) found in two independent samples of employees that supervisors were more likely to abuse low-LMX subordinates when they perceived higher coworker relationship conflict. Low-LMX subordinates may hesitate to directly retaliate against their hostile supervisors because retaliation could result in more aggressive behaviors from the supervisors (Hoobler and Brass 2006), not to mention confronting supervisors may further jeopardize their existing relationships. Taken together, abusing low-LMX subordinates when supervisors are unhappy is a less costly choice. Based on the arguments above, the following hypothesis is formulated,

Hypothesis 2

LMX buffers the positive relationship between supervisors’ negative affect at work and abusive supervision, such that the relationship is weaker among dyads with higher LMX, compared to those with lower LMX.

Impacts on Subordinates

Supervisors’ expressions of negative affect, manifested in abusive supervision, will further influence subordinates’ well-being. We focus on two important indicators of well-being that are especially related to individuals’ work outcomes: negative affect and job satisfaction (for meta-analyses, see: Judge et al. 2001; Kaplan et al. 2009). Our choice of these two indicators also aligns with Diener and colleagues’ conceptualization of well-being as “a broad category of phenomena that includes people’s emotional responses, domain satisfactions, and global judgment of life satisfaction” (1999, p. 277). Given that being a victim of hostile behaviors is inherently a negative event, it leads to negative affective reactions and influences employees’ job attitudes (Weiss and Cropanzano 1996). Research has shown that interacting with abusive supervisors evokes long-lasting negative affect, such as tension, anxiety, and emotional exhaustion (Tepper 2007). Further, subordinates are usually less satisfied with their jobs when they work with abusive supervisors (Bowling and Michel 2011; Schyns and Schilling 2013).

Not only would working with abusive supervisors influence subordinates’ well-being, it also impacts their work-related behaviors. Social exchange theory (Cropanzano and Mitchell 2005) suggests that reciprocity is an important component in supervisor–subordinate interaction; that is, subordinates tend to react to supervisors’ positive treatment by enacting positive behaviors and repay supervisors’ negative treatment by offering unfavorable returns. After being abused, subordinates might retaliate by engaging in interpersonal deviant behaviors toward their supervisors or colleagues (e.g., Lian et al. 2014; Mitchell and Ambrose 2007). Because deviant behaviors are mostly sanctioned in organizations, subordinates might also react to abusive supervision in a more subtle and passive-aggressive way, by reducing their reciprocation and withholding efforts in extra-role behaviors. Research has supported that abusive supervision was related to fewer citizenship behaviors from subordinates (e.g., Aryee et al. 2007; Rafferty and Restubog 2011; Zellars et al. 2002).

In this study, we focus on another way for subordinates to retaliate against abusive supervisors, which is to limit their personal initiative at work. Personal initiative is conceptualized as employees’ self-starting behaviors to overcome barriers to achieve a goal (Frese et al. 1996). What distinguishes personal initiative from citizenship behavior is a long-term orientation. Employees who enact more initiative behaviors contribute to the organizations in the long term as they persist regardless of barriers placed upon them to achieve their goals. However, in the short term, initiative behaviors might not be favorable to supervisors as employees may voice too often and push too hard for changes (Frese and Fay 2001; Frese et al. 1996). Thus, withholding personal initiative is an even more subtle passive-aggressive way to reduce one’s reciprocation to abusive supervisors as it actually lessens supervisors’ burden of changes in the short term, yet one that will hamper organizational performance in the long run.

Given the potential negative impacts of abusive supervision on subordinates, it is straightforward to expect indirect effects from supervisors’ negative affect to subordinate outcomes (i.e., negative affect at work, job satisfaction, and personal initiative) via abusive supervision. This is in accordance with the crossover literature, which specifies that one’s negative feelings and perception will cross over to significant others through negative social interactions (Westman 2001). For example, when school principals experience strain at work, they are more likely to engage in social undermining behaviors, which enhance teachers’ strain levels (Westman and Etzion 1999). Integrating this idea with the moderating effect of LMX discussed earlier, we propose,

Hypothesis 3

The indirect effects from supervisors’ negative affect to subordinate outcomes, including subordinates’ negative affect at work, job satisfaction, and personal initiative at work, via abusive supervision are stronger among dyads with lower LMX, compared to those among dyads with higher LMX.

Method

Participants and Procedures

We recruited participants from employees working in five hotels in Shanghai, China. One of the authors was introduced to the supervisors in these five hotels by the directors at the Human Resource (HR) Departments. This author introduced our study, invited the supervisors to participate and emphasized that participation was voluntary and anonymous. Upon receiving their consents, we distributed the package of questionnaires to each supervisor. This package of questionnaires includes one questionnaire booklet for the supervisor (measuring demographic information, supervisors’ negative affect at work, and subordinates’ personal initiative) and the other for one of their immediate subordinates (measuring demographic information, LMX, abusive supervision, subordinates’ negative affect at work, and job satisfaction). Each pair of supervisor and subordinate questionnaires was assigned with a code, which was stamped on the questionnaires, so that we were able to match the supervisor–subordinate dyads. Supervisors were required to pass the subordinate booklet to the first subordinate they encountered the next workday after receiving the questionnaire booklet in order to prevent potential biases caused by supervisors’ choices of subordinates. On the first page of the subordinate booklet, we introduced this study and ensured their confidentiality. To further ensure confidentiality, we attached empty envelopes for both supervisors and subordinates and asked them to seal the completed questionnaires in the envelopes and drop the envelopes in the collection box located outside the HR Department office, respectively.

Among the 200 questionnaire packages distributed, we received 189 supervisor questionnaires and 189 subordinate questionnaires, with a response rate of 94.5 %. One hundred and eighty pairs of supervisor and subordinate questionnaires were matched. Among these one hundred and eighty matched supervisor–subordinate dyads, 36.7 % are from Food and Beverage Department, 24.4 % are from Front Office, 13.9 % come from Housekeeping Department, and 13.9 % come from Marketing Department. Among supervisors, 57 % are male, with an average age of 36 years old. Among subordinates, 45 % are male, with an average age of 31 years old. On average, subordinates have been working with these supervisors for 4 years.

Measures

Because all measures of the main variables used in this study were developed in English, we followed Brislin’s (1980) back-translation procedure to translate them into Chinese.

Supervisors’ and Subordinates’ Negative Affect at Work

We measured supervisors’ and subordinates’ negative affect at work using the 10 adjectives from the Positive and Negative Affect Schedule (PANAS, Watson et al. 1988). Participants were asked to indicate the frequency of their negative affective experiences at work during the past one month, on a 5-point Likert scale (1 = very slightly or not at all, 5 = extremely). Sample negative affect adjectives include “irritable,” “distress,” “nervous,” and “hostile.” Cronbach’s α for this scale was .84 among supervisors and .82 among subordinates.

Abusive Supervision

We measured abusive supervision using the 15-item scale developed by Tepper (2000). Subordinates were asked to rate how frequently their supervisors engaged in the 15 behaviors during the past one month, on a 5-point Likert scale (1 = never, 5 = very often). Sample items of abusive supervision include, “My supervisor ridiculed me,” and “My supervisor told me I was incompetent.” Cronbach’s α for this scale was .93.

LMX

The quality of LMX was measured using the 7-item scale used in the study by Janssen and Van Yperen (2004). This scale was developed based on several versions of LMX scales and had a high internal reliability. Subordinates were asked to indicate whether they agreed or disagreed with the statements about the relationship quality with their supervisors on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Sample statements include “My supervisor would be personally inclined to help me solve problems in my work” and “My supervisor understands my problems and needs.” Cronbach’s α for this scale was .86.

Job Satisfaction

We adopted the five-item scale from Brayfield and Rothe (1951) to measure subordinates’ current job satisfaction. We asked subordinates to rate the extent to which they agreed with the statements using a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Sample items were “I feel fairly satisfied with my present job” and “I am finding real enjoyment in my work.” Cronbach’s α for this scale was .77.

Personal Initiative

Subordinates’ personal initiative was assessed using an adapted version of the personal initiative scale developed by Frese et al. (1997). The original scale was developed for self-assessment. We adapted the six items to the supervisors’ perspective. Supervisors were required to rate the extent to which they agreed with the statements describing the designated subordinates’ personal initiative in the past one month on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Sample items were “Whenever there was a chance to get actively involved at work, this subordinate took it” and “This subordinate took initiative immediately even when others didn’t.” Cronbach’s α for this scale was .87.

Control Variables

We controlled for subordinates’ gender and subordinates’ relationship tenure with their supervisors, because pervious research indicated that these two demographic variables not only accounted for the variance in abusive supervision (Aryee et al. 2007; Hoobler and Brass 2006; Lian et al. 2014; Wu et al. 2013), but also influenced subordinates’ reactions toward abusive supervisors (Haggard et al. 2011; Mitchell and Ambrose 2007). Subordinates’ gender was coded as 0 for female and 1 for male. Dyadic tenure was reported by the subordinates. Each subordinate was asked to indicate how long s/he had been working with the current supervisor (in the number of months).

Results

Confirmatory Factor Analysis

Prior to test our measurement model, we used the parceling strategy to form parcels for supervisors’ negative affect, abusive supervision, and subordinates’ negative affect. The parceling strategy is preferred when the sample size is small (Little et al. 2002), because by using parcels we are able to maintain a reasonable ratio of estimated parameters to sample size and provide more accurate parameter estimates. Therefore, the 10-item negative affect scale, for both supervisors and subordinates, were reduced to three parcels by randomly assigning items into each parcel (with two 3-item parcels and one 4-item parcel, respectively) and the 15-item abusive supervision scale was reduced to five parcels (with three items in one parcel).

We conducted confirmatory factor analysis (CFA) and compared the Chi-square scores and the fit indexes of the four models. The first model consisted of six distinctive factors, including supervisors’ negative affect, abusive supervision, LMX, subordinates’ negative affect, subordinates’ job satisfaction, and subordinates’ personal initiative. The second model was built upon the first model and included five factors, with abusive supervision and LMX loaded on a single factor. The third model included two factors, with one factor consisting of items rated by supervisors (i.e., supervisors’ negative affect and subordinates’ personal initiative) and the other factor consisting of items rated by subordinates (i.e., abusive supervision, LMX, subordinates’ negative affect and subordinates’ job satisfaction). The last model included all items used to measure the main variables under one general factor. Results revealed that the six factor model (χ 2 = 778.87, df = 362, CFI = .91, NFI = .85, NNFI = .90, SRMR = .07, RMSEA = .08) fit better than the five-factor model (χ 2 = 1106.75, df = 367, CFI = .85, NFI = .79, NNFI = .83, SRMR = .10, RMSEA = .12; Δχ 2 = 327.88, Δdf = 5, p < .05), the two-factor model (χ 2 = 1619.49, df = 376, CFI = .74, NFI = .69, NNFI = .72, SRMR = .12, RMSEA = .15; Δχ 2 = 840.62, Δdf = 14, p < .05), and the one-factor model (χ 2 = 1982.29, df = 377, CFI = .67, NFI = .62, NNFI = .64, SRMR = .14, RMSEA = .18; Δχ 2 = 1203.42, Δdf = 15, p < .05). Therefore, the main variables are sufficiently distinct.

Tests of Hypotheses

Table 1 shows the means, standard deviations, and correlations among the study variables. Results showed that supervisors’ negative affect was positively correlated with abusive supervision (r = .16, p < .05) and abusive supervision was positively correlated with subordinates’ negative affect (r = .23, p < .01), negatively correlated with subordinates’ job satisfaction (r = −.49, p < .01) and subordinates’ personal initiative (r = −.26, p < .01). These findings provided preliminary support for the hypothesized relationships.

Table 1 Means, standard deviations, correlations, and reliability among variables
Full size table

We tested all the hypotheses in a path-analysis framework using Mplus 7.0 (Muthén and Muthén 1998–2012) and presented the results in Table 2. In all analyses, we controlled for subordinates’ gender and dyadic tenure. Hypothesis 1 states that supervisors’ negative affect is positively related to abusive supervision. Results from Model 1 showed a positive relationship between supervisors’ negative affect and abusive supervision (B = .20, p < .05), supporting Hypothesis 1.

Table 2 Unstandardized coefficient estimates of theoretical model in Mplus
Full size table

Next, we examined whether LMX moderated the link between supervisors’ negative affect and abusive supervision. Hypothesis 2 states that the positive relationship between supervisors’ negative affect and abusive supervision is weaker when LMX is higher. Results from Model 2 revealed that the interaction between supervisors’ negative affect and LMX significantly predicted abusive supervision (B = −.28, p < .05). We followed the procedure suggested by Aiken and West (1991) to plot the interaction at the two conditional values of LMX (one standard deviation above and below the mean). We presented the interaction in Fig. 2. Simple slope analyses showed that, when LMX was low, the positive relationship between supervisors’ negative affect and abusive supervision was significant (simple slope = .32, p < .01); whereas when LMX was high, the relationship between supervisors’ negative affect and abusive supervision was not significant (simple slope = .02, p > .10). Therefore, Hypothesis 2 was supported.

Fig. 2
figure 2

Interaction between supervisors’ negative affect and LMX on abusive supervision. Note NA negative affect, LMX leader–member exchange

Full size image

To test Hypothesis 3, we estimated the indirect effects of supervisors’ negative affect on subordinate outcomes via abusive supervision when LMX is low (one standard deviation below the mean) and those when LMX is high (one standard deviation above the mean), using the bootstrapping procedure in Mplus. We used 1000 bootstrap samples to estimate the indirect effects using 95 % bias-corrected confidence intervals (CI). An indirect effect is significant when the 95 % CI does not include zero. For subordinates’ negative affect, the indirect effect was stronger when LMX was low (indirect effect = .05, 95 % CI [.01, .10]) than that when LMX was high (indirect effect = .00, 95 % CI [−.29, .03]). For subordinates’ job satisfaction, the indirect effect was stronger when LMX was low (indirect effect = −.17, 95 % CI [−.28, −.06]) than that when LMX was high (indirect effect = −.01, 95 % CI [−.10, .08]). For subordinates’ personal initiative, the indirect effect was stronger when LMX was low (indirect effect = −.09, 95 % CI [−.15, −.03]) than that when LMX was high (indirect effect = −.00, 95 % CI [−.05, .05]). Taken these results together, Hypothesis 3 was supported.Footnote 1

Discussion

General Discussion and Theoretical Implications

The results of this study supported our proposed model to explain how and why supervisors’ negative affect had profound influences on subordinates’ well-being and behavior. Consistent with the cognitive-neoassociationistic model, the present study demonstrated that although the experience of negative affect enhanced supervisors’ tendency to engage in abusive behaviors, who they expressed their negative affect to was influenced by factors that might affect their cognitive appraisal. Specifically, LMX is a key factor determining who aroused supervisors express their aggressive intention to. Drawing from research on power/resource interdependence and victim precipitation theory, the present study sheds new light on the notion that abusive supervision may be strategic (Tepper et al. 2012). We found support that LMX buffered the positive relationship between supervisors’ negative affect and abusive supervision. We reasoned that, because high-LMX subordinates possess more valuable resources to improve work unit performance and to facilitate supervisors to achieve their personal career goals, the aftermath of engaging in abusive supervision toward high-LMX subordinates may be more costly. Therefore, it is more strategic to vent negative affect at low-LMX subordinates instead of high-LMX subordinates.

Besides, we complement this study by examining the consequences of supervisors’ negative affect on subordinates’ well-being and behavior via abusive supervision. Results showed that, among low-LMX dyads, supervisors’ negative affect at work cascaded to influence subordinates’ negative affect at work, job satisfaction, and personal initiative. Such indirect effects were not found in high-LMX dyads. Aside from the hypotheses we proposed, it is important to note that supervisors’ negative affect had a direct effect only on subordinates’ negative affect (see Table 2 Model 2, B = .23, p < .01), but not on subordinates’ job satisfaction (B = −.00, ns) and personal initiative (B = −.03, ns). This can be explained by the emotional contagion literature. The emotion contagion effect happens when one person subconsciously reads and mimics the other person’s facial and verbal expressions through daily interactions, forming similar emotions or mood as the other person displays. Previous studies found a positive relationship between supervisors’ negative affect and subordinates’ negative affect (Hatfield et al. 1994; Johnson 2008; Sy et al. 2005). Consistent with these findings, we found that negative affect was contagious from supervisors to subordinates, whereas the effects of supervisors’ negative affect on subordinates’ job satisfaction and personal initiative only took place via abusive supervision.

Taken together, this study offers three important theoretical contributions to the existing literature. First, although research has found that abusive supervision is more likely to happen when supervisors are depleted and unable to control their impulses (Barnes et al. 2015; Yam et al. 2015), some instances of abusive supervision might be driven by “rational weighting of the costs and benefits associated with abuse” (Tepper et al. 2012, p. 196). Research has shown that subordinates’ attributes (e.g., negative affectivity and performance, Tepper et al. 2006, 2011) are important factors determining supervisors’ decisions of whom to abuse when they encounter negative experiences, we contribute to this line of research by drawing on the cognitive-neoassociationistic model of aggression and examining it as the product of negative affect and supervisor–subordinate relationship quality. In addition, by examining multiple antecedents of abusive supervision, we add knowledge to the limited yet increasing research on the antecedents of abusive supervision (Tepper 2007).

Second, by highlighting the unfavorable cascading effect of supervisors’ negative affect at work, this study adds evidence to the growing literature which emphasizes the role of affect and emotions in the leadership process (e.g., Brief and Weiss 2002; Sy et al. 2005) with a focus on supervisors’ negative feelings. We contribute to the leadership literature by highlighting a behavioral mechanism of how supervisors’ negative feelings influence subordinates. Summarizing studies on the influences of leaders’ feelings on followers’ outcomes, Gooty and colleagues highlighted that, what was missing in these empirical studies was the “demarcation between what leaders feel and what they display” (2010, p. 989) as supervisors might not always display how they feel in daily interactions with subordinates. The current study bridges this link by pointing out that the more negative feelings a supervisor has at work, the more likely s/he would express his/her negative feelings by abusing subordinates; and it is such negative expression that transmits the negative impacts from supervisors to subordinates. Integrating LMX as the boundary condition, we highlight that the relationship between what leaders feel and what they display is affected by dyadic relationship quality.

Third, the present study broadens our understanding of the consequences of abusive supervision on subordinates (Schyns and Schilling 2013), especially in the Chinese context. To establish an image of authority and obtain compliance from subordinates, Chinese supervisors are accustomed to engaging in hostile behaviors when interacting with their subordinates, such as belittling their subordinates or scolding them in the presence of others, which are typical indicators of abusive supervision (Cheng 1995). Hence, Chinese subordinates generally reported higher level of abusive supervision than subordinates in other countries (Hu et al. 2011). In accordance with prior research (see Tepper 2007 for a review), our research results showed that abusive supervision was detrimental to subordinates’ psychological well-being (i.e., higher negative affect and lower job satisfaction). Perhaps more interesting is our finding that Chinese subordinates retaliated against supervisors’ mistreatment in a very subtle, passive-aggressive manner by reducing their personal initiative. This study moves beyond existing research examining subordinates’ direct hostile retaliation against their abusive supervisors (Lian et al. 2014; Mitchell and Ambrose 2007) and highlights that subordinates in a high power distance country, like China, may react differently to abusive supervisors. Indeed, Wang et al. (2012) found that subordinates with low power distance value were more likely to engage in interpersonal deviant behaviors when being abused by their supervisors, as compared to subordinates with high power distance value. In a society where compliance to the superiors is a norm, being less proactive at work may be a more proper way for subordinates to retaliate against supervisors’ hostile treatment.

Limitations and Future Directions

Despite the interesting findings in this study, there are still some concerns. First, although we collected data from two different sources (i.e., supervisors and subordinates), common method variance (CMV) may still be a potential issue. For example, the relationships between abusive supervision and subordinates’ negative affect or job satisfaction might be inflated because these variables were measured by the same person at the same time. Nevertheless, we tried to mitigate the threats of CMV problems by including a supervisor-rated outcome variable (i.e., personal initiative) and conducting CFA to test whether the study variables shared substantial common variance (Podsakoff et al. 2003). Yet, it might be biased when abusive supervisors evaluate subordinates’ performance. Hence, future studies are suggested to use objective indicators to examine the relationships between abusive supervision and subordinate outcomes. In addition, due to the cross-sectional design, we are not able to infer causal relationships among the study variables. For instance, subordinates’ negative feelings at work might restrict them from performing well at work, and research has shown that poor performers are subjected to more abusive supervision at work (Tepper et al. 2011). Future studies can adopt a time-lagged design, longitudinal design, or experimental design, to clarify the causal relationships of the study variables.

Another concern lies in our measure of LMX from the subordinates’ perspective despite of our discussion of LMX from the supervisors’ perspective. However, empirical evidence showed that subordinates’ rating of LMX is a close indicator of supervisors’ rating of LMX. First, a meta-analysis by Sin et al. (2009) showed that supervisors’ rating of LMX positively predicted subordinates’ rating of LMX (ρ = .37). Second, Sin et al. (2009) also found that LMX agreement increased when the length of dyadic tenure increased. Since the average dyadic tenure in our sample is relatively longer (4 years) than those studies in the meta-analysis by Sin and colleagues, it implies that there will be a high agreement between supervisors’ perception of LMX and subordinates’ perception of LMX. Taken together, we believe that our research results will not be severely affected using subordinates’ rating of LMX, though having supervisors’ rating of LMX would provide a more rigorous design.

Moreover, although we discussed the moderating effect of LMX based on the role of cognitive processes in the cognitive-neoassociationistic model, we did not directly measure supervisors’ cognitive processes when they decided who the safe target was. Variables capturing supervisors’ cognitive processes could act as the mediators explaining why LMX moderates the relationship between supervisors’ negative affect and abusive supervision. Future studies are recommended to directly measure supervisors’ perceived cost of abusing certain subordinates and test whether perceived cost mediates the moderating effect of LMX to provide further empirical support to the application of the cognitive-neoassociationistic model to the abusive supervision literature.

Lastly, although we adopted a dyadic design in this study, future studies are recommended to move beyond and use a multilevel design. Using a dyadic design, our results can only show that the negative crossover effect is more likely to happen among dyads in low LMX. Using a multilevel design, we can further examine, given the same supervisor, whether s/he decides to abuse specific subordinates (i.e., those in worse relationships with him/her) but not others. That is, we can test whether the relationship between a supervisor’s negative affect and his/her subordinates’ perceived abusive supervision is moderated by each subordinate’s relationship with this supervisor.

Practical Implications

This study examined the impact of unhappy supervisors on their subordinates. Managers are usually expected to manage subordinates’ affect/emotions and maintain a “emotionally healthy” organization (Ashkanasy and Daus 2002; Toegel et al. 2013), the demands of which result in managers’ increased work stress and worse psychological well-being (Mignonac and Herrbach 2004; Ten Brummelhuis et al. 2014). The findings of the unfavorable cascading effects bring our attention to the importance of preventing the negative spiral from forming at the very beginning, by paying more attention to managers’ negative feelings at work and enhancing their emotion regulation abilities. Organizations are suggested to provide stress management training and emotion regulation training designed for managers. For example, mindfulness training, which includes mindfulness meditation and some informal daily awareness exercises, is helpful to improve individuals’ self-control abilities and emotion regulation abilities (Hülsheger et al. 2013). In addition, because supervisors are more likely to abuse those subordinates in worse relationships, cultivating higher quality supervisor–subordinate relationships is important in reducing the occurrence of abusive supervision. Organizations can facilitate the forming of higher quality relationships by organizing more social activities, such as New Year party and Friendship Day, and encouraging managers to be part of these activities.

Conclusion

Leaders are experiencing higher negative affect at work due to the fierce competition in today’s business world, resulting in severe impacts on subordinates’ well-being and behavior. The present research examined this phenomenon by testing a behavioral mechanism (i.e., abusive supervision) and a relational boundary condition (i.e., LMX) in the relationships between supervisors’ negative affect and subordinate outcomes. Integrating the cognitive-neoassociationistic model of aggression, research on power/resource interdependence and victim precipitation theory, we found that supervisors only vented negative affect to low-LMX subordinates by abusing them, and therefore only low-LMX subordinates’ well-being and personal initiative were influenced. We hope our study draws attention to the notion that abusive supervision may be strategic and stimulates future research on supervisors’ cognitive processes behind abusive supervision.