Introduction
Daily interactions of subordinates with their supervisors have a considerable impact on their work-related motivation, well-being, and strain (Kuoppala et al., 2008; Nyberg et al., 2011; Schyns & Schilling, 2012; Skakon et al., 2010). Because leaders’ actions are central and important to employees (van Dierendonck et al., 2004), they likely influence an individual’s self-concept (Semmer et al., 2007; Van Dick & Haslam, 2012). For instance, earlier research indicated that transformational leaders strengthen their subordinates’ self-efficacy, which increases their well-being (Nielsen & Munir, 2009). Based on recent studies, we proposed a complementary pattern for destructive leadership (Jiang et al., 2016; Priesemuth et al., 2014). As highlighted by the concept of stress as offense to self (SOS; Semmer et al., 2007), disrespect and a lack of appreciation by leaders can seriously threaten one’s sense of self and increase the risk of stress responses. According to conservations of resources theory (COR; Hobfoll, 1989, 2012), such a threat to a valuable resource, such as occupational self-efficacy, can result in a vicious circle of resource loss that reduces work engagement and causes strain. Because (the threat of) resource loss is more salient than resource gain (Hobfoll, 2012), the negative impact of destructive leadership on well-being and health is likely stronger than the positive effects of constructive leadership (cf. Schyns & Schilling, 2012). Moreover, fluctuations in abusive behaviours (Barnes et al., 2015; Courtright et al., 2016; Johnsen et al., 2012) could be experienced as particularly stressful because the behaviour of the supervisor is less predictable (Matta et al., 2017). Thus, focusing on abusive supervision as one example of destructive leadership, we proposed the negative effect of abusive supervision on well-being is mediated by lowered self-efficacy of subordinates.
Considering occupational self-efficacy as a central personal resource in the relationship between abusive supervision and well-being might also help to understand the mixed results regarding gender in that context (Burton & Hoobler, 2006; Rafferty & Restuborg, 2011; Wang et al., 2016). Recent research on abusive supervision, aggressive workplace behaviour, and bullying suggests that men and women perceive, label, and react to hostile behaviours differently (Escartín et al., 2011; Ouyang et al., 2015; Salin & Hoel, 2013; Salin, 2015; Schyns et al., 2018; Stempel & Rigotti, 2018). Occupational self-efficacy might be an important explanatory mechanism because self and other expectations regarding one’s competencies and abilities are assumed to differ between men and women (Ridgeway, 2001). Addressing potential gender differences in response to abusive supervision is not only important for furthering the theoretical framework on gender and leadership but also the implementation of tailored intervention programmes in practice. Drawing on expectations state theory (Ridgeway, 2011), COR theory (Hobfoll, 2012), and the SOS model (Semmer et al., 2007), we examined how male and female subordinates’ occupational self-efficacy is affected by abusive supervision and how this relates to their well-being and health. Overall, we suggested a moderated mediation model in which the indirect relationship of abusive supervision with well-being and health via occupational self-efficacy is contingent upon subordinates’ gender.
With our study, we extend existing research by addressing occupational self-efficacy as a central resource and potential psychological mechanism that explains how changes in abusive leadership behaviours affect the well-being of subordinates. Moreover, by drawing on the SOS model (Semmer et al., 2007), we further the theoretical foundations necessary to explain the unique mechanisms of destructive, as compared to constructive, leadership concerning well-being. We contribute to research on abusive supervision by suggesting that such leadership behaviour is seen as an offence to the self that is manifested in low occupational self-efficacy, which relates to reduced motivation and higher strain.
Additionally, we explored the role of subordinates’ gender in response to abusive supervision to gain insight into potentially divergent mechanisms for women and men. Thus, we contribute empirical evidence to foster theoretical refinement concerning gender and destructive leadership. We tested our hypotheses in a longitudinal study with separated points of measurement for the independent and the dependent variables.
Leadership and subordinate well-being
Supervisors influence their subordinates not only by their traits and direct behaviours but also by defining and controlling work characteristics and resources (Nielsen et al., 2008; Piccolo & Colquitt, 2006; Tims et al., 2011). In reference to COR theory (Hobfoll, 1989, 2012), leaders are a resource for employees in the work context because they potentially offer valuable assets, including support, inspiration, or access to information. According to COR, people strive to gain and sustain important resources and to defend them against threats. Hobfoll (2012) suggested that people who can draw on plenty of resources are more resilient against resource loss and can even expand their resources, stimulating a circle of resource gain. On the contrary, a lack of resources can trigger a loss circle, which is considered more salient than resource gain (Hobfoll, 2012). Thus, whereas a constructive leader can function as a source for development, well-being, and health (e.g., Nielsen & Munir, 2009), destructive leaders likely have a detrimental influence (Schyns & Schilling, 2012).
Abusive supervision, which is defined as “subordinates’ perceptions of the extent to which supervisors engage in the sustained display of hostile verbal and nonverbal behaviours, excluding physical contact” (Tepper, 2000, p. 178), is one form of destructive leadership. Tepper (2007) defined abusive supervision as stable over time and situations and explicitly involves a certain wilfulness by the leader. However, recent research challenges the static nature of abusive supervision by examining fluctuations in abusive leadership behaviours (Barnes, 2015; Courtright, 2016; Johnson, 2012). Notably, abusive supervisor behaviours were shown to vary due to a lack of sleep or work-family conflict (Barnes, 2015; Courtright, 2016). Introducing the concept of justice variability, Matta et al. (2017) argued that consistent and predictable (un)fair treatment by the supervisor is associated with less employee stress than variable fair treatment. Because subordinates often interact with their supervisor daily, unpredictable disrespectful and unfair supervisor behaviours might be particularly devastating to their self-conviction of competence (Duffy et al., 2002; Meier et al., 2012). To capture the impact of this variability on occupational self-efficacy and well-being, we examined changes in abusive supervision over time.
Furthermore, Mitchell and Ambrose (2007) distinguished two facets of abusive supervision: active and passive acts of abuse. We focused on active interpersonal abuse in our study, because these actions seem more tangible and easier to observe as opposed to passive forms. Thus, we considered changes in active abusive behaviour by the supervisor to be a significant stressor for subordinates (Lee & Brotheridge, 2011; Meier et al., 2012; Semmer et al., 2007).
Unsurprisingly, plenty of research links abusive supervision to various detrimental outcomes for subordinates’ well-being and health. Empirical findings illustrate that subordinates of abusive supervisors show higher levels of anxiety, emotional exhaustion, depression, and burnout (see Martinko et al., 2013; Schyns & Schilling, 2012), as well as lower life and job satisfaction, higher turnover intentions, and more work-family conflict (Bowling & Michel, 2011; Carlson et al., 2011; Lin et al., 2013; Schyns & Schilling, 2012; Tepper, 2000, 2007).
The mediating role of occupational self-efficacy
In the work context, a central source for one’s self-esteem is occupational self-efficacy (Judge & Bono, 2001; Loeb et al., 2016; Schyns & Collani, 2002). Occupational self-efficacy is defined as the belief in one’s ability to master occupational challenges and to pursue a career successfully independent from a specific occupation (Bandura, 1997; Higgins et al., 2008). People with high occupational self-efficacy are more likely to shape their work environment actively, which is shown to have a positive impact not only on their performance but also their psychological well-being (Tims et al., 2014).
In the COR framework, Hobfoll (2012) highlighted occupational self-efficacy as a key personal resource in the context of work that is positively related to well-being and favourable organisational outcomes (Avy et al., 2010; Heuven et al., 2006; Judge & Bono, 2001; Schyns & Collani, 2002) and negatively related to strain (Mohr et al., 2006). Given that occupational self-efficacy as a personal resource is susceptible to change (Fan et al., 2012; Perko et al., 2014), it can be influenced by leadership behaviours. Accordingly, Nielsen and Munir (2009) and Perko et al. (2014) suggested supportive supervisor behaviours, such as transformational leadership, are a work-related resource that fosters occupational self-efficacy, which stimulates affective well-being and lowers depressive symptoms.
Semmer and colleagues (2007) emphasised in their conceptualisation of the SOS model that appreciation and social support can be powerful resources that provide a boost to the self. However, it is not only positive social interactions that shape employees’ self-concepts and, consequently, their well-being (van Knippenberg, 2000). Based on social identity theory, van Dick and Haslam (2012) argued that the experience of stress in the workplace depends on an individual’s identification with a group, unit, or organisation. Toxic interactions with a supervisor have the potential to undermine social identification processes and thus impact subsequent well-being and health (Meier et al. 2012; Semmer et al., 2007). Here, the SOS model links to the COR (Hobfoll, 2012) assumption that a threat to a central personal resource, such as occupational self-efficacy, elicits the subjective experience of stress. According to the model, people’s self-esteem is seriously threatened through the experience of deficiencies regarding their competencies or personal attributes (Semmer et al., 2007). Moreover, because social inclusion and acceptance is a basic motivation for most individuals, verbal attacks and disrespect from significant others can damage the self-evaluation of one’s abilities and skills (Meier et al., 2012). Thus, the lack of appreciation and the permanent exertion of hostile behaviour by the leader make abusive supervision a primary threat to the organisational self-esteem of their subordinates (Rafferty & Restubog, 2011). Not only does an abusive leader fail to provide feedback and recognition or to assign important tasks, but also they actively disrespect subordinates’ competency (sample item: “My boss tells me I’m incompetent”). Hence, abusive leadership threatens the core of the subordinates’ occupational self-efficacy convictions. Indeed, the results of a study on social undermining by supervisors (Duffy et al., 2002) suggest that abusive supervision likely lessens occupational self-efficacy. We extended this proposed relationship by investigating occupational self-efficacy as a mediator that consequently reduces subordinate well-being and health.
In our study, we focused on work engagement and depressive symptoms as relevant outcomes because they are positive and negative indicators of psychological well-being and health as defined by the WHO (1946). Work engagement is a positive work-related state of mind where people feel energetic and confident that they can manage work demands (Schaufeli et al., 2006). Within the work context, Hobfoll (2012) explicitly stressed the importance of key personal resources, such as occupational self-efficacy, for work engagement, a proposition that is empirically supported by various studies (e.g., Salanova et al., 2011; Xanthopolou et al., 2013).
In contrast, experiencing depressive symptoms, such as reduced appetite, loss of energy, or loss of interest, are strong signs of psychological strain, regardless of whether they meet the criteria for a formal mental health diagnosis (Perko et al., 2014). Based on the conceptualisation of occupational self-efficacy as a resource, Perko et al. (2014) assumed a negative relationship of self-efficacy with depressive symptoms. Their results show that occupational self-efficacy fully mediated the relationship between transformational leadership and depressive symptoms.
The two chosen indicators of well-being represent positive motivational aspects as well as psychological health impairment. Therefore, we assumed that abusive supervision as a threat to one’s occupational self-efficacy has detrimental effects on well-being evidenced by decreased work engagement and increased depressive symptoms. Furthermore, based on the justice variability concept (Matta et al., 2017) and the findings regarding the fluctuations of abusive supervision over time and situations within a supervisor (Barnes et al., 2015; Courtright et al., 2016), we analysed whether changing abusive supervision across time shows a relationship to changes in self-efficacy and subsequent well-being.
H1a: Abusive supervision shows a negative indirect relationship with work engagement via occupational self-efficacy.
H1b: Abusive supervision shows a positive indirect relationship with depressive symptoms via occupational self-efficacy.
Gender in the relationship between abusive supervision, self-efficacy, and well-being
Einarsen and Nielsen (2015) highlighted that gender differences in the relationship between hostile behaviours and consequences for subordinates are seldom considered despite their potential importance. The empirical evidence is mixed concerning whether abusive supervisors provoke different health and well-being reactions in female and male subordinates: On the one hand, women were found to report slightly more emotional exhaustion and turnover intention as well as lower state self-esteem in response to abusive supervision (Burton & Hoobler, 2006; Wang et al., 2016). On the other hand, Einarsen and Nielsen (2015) showed that exposure to hostile behaviours was only associated with symptoms of distress for male participants. Similarly, Liu et al. (2008) found that men exhibit more negative emotions than women in response to general interpersonal conflict at work. Likewise, Hoel et al. (2004) showed that the relationship between bullying and health was stronger for male than female victims.
To understand these contradictory findings regarding gender in the relationship between abusive supervision and well-being, it seems worthwhile to re-examine occupational self-efficacy. Some empirical results showed that men report higher occupational self-efficacy than women (Abele & Spurk, 2009; Loeb et al., 2016); whereas, others found no significant differences (Hartman & Barber, 2020). Expectation states theory offers a theoretical basis to examine potential gender differences in self and other expectations regarding competency and influence (Ridgeway, 2001; Stempel & Rigotti, 2018). According to this theory, men are ascribed a higher status role as compared to women in the workplace and thus not only receive more favourable evaluations regarding competency and influence but also internalise these advantageous expectations to a greater extent (Ridgeway, 2001). Hence, one could assume that men should have more positive convictions regarding their occupational self-efficacy as compared to women (Ridgeway, 2001).
Unsurprisingly, the consequences of these expectations regarding the abilities and competencies of women and men can be evidenced in daily work life. Empirical findings show that men are allocated more challenging tasks at work (De Pater et al., 2009; De Pater et al., 2010) and receive more gratification in terms of bonuses, payment, and promotion (Amanatullah & Tinsley, 2013; Kulich et al., 2011). Regarding self-concept, men show higher work centrality (Isaksson & Johannsson, 2000) and career aspirations (Chevalier, 2007; Hartman & Barber, 2020; Ng & Feldman, 2008). Such gender differences regarding work performance expectations likely impact men’s and women’s self-assessment of occupational self-efficacy. Thus, on the one hand, men’s presumably stronger self-expectations regarding occupational efficacy might be a valuable resource that stimulates circles of resource gain as postulated by COR theory (Hobfoll, 2012). On the other hand, precisely this more pronounced occupational self-efficacy could make them more susceptible to threats, because resource threat or loss is more salient than gain (Hobfoll, 2012). Based on the SOS model and expectation states theory (Semmer et al., 2007), we assumed that the disrespect of the supervisor regarding the competencies and abilities pose a greater threat to the occupational self-efficacy beliefs of men than of women, subsequently impacting their well-being and health.
H2: The indirect effect from abusive supervision via self-efficacy on (a) work engagement and (b) depression is stronger for men than for women.
To test our hypotheses, we conducted a study with two separate points of measurement for the independent and the dependent variables. Due to the large sample size, we were able to apply a full SEM model (see Figure 1).
Figure 1
Conceptual Model of the Proposed Moderated Mediation.
Method
Participants and procedure
The sample came from an international project on leadership and well-being (Rigotti et al., 2014). To investigate the impact of leadership on well-being, the study targeted individuals in Germany who have frequent (daily or at least weekly) contact with their immediate supervisors. Following the Helsinki Declaration, we assured the voluntary nature of participation in the study and the confidentiality of the data. Every participant was matched based on an individual and a team code.
In total, we invited 2,273 subordinates from public (municipalities) and private organisations (banks and medium-sized companies) to participate in 2012 (T1). In total, 1,405 individuals took part, resulting in a response rate of 61.8%. Six months later (T2) 2,273 questionnaires were sent, and 1,203 individuals responded, resulting in a response rate of 52.9%. The participants had the option to complete written questionnaires (10.2%) or use the online format via a link in an email (89.2%). We classified 709 participants (76.4%) as working in the private sector.
We excluded participants who did not participate in T1 or had missing data, and only those who indicated having the same direct supervisor as during T1 remained in the sample, resulting in a final sample of 928 subordinates who participated at both points of measurement. These individuals belonged to 188 teams. Of these 928 individuals, 726 (78.2%) were female and 202 (21.8%) were male. Two thirds of the participants (64.7%) had a female leader and one third had a male leader (35.3%).
The participants were, on average, 40.5 years old (SD = 9.93) and had worked for their current supervisor for an average of 4.50 years (SD = 3.95) and their current employer for an average of 15.20 years (SD = 8.89). The majority held a permanent contract (90.9%), and the average working hours per week were 37.18 (SD = 4.39), with women working on average 36.59 hours (SD = 4.50) and men 38.47 hours (SD = 4.19) per week.
Dropout analysis on all employed variables revealed that participants who dropped out of the sample at T2 reported higher levels of abusive supervision (t = 1.98, p < 0.05) at T1, were older (t = 1.98, p < 0.05), filled out the questionnaire online (χ2 = 6.65, p < 0.05), were more likely from the public sector (χ2 = 8.33, p < 0.01), and were female (χ2 = 7.44, p < 0.01).
Measures
Abusive supervision
We measured abusive supervision using a shortened 5-item version of Tepper’s (2000) abusive supervision scale developed by Mitchell and Ambrose (2007), which reflects active, and therefore more easily perceivable, aspects of abusive supervision than the 15-item version. Participants were asked to indicate on a 7-point scale how much they agree with statements such as “My boss tells me I’m incompetent” (1 = strongly disagree; 7 = strongly agree). The internal consistency for the scale was αT1/T2 = 0.92/0.91.
Occupational self-efficacy
To measure occupational self-efficacy, we employed a short 6-item scale by Rigotti, Schyns, and Mohr (2008). Participants indicated on a 7-point scale how much they agree with statements like “I meet the goals that I set for myself in my job” (1 = strongly disagree; 7 = strongly agree). The internal consistency was αT1/T2 = 0.79/0.82.
Work engagement
Work engagement was measured using six items from the Utrecht Work Engagement Scale by Schaufeli et al. (2006) capturing the core components of vigour and dedication (Mauno et al., 2007; αT1/T2 = 0.92/0.94). Participants had to indicate on a 7-point scale how often a certain behaviour occurred (0 = never; 6 = every day). A sample item for vigour is “I am enthusiastic about my job,” and for dedication, “My job inspires me.”
Depressive symptoms
To measure depressive symptoms, we used the 12-item Major Depression Inventory validated by Bech et al (2001). Reflecting on the past 2 weeks, participants were asked to indicate on a 6-point scale how often they felt “low in spirits or sad” (0 = at no time; 5 = all the time). The scale covers symptoms including changes in appetite (reduced or increased) and energy level (subdued or restless), with two polar items in each case. Here, only the item with the higher value is used to calculate the overall score for depressive symptoms. The internal consistency was αT1/T2 = 0.92/0.91.
Gender
The gender of the participants and their respective supervisors was coded 0 for females and 1 for males. The information on the gender of the supervisor was included as a control variable because of its potential impact on the supervisor–subordinate relationship.
Controls
Age was included as a control variable because it is a relevant variable for self-efficacy (Molter et al., 2013), as well as for well-being at work (Mauno et al., 2013). Additionally, we included the variable sector (public = 0 and private = 1) as a control in our analysis, because past research on gender differences in occupational stress research (e.g., Garcia-Retamero & López-Zafra, 2006; Stuhlmacher & Poitras, 2010) highlighted the importance of the occupational context. Because abusive supervision is based on the perceptions of the subordinates, we also controlled for tenure with the current supervisor (“How long have you been working under the supervision of this leader?”) to consider the length of the leader-subordinate relationship.
Analyses
We tested all hypotheses employing SEM using MPlus. To account for the nested data structure with individuals being grouped into teams, we used TYPE=COMPLEX in all subsequent analyses. This method accounts for the nested data structure by calculating adjusted standard errors using a so-called sandwich estimation procedure (Zeileis et al., 2020). Relationships among variables were modelled using latent difference scores (cf. MacKinnon, 2017) following the syntax examples provided by Selig and Preacher (2009). We used latent difference scores for the mediation model to (a) map the dynamic over time, foregoing the ambiguous options for mediational models using two measurements in time, and (b) follow recent calls that abusive supervision is fluctuating across time and situations (Barnes et al., 2015; Courtright et al., 2016). All variables were included as latent constructs represented by single items as manifest indicators for both times of measurement. Additionally, we allowed the error variances of the same items to correlate across time points.
We regressed the latent difference score of occupational self-efficacy on the latent difference score of abusive supervision, representing the a-path in the mediation model. Latent differences scores of work engagement as well as depressive symptoms were regressed on the latent difference score of occupational self-efficacy, representing the two b-paths in the model. The direct effect was also included in the model. Furthermore, we controlled for age, sector, the gender of the supervisor, and time spent working together with the supervisor.
In our second two hypotheses (H2a/H2b), we proposed the gender of subordinates moderated abusive supervision and occupational self-efficacy. For this model, we introduced the interaction between the latent change score of abusive supervision and gender as a latent variable as a further predictor for the latent difference scores of occupational self-efficacy. Due to the random element of the latent interaction between the latent difference score of abusive supervision and gender, bootstrapping and standardised estimates were not available, but we present the 95% confidence intervals of indirect effects.
We display correlations, means, and standard deviations in Table 1. Abusive supervision correlates significantly negatively with occupational self-efficacy and work engagement and positively with depressive symptoms within, as well as between, time points. To check for differences in the absolute reporting of abusive supervision and stress responses, we conducted a correlational analysis. Correlations indicated that men reported slightly higher occupational self-efficacy as compared to women (T1: r = 0.07, p < 0.05; T2: r = 0.08, p < 0.05) but lower work engagement, at least at T2 (r = –0.08, p < 0.05). Significant differences did not appear between men and women in their ratings of depressive symptoms or abusive supervision, neither did we find differences for male and female abusive leaders.
Table 1
Means, Standard Deviations, Correlations, and Internal Consistencies for all Variables.
| MALES | FEMALES | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | |||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| M | SD | M | SD | |||||||||||||||||
| 1 | Abusive Superv. (T1) | 1.31 | 0.85 | 1.30 | 0.71 | (.92) | ||||||||||||||
| 2 | Abusive Superv. (T2) | 1.38 | 0.92 | 1.32 | 0.72 | .57** | (.91) | |||||||||||||
| 3 | Self-Efficacy (T1) | 5.14 | 0.84 | 5.01 | 0.85 | –.10** | –.07* | (.79) | ||||||||||||
| 4 | Self-Efficacy (T2) | 5.21 | 0.91 | 5.05 | 0.86 | –.09** | –.14** | .70** | (.82) | |||||||||||
| 5 | Work Eng. (T1) | 3.21 | 1.21 | 3.38 | 1.20 | –.19** | –.21** | .43** | .40** | (.92) | ||||||||||
| 6 | Work Eng. (T2) | 3.03 | 1.30 | 3.28 | 1.24 | –.15** | –.21** | .38** | .48** | .81** | (.94) | |||||||||
| 7 | Depress. Symp. (T1) | 0.95 | 0.83 | 1.06 | 0.81 | .20** | .18** | –.40** | –.38** | –.41** | –.37** | (.92) | ||||||||
| 8 | Depress. Symp. (T2) | 1.02 | 0.88 | 1.06 | 0.80 | .15** | .26** | –.36** | –.45** | –.37** | –.44** | .65** | (.91) | |||||||
| 9 | Gender Subordinatea | – | – | – | – | .01 | .03 | .07* | .08* | –.06 | –.08* | –.06 | –.02 | – | ||||||
| 10 | Gender Leadera | – | – | – | – | –.03 | .02 | .07* | .05 | –.04 | –.03 | –.02 | –.02 | .09** | – | |||||
| 11 | Age (T1) | 39.86 | 10.77 | 40.54 | 9.93 | –.00 | –.03 | .05 | .08* | .04 | .07* | .01 | –.04 | –.14** | .05 | – | ||||
| 12 | Sectorb | – | – | – | – | .07* | .02 | –.18 | –.15** | –.08* | –.07* | .16** | .14** | .03 | –.11** | –.12** | – | |||
| 13 | Tenure with leader | 3.86 | 3.42 | 4.58 | 4–52 | .01 | .00 | .00 | .02 | .02 | .03 | –.03 | –.01 | –.07* | –.02 | .26** | –.13** | – | ||
| 14 | Paper-Pencil | – | – | – | – | .08* | .07 | –.15** | –.13** | –.22** | –.24** | .09** | .09** | –.00 | .10** | .04 | .47** | .01 | – | |
[i] Note: N = 866–928, a female = 0 and male = 1. b public = 0 and private = 1. * p < 0.05; ** p < 0.01.
We tested for measurement invariance between men and women and between time points (cf. Edwards et al., 2018). For all variables under study, we compared an unconstrained model with freely estimated factor loadings with a model constraining the factor loadings to be equal. If the constrained model does not show a worse fit, we can conclude that constructs have the same meaning (i.e., metric invariance; Schmitt & Kuljanin, 2008) across time or groups. Due to the non-normal distribution of variables, we used MLMV as the estimation method and accordingly employed the DIFFTEST option in MPlus to test for significance in Δχ2. The results of these analyses are presented in Table 2.
Table 2
Measurement Invariance Between Men and Women and T1/T2 measures.
| χ2 | DF | Δχ2 | ΔDF | TLI | CFI | ΔCFI | RMSEA | ||
|---|---|---|---|---|---|---|---|---|---|
| Abusive Supervision | |||||||||
| T1 (Men <> Women) | Unconstrained | 34.44 | 14 | .919 | .943 | .06 | |||
| Measurement weights | 36.71 | 19 | 7.91 | 5 | .948 | .951 | .008 | .05 | |
| T2 (Men <> Women) | Unconstrained | 32.14 | 14 | .945 | .962 | .05 | |||
| Measurement weights | 34.99 | 19 | 4.96 | 5 | .965 | .966 | .004 | .04 | |
| T1 <> T2 | Unconstrained | 135.15 | 34 | .878 | .907 | .06 | |||
| Measurement weights | 134.89 | 39 | 3.89 | 5 | .899 | .912 | .005 | .05 | |
| Occupational Self-efficacy | |||||||||
| T1 (Men <> Women) | Unconstrained | 41.35 | 23 | .964 | .973 | .04 | |||
| Measurement weights | 44.94 | 29 | 3.03 | 6 | .975 | .976 | .003 | .05 | |
| T2 (Men <> Women) | Unconstrained | 48.93 | 23 | .971 | .977 | .05 | |||
| Measurement weights | 53.42 | 29 | 4.37 | 6 | .978 | .979 | .002 | .04 | |
| T1 <> T2 | Unconstrained | 402.75 | 53 | .839 | .871 | .09 | |||
| Measurement weights | 417.57 | 59 | 10.11 | 6 | .852 | .867 | .004 | .09 | |
| Work Engagement | |||||||||
| T1 (Men <> Women) | Unconstrained | 161.88 | 23 | .922 | .940 | .12 | |||
| Measurement weights | 165.08 | 29 | 2.96 | 6 | .939 | .941 | .001 | .10 | |
| T2 (Men <> Women) | Unconstrained | 137.50 | 23 | .954 | .965 | .11 | |||
| Measurement weights | 146.69 | 29 | 4.39 | 6 | .963 | .964 | .001 | .10 | |
| T1 <> T2 | Unconstrained | 644.82 | 53 | .885 | .908 | .12 | |||
| Measurement weights | 664.04 | 59 | 8.97 | 6 | .895 | .906 | .002 | .11 | |
| Depressive Symptoms | |||||||||
| T1 (Men <> Women) | Unconstrained | 195.83 | 79 | .935 | .943 | .06 | |||
| Measurement weights | 202.70 | 89 | 14.24 | 10 | .944 | .945 | .002 | .05 | |
| T2 (Men <> Women) | Unconstrained | 188.49 | 79 | .929 | .938 | .06 | |||
| Measurement weights | 200.77 | 89 | 20.73* | 10 | .936 | .936 | .002 | .05 | |
| T1 <> T2 | Unconstrained | 716.55 | 169 | .865 | .880 | .06 | |||
| Measurement weights | 718.25 | 179 | 5.75 | 10 | .875 | .882 | .002 | .06 |
[i] Note: TLI = Tucker-Lewis Index, CFI = Comparative Fit Index, RMSEA = Root Mean Square Error of Estimation. * p < 0.05, ** p < 0.01.
The Δχ2 difference test was only significant for depressive symptoms in the comparison of men and women at T2. As the Δχ2 difference test has been criticised for being sensitive to sample size (Cheung & Rensvold, 2002), we also used ΔCFI for the evaluation, with values higher than 0.01 defined as a significant decrease of fit. In no case was this threshold surpassed. Hence, we can claim metric invariance across time points and for the comparison between men and women for all measures (see Table 2).
Results
Indirect effects
H1a and H1b claimed an indirect effect of changes in abusive supervision on changes in work engagement and depressive symptoms via changes in occupational self-efficacy. As shown in Table 3, we found significant indirect effects for work engagement but not for depressive symptoms (p = 0.056), which lends support for hypothesis H1a but leads to the rejection of H1b.
Table 3
Results of Indirect and Conditional Indirect Effects of the SEM Model.
| EFFECTS | INDIRECT EFFECTS MODEL H1A/H1B | CONDITIONAL INDIRECT EFFECTS MODEL H2A/H2B |
|---|---|---|
| Paths related to hypotheses | ||
| a-path: ΔAS → ΔOS | –.11(.05)* | –.14(.05)* |
| b-path1: ΔOS → ΔWE | .43(.07)** | .42(.07)** |
| b-path2: ΔOS → ΔDS | –.34(.07)** | –.33(.07)** |
| Interaction ΔAS (T1) x Sex → ΔOS | x | .11(.11) |
| Direct effects | ||
| ΔAS → ΔWE | –.06(.05) | –.07(.05) |
| ΔAS → ΔDS | .19(.06)** | .20(.06)** |
| Indirect effects | ||
| On Work Engagement [CI95%] | –.05(.02)* [–.09; –.00] | –.06(.03)* [–.11; –.01] |
| On Depressive Symptoms [CI95%] | .04(.02)+ [–.00; .08] | .05(.02)* [.00; .09] |
| Total effects | ||
| On Work Engagement [CI95%] | –.11(.04)* [–.20; –.03] | –.12(.05)** [–.21; –.03] |
| On Depressive Symptoms [CI95%] | .23(.06)** [.11; .35] | .24(.06)** [.12; .37] |
[i] Note: AS = Abusive Supervision, OS = Occupational Self-Efficacy, WE = Work Engagement, DE = Depressive Symptoms, ** p < 0.01, * p < 0.05, + p < 0.10, all analyses included the following control variables: sex of supervisor, sector (private vs. public), paper-pencil vs. online, age, tenure with leader. We also ran the models excluding the control variables, and the core results were the same.
Conditional indirect effects model
To examine our second hypothesis, whether gender moderated the indirect effect of abusive supervision on work engagement and depressive symptoms via occupational self-efficacy, a moderated mediation model with gender as the moderator between the latent difference score of abusive supervision and the latent difference score of occupational self-efficacy was tested. The results of these analyses are also displayed in Table 3. Our specified model shows that the indirect effects between abusive supervision and work engagement, as well as depressive symptoms, were significant for men and for women in the sample. No interaction effect could be found.
Discussion
Occupational self-efficacy is an important personal resource that positively influences well-being and shields against psychological strain (Heuven et al., 2006; Hobfoll, 2012; Mohr et al., 2006). However, a question arises as to the implications for subordinates’ well-being when an abusive supervisor seriously threatens their self-efficacy convictions as postulated in the SOS model (Semmer et al., 2007). In line with our hypotheses, the findings show that abusive supervision is linked to reduced work engagement and increased depressive symptoms. The fact that this relationship can partly be explained due to a lowered sense of occupational self-efficacy mirrors and additionally expands theory regarding positive leadership behaviours (Nielson & Munir, 2009; Perko et al., 2014; Salanova et al., 2011). We not only show that occupational self-efficacy as a mediating mechanism is translatable to destructive leadership but also underline its importance for well-being. Although plenty of research deals with circles of resource gain stimulated by constructive leadership enhancing self-efficacy (Heuven et al., 2006), we relied on the SOS model by Semmer et al. (2007) to investigate loss circles caused by abusive supervisors. Overall, the menace of occupational self-efficacy beliefs, through a lack of appreciation and the exhibition of disrespectful behaviours by supervisors, highlight a psychological mechanism by which negative leadership behaviours affect subordinates’ health and well-being.
For depressive symptoms, the indirect effect slightly failed the significance level; whereas, the direct effect was significant. For work engagement, the direct effect was no longer significant in the mediation model. However, when introducing followers’ gender as a moderator between abusive supervision and occupational self-efficacy, the indirect effect is significant for work engagement as well as depressive symptoms. On the one hand, this pattern of results suggests that there might be further mechanisms besides a deterioration of occupational self-efficacy to explain the link between abusive supervision and well-being. On the other hand, even though followers’ gender did not moderate between abusive supervision and occupational self-efficacy, taking gender into the equation still seems to be important.
In our study, we could not find that the indirect effect of abusive supervision on work engagement and depressive symptoms through occupational self-efficacy is conditional upon the gender of the subordinate. Neither did women and men differ in reporting abusive supervision in general nor did they report differences based on the leader’s gender. This result suggests that gender dissimilarity as indicated by Park et al. (2018) does not play a moderating role in our proposed mediation. Nevertheless, in line with expectations state theory (Ridgeway, 2001), we found slightly higher occupational self-efficacy and more working hours per week for men than for women, which might indicate a divergent status of work-related issues as suggested by previous research on work centrality (Isaksson & Johannsson, 2000) and career aspirations (Chavelier, 2007; Hartman & Barber, 2020; Ng & Feldman, 2008). However, these differences seem not to render the effects of abusive supervision distinct for men and women. However, considering potential gender differences of job stressors and resources might still be important to further theory development in the context of leadership and gender, because current gender theories often fall short to make clear predictions regarding destructive leadership (Stempel & Rigotti, 2018).
Limitations and implications for future research and practice
Dropout analyses revealed some non-random attrition. By including age, type of assessment, sector, and gender as control variables, we attenuated potential biases caused by a systematic dropout. Regarding the fact that loyal respondents, who provided data on both points of measurement, reported less abusive supervision might have led to a reduction in the variance of abusive supervision. Hence, our reported effects might even be underestimated.
The generalisability of our findings might also be limited because our sample was dominated by female participants. Interestingly, a meta-analysis suggested that the proportion of females in the sample does not seem to moderate the relationship between leadership and mental health (Montano et al., 2017). Nevertheless, it would be interesting to investigate the second stage moderation of gender by exploring potential gender differences in the relationship between occupational self-efficacy and well-being. For that purpose, more gender-balanced samples across various occupational backgrounds would be necessary.
Concerning our research design, one drawback of our study is using only two instead of three points of measurement for the mediation model. However, we used latent difference scores to model our independent, mediator, and dependent variables to illustrate change over time. In classical cross-lagged panel designs, the change is usually only modelled for the dependent variable and not for the independent variable. By looking at the effects of changes in abusive leadership behaviour, we opted for explicitly considering that abusive leadership behaviour can change over time.
Potentially, the effects of abusive supervision on well-being are more proximal in time as recent research on the fluctuating nature of abusive supervision suggests (Barnes et al., 2015; Courtright et al., 2016). Thus, future studies should address these time issues by, for example, investigating the static versus fluctuating nature of abusive supervision. Further interesting questions for future research concern the reverse effects of (changes in) health indicators on occupational self-efficacy and abusive supervision. Participants who feel depressed or less motivated might (a) experience less occupational self-efficacy and/or (b) elicit abusive behaviour by their supervisor. Moreover, to reduce same-source biases, it could be interesting for future research to include a variety of information sources (e.g., leader ratings, organisational health statistics) and focus on potential differences between them.
Because cultural values and norms might play a role in leadership perceptions and/or gender stereotypes, it is important to highlight that our sample stems from Germany. According to Hofstede et al. (2010), Germany scores high on masculinity, which is characterised by high competitiveness and work centrality and might pronounce the importance of occupational self-efficacy. More cross-cultural research is needed to investigate the impact of culture on differing mechanisms for male and female subordinates in the relationship between abusive supervision and well-being.
Implications for practice can be considered on several levels. First, our study findings highlight the importance of detecting and shielding against abusive supervision in the workplace to prevent detrimental impacts on well-being. Supervisor training (Gonzales-Morales et al., 2018) and opportunities for subordinates to safely voice criticism or to ask for support could help shield against the threat of abusive leaders. Second, we found occupational self-efficacy to be an important personal resource regardless of gender and to be positively related to well-being (Rigotti et al., 2008). Because men report slightly more occupational self-efficacy than women, it seems important to raise awareness regarding gendered perceptions of occupational self-efficacy. Thus, to enhance occupational self-efficacy for all employees, it seems crucial to establish gender-sensitive feedback processes as well as fair evaluation and promotion criteria.
Conclusion
We identified occupational self-efficacy as a mediator of the relationship between abusive supervision and well-being. In line with the SOS model (Semmer et al., 2007), abusive supervisor behaviours impose a considerable threat to subordinates’ occupational self-efficacy and increase their experience of stress. To further understand the underlying mechanisms, more research is needed to address theoretical and time-related issues and to consider individual characteristics.
Additional Files
The additional files for this article can be found as follows:
Transparency Table
Data transparency table RE-SU-LEAD, June 2020. DOI: https://doi.org/10.16993/sjwop.124.s2
Competing Interests
The authors have no competing interests to declare.