Lead+D Lab research synthesis · 29 studies digested · 2026-07-28
This systematic review synthesizes evidence on the relationship between abusive, destructive, and toxic supervision and employee turnover, drawing on two meta-analyses (Schyns & Schilling, 2013; Zhang & Liao, 2015), the foundational Tepper (2000) primary study, and Tepper et al.'s (2017) Annual Review, contextualized by 25 recent primary studies. Abusive supervision is reliably associated with turnover intention (r roughly .30 to .31). The literature, however, overwhelmingly measures intention rather than behavior: meta-analytic actual-turnover effects are near zero (r = .039, ns), against Tepper's (2000) single-study beta of -.37. Organizational justice, especially interactional justice, is the dominant mediating mechanism, with social exchange, resource depletion, and emotional exhaustion also implicated. Power distance critically moderates the relationship (significant in low-power-distance North America, non-significant in high-power-distance Asia), raising generalizability concerns. The evidence base is predominantly cross-sectional, self-reported, and correlational, with negligible interventional data. Priorities for behavioral-outcome, cross-cultural, and intervention research are identified.
Method: OpenAlex citation-chain sweep + keyword search from 5 canonical seeds. 625 records pooled → 215 included after title screening (410 excluded). Four meta-analyses / reviews digested in full text; 25 core primary studies at abstract level. No statistic reported here is un-sourced.
| Statistic | What it says | Source |
|---|---|---|
| r = .30, 95% CI [.27, .32]; 13 effect sizes, N = 5,950 | Abusive supervision is moderately associated with turnover intention | Zhang & Liao 2015 |
| r = .313, 95% CI [.290, .335]; k = 11, N = 6,034, Qw = 131.77 (p < .001) | Destructive leadership correlates with turnover intention (raw correlation, with large heterogeneity) | Schyns & Schilling 2013 |
| standardized logistic beta = -.37, p < .01; Δχ²(2) = 11.71, p < .01 | The single primary study of actual voluntary turnover found abusive supervision predicts quitting six months later | Tepper 2000 |
| r = .039, 95% CI [-.069, .146]; k = 2, N = 333, ns | Meta-analytic actual/objective turnover is near zero and non-significant | Schyns & Schilling 2013 |
| r = .222 (Tepper scale) vs r = .339 (other); Qb = 14.306, p < .001 | Effect size depends on instrument: the Tepper scale yields a smaller turnover-intention correlation than other measures | Schyns & Schilling 2013 |
| low-PD r = .26 [.23, .29] vs high-PD r = .04 [-.01, .10], CI includes 0 | Power distance moderates the link: significant in low-PD North America, non-significant in high-PD Asia | Zhang & Liao 2015 |
| incremental abusive-supervision effect non-significant after justice (full mediation) | Organizational justice fully mediated the abusive-supervision-to-turnover path | Tepper 2000 |
| attitudes toward leader r = -.571; CWB r = .377; turnover intention r = .313 | Strongest effect in the destructive-leadership meta-analysis was on attitudes toward the leader, exceeding turnover intention | Schyns & Schilling 2013 |
| approximately 10% of employees, roughly 3x higher in college athletics | Estimated population exposure to abusive supervision, with college athletics an outlier | Tepper et al. 2017 |
Two independent meta-analyses converge on a moderate positive association between abusive supervision and turnover intention. Zhang and Liao (2015) report r = .30, 95% CI [.27, .32], across 13 independent effect sizes and N = 5,950. Schyns and Schilling (2013) report r = .313, 95% CI [.290, .335], k = 11, N = 6,034, with substantial within-study heterogeneity (Qw = 131.77, p < .001). By magnitude, the turnover-intention effect sits between the AS effect on job satisfaction (r = -.336 to -.35) and the markedly stronger effects on attitudes toward the leader (r = -.571, the largest in the Schyns and Schilling synthesis) and counterproductive work behavior (r = .377).
The crucial caveat is that the literature overwhelmingly measures turnover INTENTION, not actual turnover behavior. The only primary study linking abusive supervision to objectively coded actual voluntary turnover is Tepper (2000), which reported a standardized logistic-regression beta of -.37 (p < .01; model chi-square change for abusive supervision plus mobility of 11.71, p < .01) over a six-month time lag in a sample of 119 quitters and 475 stayers. Yet when actual or objective turnover is pooled meta-analytically the effect collapses toward zero: Schyns and Schilling's organizational-performance composite, driven largely by Detert, Trevino, Burris, and Andiappan (2007), yields r = .039, 95% CI [-.069, .146], k = 2, N = 333, non-significant, and Detert et al. found no significant relationship with actual turnover. Zhang and Liao code no actual-turnover effect at all.
This intention-behavior gap is the central inferential problem of the review area. Intentions are robust and consistent, but the evidence that abusive supervision causes employees to actually leave is thin, drawn from very few studies, and methodologically fragile: Tepper's binary outcome likely captured some nonvoluntary departures and excluded intended quitters whose supervisors left first, compromising the power of even that single test. Plainly stated, we know a great deal about why subordinates want to quit and very little about whether they in fact do.
The dominant and best-supported mechanism is organizational justice, particularly interactional justice. Tepper (2000) found that the three justice dimensions fully mediated the abusive-supervision-to-voluntary-turnover path: the incremental effect of abusive supervision was non-significant once justice was controlled (Baron and Kenny), and interactional justice emerged as a comparatively strong predictor of turnover, commitment, and distress. Zhang and Liao (2015) show abusive supervision is strongly negatively related to all three justice forms, interactional r = -.51, procedural r = -.34, distributive r = -.31. Critically, however, neither meta-analysis quantitatively tested mediation: Zhang and Liao explicitly state they did not meta-analyze mediating variables and flag meta-path analysis as a future need, and Schyns and Schilling (2013) only propose (do not test) their mechanisms. Justice-as-mediation is therefore the only pathway with a formal turnover test in data; the rest are theoretically motivated.
The causal narrative, as assembled across the four sources, runs as follows. Abusive supervision signals disrespect and procedural unfairness, depressing justice perceptions (especially interactional), which erode job satisfaction (r = -.35), affective or organizational commitment (r = -.30), and engagement (r = -.29), which in turn raise turnover intention. Several complementary pathways are theorized from adjacent outcomes. Social exchange (Blau; Cropanzano and Mitchell) frames abusive supervision as a violation of the reciprocal exchange relationship, reciprocated through withdrawn commitment, OCB, and eventual exit intent. Conservation of resources and ego-depletion theory (Hobfoll) treats abusive supervision as a chronic stressor draining personal resources such as psychological capital, self-esteem, and energy, prompting protective withdrawal; recent primary studies operationalize this through ego depletion (Chinese hospitality supervisor incivility; Chinese rural kindergarten teachers) and psychological capital (Chinese factory workers). The stressor-emotion model (Spector and Fox) casts abusive supervision as an extreme social stressor generating negative affect and emotional exhaustion, which Tepper (2000) showed partially mediated distress outcomes and which the nursing literature links to burnout and subsequent turnover intent. Finally, overspill (Burris et al., 2008) holds that negative feelings toward the supervisor generalize to negative attitudes toward the organization, helping explain commitment and turnover effects.
An honest assessment is warranted. Tepper, Simon, and Park (2017) criticize the piecemeal nature of mediation testing: studies examine one, rarely two, mechanisms at a time, so the relative explanatory power of justice versus social exchange versus conservation of resources versus emotion is essentially unknown. Justice is the empirically anchored pathway; the others are theoretically plausible but triangulated from outcomes other than turnover.
The single most consequential moderator is power distance. Zhang and Liao (2015) report that the abusive-supervision-to-turnover-intention link is significant in low-power-distance North America (r = .26, 95% CI [.23, .29]) but NOT significant in high-power-distance Asia (r = .04, 95% CI [-.01, .10], confidence interval includes zero). The implication is serious for generalizability: the field's headline turnover-intention effect may not transfer to high-power-distance Asian contexts, where subordinates may feel unable to exit, may develop learned helplessness (Schyns and Schilling, 2013, proposed but not tested), or may read hierarchical harshness through a paternalistic or collectivist lens. Two caveats temper this. First, Zhang and Liao proxied power distance by geographic region (Asia versus North America), conflating power distance with collectivism, paternalism, and other cultural dimensions, so the specific power-distance mechanism is inferred rather than isolated. Second, the finding sits in tension with corroborating cross-cultural signals: Vogel et al. (2015, cited in Tepper et al., 2017) found abusive-supervision effects on interpersonal justice were stronger in Anglo than Confucian clusters with power distance explaining the difference, and Mackey et al. (2015) report abusive supervision occurs more frequently in Asian than US samples. Higher exposure yet a weaker turnover response is a paradox the literature has not reconciled.
Several person-and-context moderators further qualify the effect (Zhang and Liao, 2015). Age attenuates harm: effects on anger (B = -.05) and emotional exhaustion (B = -.19) are weaker for older subordinates. Organizational tenure moderates commitment (B = -.07), job satisfaction, supervisor-directed deviance (B = -.07), and interactional justice (B = -.12). Time spent with the supervisor moderates anger, distributive and interpersonal justice, emotional exhaustion, and supervisor-directed deviance. Rating source matters sharply for supervisor-directed deviance: self-rated r = .53 versus supervisor-rated r = .11, with no overlap in confidence intervals, a clear common-source inflation signal.
Measurement instrument is a demonstrated moderator (Schyns and Schilling, 2013): for turnover intention, Tepper's abusive-supervision scale yields r = .222 versus r = .339 for other instruments (Q-between = 14.306, p < .001), attributed to specificity matching, since Tepper items carry a more personal connotation whereas other instruments are more job-relevant. Job mobility (Tepper, 2000) makes abusive supervision less detrimental for satisfaction and depression but did NOT moderate turnover. Tepper et al. (2017) add relational boundary conditions: being singled out (single-target) strengthens harm whereas co-targeting with peers buffers it, including intent to quit (Duffy et al., 2006); targets who engage in more upward hostility against the supervisor suffer less (Tepper et al., 2015); and supervisors who toggle between abusive and supportive behavior injure subordinates more than consistently hostile ones (Duffy et al., 2002; Lian et al., 2012). Finally, time lag showed no moderating effect on any abusive-supervision consequence in Zhang and Liao, and neither meta-analysis examined industry.
The leader-behavior construct is dominated almost completely by the Tepper (2000) 15-item abusive supervision scale (alpha = .90 in the original study, on a five-point frequency response from cannot-remember-ever to uses-very-often). Zhang and Liao (2015) made use of the Tepper scale or a short version an explicit inclusion criterion. Schyns and Schilling (2013) found it by far the most-used instrument, with a long tail of alternatives appearing in only a handful of studies each: petty tyranny (Ashforth, 1997), social undermining (Duffy et al., 2002), despotic leadership (De Hoogh and Den Hartog, 2008), destructive or tyrannical leadership (Einarsen et al., 2002), aversive leadership, unsupportive managerial behaviors, coercive power, supervisor verbal abuse, and supervisory abuse. The construct is operationalized as subordinates' perceptions of the sustained display of hostile verbal and nonverbal behaviors, excluding physical contact, reported by the single focal subordinate about the immediate supervisor, that is, perceptual and almost always single-source.
Turnover is measured with a striking asymmetry. Tepper (2000), the foundational study, measured ACTUAL voluntary turnover and not intention: two independent coders classified verbatim accounts of how the supervisor relationship ended into voluntarily quit versus terminated or supervisor left (inter-rater kappa = .89), coded binary (quit = 0, same supervisor at T1 and T2 = 1), across a six-month lag, and no turnover-intention scale such as the TIS-6 was administered. By contrast, the entire meta-analytic record on the turnover outcome codes self-reported turnover INTENTION (intention to quit, turnover intent, turnover intention). Actual or objective turnover appears meta-analytically only once, at the organizational level (Detert et al., 2007).
Common-method and single-source concerns are therefore systemic. Both meta-analyses draw on predominantly cross-sectional, self-report primary studies; Tepper (2000) mitigated this only partially through its time-lagged design, 15-factor confirmatory factor analysis, and significant moderation effects. The instrument-dependence of effect sizes is now established: the Tepper scale produces a meaningfully smaller turnover-intention correlation (r = .222) than other instruments (r = .339), so the chosen measure partly determines the reported effect. Tepper et al. (2017) add two further validity problems. First, a performance-outcome confound, because one form of abuse is delivered through performance ratings, so supervisor-rated performance may tap the same construct as subordinate-reported abusive supervision. Second, intersubjective agreement among subordinates does not establish objectivity, and supervisor self-reports are no more objective than subordinate reports; experimental studies rely on low-ecological-validity vignettes. The review explicitly calls for hard, objective measures of withdrawal (tardiness, absenteeism, actual turnover) and performance, which remain scarce.
The honest assessment is that there is almost no interventional evidence in this literature. The evidence base is overwhelmingly correlational and predictive rather than experimental or quasi-experimental, and neither meta-analysis nor the Annual Review evaluates a supervised intervention, such as supervisor selection, leadership training, or organizational change, shown to causally reduce abusive supervision or turnover. Tepper et al. (2017) note that the lone counter-result, an inverted-U between abusive supervision and creativity (Lee et al., 2013), is unreplicated, and they identify no tested intervention. Proposals in the mechanism literature, for example checks-and-balances against permissive cultures via the Padilla et al. (2007) toxic triangle, remain untested as remedies.
What does exist is a set of moderator-derived boundary conditions that attenuate the abusive-supervision effect and thus suggest intervention targets, though these are inferred rather than demonstrated. Job mobility buffers harm: abusive supervision is less detrimental for high-mobility subordinates on satisfaction and depression (Tepper, 2000), implying that internal labor-market access and exit options protect employees, although mobility did not moderate turnover itself. Co-targeting, being abused alongside peers rather than singled out, buffers individual effects including intent to quit (Duffy et al., 2006). Perceived organizational support alleviates the supervisor-incivility-to-turnover-intention link in a Chinese hospitality sample, though not the ego-depletion pathway. Workplace friendship weakens the mediated abusive-supervision-to-turnover-intention path through voice behavior in an Indian sample. Paradoxically, supportive-leadership stability matters: consistently hostile supervisors injure subordinates less than supervisors who toggle between abusive and supportive behavior (Duffy et al., 2002), suggesting predictability is itself protective. The reasonable, theory-consistent practitioner inference is to select against abusive traits, train interactional-justice and respectful-engagement behaviors, build psychological safety and credible exit or mobility options, and install structural checks against permissive cultures. This inference, however, is evidence-inferred, not evidence-tested. The field still needs pre-registered randomized and quasi-experimental designs showing that reducing abusive supervision reduces actual turnover, not merely stated intent.
Read as a body, the 25 recent core studies display four clear patterns. First, geography has shifted decisively away from the US samples that founded the field. There is heavy Asian representation (China: factory workers, rural kindergarten teachers, university faculty, hospitality; India: a northern-India voice-behavior study; Turkey: nurses; Indonesia: public organizations), an unusually strong Pakistan cluster (banking employees, call-centre workers, and banks), plus the UAE (hospitality frontliners), South Africa (manufacturing), and Canada (new-graduate nurses). Many of these studies are situated precisely in the high-power-distance Asian contexts where Zhang and Liao (2015) found the abusive-supervision-to-turnover link non-significant, yet they almost universally report significant intention effects, a tension the field has not reconciled and that most likely reflects region being a poor proxy for power distance coupled with cross-sample design differences. Second, industries concentrate in nursing and healthcare (Turkish and Canadian nurses, Indonesian hospitals, abusive-supervision-climate nurses), hospitality and frontline service (China, UAE), and banking (Pakistan), with education (kindergarten teachers, university faculty), manufacturing, and call centres recurring. Third, there is marked construct drift away from abusive supervision specifically: studies variously label the construct toxic leadership, despotic leadership, destructive or derailed leadership, supervisor incivility, workplace bullying, and corporate psychopathy, often measuring overlapping phenomena with different instruments, consistent with Schyns and Schilling's (2013) instrument-dependence finding and creating real comparability problems. Fourth, design is overwhelmingly cross-sectional self-report analyzed via regression or structural equation modeling, with only a minority using two- or three-wave time-lagged data, and virtually none measuring actual turnover behavior. The contemporary frontier is thus mechanism-rich (psychological capital, ego depletion, psychological empowerment, psychological-contract breach, organization-based self-esteem, voice behavior, resilience) and moderator-rich (resilience, workplace friendship, organizational support, ostracism), but methodologically narrow: more mediators and moderators are layered onto the same cross-sectional intention outcome rather than stronger designs or behavioral criteria being adopted.