Lead+D Lab research synthesis · leadership-anchored · 3-source triangulation · 2026-07-28
Objective: To synthesize the consequence network of bad management and toxic work environments, treating abusive, destructive, and despotic supervision as the spine and workplace bullying as a supplementary, environment-level signal. Methods: We integrated four abusive-supervision syntheses (Mackey et al., 2017, k=140; Zhang & Liao, 2015, k=119; Schyns & Schilling, 2013, k=57; Tepper, 2000), the Frazier et al. (2017) psychological-safety meta-analysis (k=136), three bullying meta-analyses (Nielsen & Einarsen, 2012; Verkuil et al., 2015; Nielsen et al., 2016), Hershcovis (2011) construct-reconciliation evidence, and field-wide metaBUS correlations. Screening was on definition and measure, not label. Results: Effects are strongest for relational and justice rupture (interpersonal justice ρ = −.66; leader-member exchange −.54), supervisor-directed deviance (ρ = .53), and emotional exhaustion (.36), and weakest for actual or objective turnover (near zero) and objective health. Psychological safety adds incremental variance beyond leadership. Bullying converges on mental health and sickness absence but its perpetrator is not necessarily the manager. Conclusion: The boss leaves the deepest scar on justice, deviance, and strain; behavioral withdrawal is under-supported and high power-distance generalizability remains thin.
The aggression-at-work literature suffers from severe jingle-jangle drift: abusive supervision, destructive and despotic leadership, workplace bullying, incivility, social undermining, and corporate psychopathy are label-proliferating constructs that all describe hostile interpersonal treatment at work. Hershcovis (2011) demonstrated that this proliferation is largely nominal. At the item level, the scales share literal content (insulting, silent treatment, rumor-spreading, derogatory remarks, and rudeness recur across measures), and across 25 pairwise construct-outcome comparisons only 7 were statistically distinct while 18 (72%) showed overlapping confidence intervals. The theoretically predicted superiority of the more severe constructs failed: abusive supervision was not significantly stronger than incivility on any outcome, and bullying exceeded incivility on only one (physical well-being). No construct showed a uniformly stronger pattern.
We therefore screened on definition and measure, not label. We treat the family as hostile leadership or workplace aggression, but preserve member distinctions only where a study operationally isolated the distinguishing feature (perpetrator identity, intensity, intent, frequency, power). The spine of this review is constructs whose harm source is the supervisor (abusive, destructive, despotic supervision); bullying, incivility, and social undermining enter only as a clearly labelled environment-level supplement and are never allowed to masquerade as leadership evidence. This follows Hershcovis's three screens (item overlap, whether the distinguishing feature was actually measured, outcome convergence) and her recommendation to code presumed distinctions as study-level moderators rather than separate constructs.
We also report two grains throughout. The abusive-supervision-specific grain uses the Tepper (2000) 15-item scale with the supervisor as perpetrator; the broader destructive-leadership umbrella pools petty tyranny, despotic, tyrannical, and aversive leadership instruments. They diverge systematically. For turnover intention, Schyns and Schilling (2013) found r = .222 under the Tepper scale versus r = .339 under other instruments (Qb = 14.306, p < .001), attributable to specificity matching: the Tepper items carry a personal connotation and correlate more with affectivity, well-being, leader-directed attitudes, and counterproductive behavior, whereas the broader instruments align more with resistance, justice, turnover, and performance.
| Construct | Definition | Typical measure | Source of harm | Role |
|---|---|---|---|---|
| Abusive supervision | Subordinates' perceptions of the extent to which supervisors engage in the sustained display of hostile verbal and nonverbal behaviors, excluding physical contact (Tepper, 2000, p. 178). | Tepper (2000) 15-item scale (or short forms), 5-point frequency anchors; mean alpha .92 across adaptations (Mackey et al., 2017). | Immediate supervisor (the boss). | core |
| Destructive leadership | Systematic behavior by a leader that violates the legitimate interests of the organization by undermining its goals, tasks, resources, and effectiveness (Einarsen et al., 2002). | Heterogeneous: petty tyranny (Ashforth, 1997), tyrannical/destructive leadership (Einarsen et al., 2002), aversive leadership (Bligh et al., 2007), coercive power (Elangovan & Xie, 2000). | Leader. | core |
| Despotic leadership | Self-serving, autocratic, and controlling leader behavior that dominates subordinates for personal rather than organizational ends (De Hoogh & Den Hartog, 2008). | Despotic leadership scale (De Hoogh & Den Hartog, 2008). | Leader. | core |
| Workplace bullying | A person repeatedly and over time exposed to negative acts (abuse, offensive remarks, ridicule, social exclusion) on the part of coworkers, supervisors, or subordinates (Einarsen, 2000). | Negative Acts Questionnaire (behavioral-experience) or self-labelling with definition; behavioral method yields larger effects (Nielsen & Einarsen, 2012). | Any organizational insider (perpetrator not necessarily the manager). | context |
| Workplace incivility | Low-intensity deviant acts, rude and discourteous verbal and nonverbal behaviors with ambiguous intent to harm (Andersson & Pearson, 1999). | Incivility scale (Cortina et al.). | Any organizational member. | context |
| Social undermining | Behavior intended to hinder, over time, the ability to establish and maintain positive relationships, work-related success, and favorable reputation (Duffy et al., 2002, p. 332). | Undermining scale (Duffy et al., 2002), separately identifying supervisor versus coworker source. | Supervisor or coworker. | context |
Shaded rows = the leadership spine of this review. Unshaded = related constructs used only as labelled context.
Strongest domain; the proximal fairness and trust violation, with corrected effects in the −.50 to −.66 range.
Abuse is read first as a fairness and exchange violation, and the relational-rupture effects are the largest in the entire matrix. Mackey et al. (2017) report interpersonal justice ρ = −.66 (k = 5), interactional justice −.55 (k = 7), procedural justice −.36, and distributive justice −.25, alongside leader-member exchange ρ = −.54, ethical leadership −.50, and perceived organizational support −.40. Zhang and Liao (2015) converge on interactional justice r = −.51, procedural −.34, distributive −.31, and the original Tepper (2000) study found interactional justice r = −.53, procedural −.48, and distributive −.39, with interactional justice emerging as the comparatively strong predictor of both turnover and distress. The independent metaBUS node confirms the magnitude: interpersonal justice r = −0.554, procedural −0.273, justice overall −0.238, and trust in management −0.139 (k_articles = 23). This is where the supervisor does the most measurable damage: the target's sense of fairness, trust, and standing in the relationship collapses before behavior or health do.
Very strong; supervisor-directed retaliation is the single largest behavioral consequence and the densest metaBUS evidence (k_articles = 38).
Retaliation aimed back at the supervisor is the dominant behavioral response. Mackey et al. (2017) report supervisor-directed (interpersonal) deviance ρ = .53 (k = 14, N = 5,223), counterproductive work behavior ρ = .41 (k = 6), organization-directed deviance ρ = .41 (k = 11), and interpersonal deviance ρ = .35. Zhang and Liao (2015) replicate the pattern: supervisor-directed deviance r = .51, organization-directed r = .38, interpersonal-directed r = .37, and Schyns and Schilling (2013) pool counterproductive work behavior at r = .377 (k = 19). The field-wide metaBUS node is both the strongest and the densest source of triangulation here: counterproductive behavior r = 0.431 (k_articles = 38), counterproductive behavior toward individuals r = 0.458 (k_articles = 20), and toward the organization r = 0.388 (k_articles = 16). The displaced-aggression and social-learning mechanisms proposed by Schyns and Schilling (2013) fit a target who retaliates against the boss because direct resistance is unsafe, and who displaces hostility onto the organization.
Strong and consistent across all sources; attitudes cluster around ρ = −.30 to −.40.
Attitudinal damage is robust and convergent. Mackey et al. (2017) report perceived organizational support ρ = −.40, job satisfaction −.34 (k = 20, N = 6,469), and affective commitment −.26. Zhang and Liao (2015) report job satisfaction r = −.35, affective commitment −.30, and organizational identification −.22, and Schyns and Schilling (2013) report job satisfaction r = −.336 (k = 21). The metaBUS node confirms the cluster: job satisfaction r = −0.301 (k_articles = 13), affective commitment −0.198, organizational commitment −0.204, and trust in management −0.139. Notably, the strongest single attitudinal effect in Schyns and Schilling (2013) is attitudes toward the leader at r = −.571, locating the attitudinal collapse at the supervisor relationship itself before it generalizes to the organization. These effects are consistent in sign and rank order across meta-analysis, primary study, and the independent metaBUS database.
Moderate-to-strong on strain (exhaustion ρ = .36); weak on objective and clinical health (metaBUS health r = −0.064).
The abusive-supervision spine shows clear strain effects but thinner clinical evidence. Mackey et al. (2017) report emotional exhaustion ρ = .36 (k = 22, N = 7,761), work-to-family conflict ρ = .35, job tension ρ = .24, and depression ρ = .24 (k = 5). The metaBUS node places emotional exhaustion at r = 0.374, stress at 0.336, exhaustion at 0.335, and anxiety at a weaker 0.173, while objective health is near-zero at r = −0.064 (k_articles = 8). Psychological safety supplies the mechanism that transmits these effects: Frazier et al. (2017) show positive leader relations build safety (ρ̂ = .44), and safety in turn predicts satisfaction (ρ̂ = .53), engagement (.45), and task performance (.43). The strain column is therefore real and convergent, but it is dominated by self-reported exhaustion and tension; the harder clinical and physiological outcomes remain under-powered, and the bullying supplement (below) is what fills the mental-health gap.
Moderate but methodologically fragile; task performance ρ = −.19, with a supervisor-rating confound.
Performance damage is real but modest, and the estimates are methodologically fragile. Mackey et al. (2017) report task performance ρ = −.19 (k = 13, N = 2,872) and organizational citizenship behavior ρ = −.24 (k = 5). Zhang and Liao (2015) report engagement r = −.29, overall citizenship r = −.24, individual-directed citizenship −.21, organizational-directed citizenship −.17, and subordinate work performance −.16. The metaBUS node, despite dense sampling, returns weak performance effects: task performance r = −0.128 (k_articles = 32), in-role performance −0.126 (k_articles = 18), extra-role or citizenship −0.137 (k_articles = 14), and engagement −0.282 from only k_articles = 2. Two caveats bite here. First, supervisor-rated performance is partly a vehicle through which abuse is delivered, confounding the predictor with the criterion (Tepper et al., 2017). Second, Schyns and Schilling (2013) found the hard organizational-performance composite essentially null at r = .039 (k = 2, N = 333), so the performance column should be read as attitudinal and self-reported rather than bottom-line.
Weakest-supported consequence; turnover intention is moderate (r ≈ .25-.31) but actual or objective turnover is near zero.
This is the weakest link in the consequence network. Turnover intention is moderate: Zhang and Liao (2015) report r = .30 (k = 13, N = 5,950) and Schyns and Schilling (2013) report r = .313 overall, but split sharply by instrument (Tepper scale r = .222 versus other destructive-leadership instruments r = .339, Qb = 14.306, p < .001), with the independent metaBUS node at r = 0.253 (k_articles = 6). Actual or objective turnover, however, is barely supported: the only primary evidence is Tepper (2000), who found self-reported voluntary turnover predicted at standardized beta = −.37 with organizational justice fully mediating the path, while the lone hard-criterion composite in Schyns and Schilling (2013) is r = .039 and non-significant. metaBUS engagement and voice are also thin (k_articles = 2 and 4). The honest reading is that abuse reliably produces the intention to leave but the evidence that it actually drives people out the door, on objective criteria, is close to absent.
metaBUS is a curated database of the field's meta-analytic correlations, indexing "abusive supervision" as a single de-duplicated node. It is an independent check on the meta-analyses above, and its article count (k) shows where the evidence base is thick versus thin.
| Outcome | metaBUS r | k (articles) |
|---|---|---|
| Turnover intention | +0.253 | 6 |
| Job satisfaction | -0.301 | 13 |
| Affective commitment | -0.198 | 9 |
| Organizational commitment | -0.204 | 10 |
| Counterproductive behavior (CWB) | +0.431 | 38 |
| CWB toward individuals | +0.458 | 20 |
| CWB toward organization | +0.388 | 16 |
| Interpersonal justice | -0.554 | 6 |
| Procedural justice | -0.273 | 9 |
| Justice (overall) | -0.238 | 20 |
| Emotional exhaustion | +0.374 | 7 |
| Stress | +0.336 | 14 |
| Exhaustion | +0.335 | 11 |
| Engagement | -0.282 | 2 |
| Task performance | -0.128 | 32 |
| In-role performance | -0.126 | 18 |
| Extra-role/OCB | -0.137 | 14 |
| Trust in management | -0.139 | 23 |
| Voice behavior | -0.135 | 4 |
| Health | -0.064 | 8 |
| Anxiety | +0.173 | 4 |
Convergence across the abusive-supervision meta-analyses, the original primary study, and the independent metaBUS field-wide node is strong and reassuring, and the rank order of consequences holds in all three sources. metaBUS counterproductive behavior r = 0.431 (k_articles = 38, the densest cell) confirms Mackey (.41 to .53) and Zhang (.38 to .51); metaBUS interpersonal justice r = −0.554 confirms Mackey ρ = −.66; metaBUS job satisfaction r = −0.301 confirms Mackey −.34, Zhang −.35, and Schyns −.336; and metaBUS emotional exhaustion r = 0.374 confirms Mackey ρ = .36. Evidence density, read from metaBUS k_articles, is thickest for counterproductive behavior (38), task performance (32), trust in management (23), job satisfaction (13), and exhaustion or stress (11 to 14). It is thinnest for turnover intention (k_articles = 6), anxiety (4), voice (4), engagement (2), and objective health (8, with r = −0.064 near zero). Actual or objective turnover is the thinnest cell of all: a single primary study in the synthesis and a non-significant composite (r = .039), so behavioral exit is the least-supported claim in the network while deviance, performance, satisfaction, justice, and exhaustion are thickly and consistently established.
Three mechanisms carry the consequence network, with differential empirical support. Organizational justice is the dominant tested mediator: Tepper (2000) showed that justice perceptions fully mediated the abusive-supervision to voluntary-turnover path (the incremental abusive-supervision effect was non-significant once interactional, procedural, and distributive justice were controlled), and both Mackey et al. (2017) and Zhang and Liao (2015) document that abuse sharply erodes all three justice forms (interactional ρ up to −.55; r −.51). Psychological safety is the theoretically implied but under-tested complement: Frazier et al. (2017) show positive leader relations build psychological safety (ρ̂ = .44, k = 30, N = 10,180; transformational leadership .42; trust in leader .39; leader-member exchange .38; inclusive leadership .36), and psychological safety in turn predicts satisfaction (ρ̂ = .53), engagement (.45), task performance (.43), citizenship (.32), and voice (.31); critically, safety added incremental variance in task performance beyond leadership (ΔR² = .18 to .23 versus a reverse ΔR² of .01 to .08), implying abuse destroys the very safety state that transmits leadership effects, although no abusive-supervision meta-analysis has formally modeled safety as a mediator. Emotional exhaustion and social-exchange or conservation-of-resources routes round out the picture (emotional exhaustion ρ = .36), but these remain theorized rather than meta-analytically quantified. Honest caveats apply: Frazier et al. (2017) ran no formal indirect-effect or longitudinal test, only 13% of correlations were cross-source (same-source ρ = .45 versus different-source .29), objective task performance was only ρ = .07 versus subjective .40, and file-drawer effects appeared for performance, citizenship, and learning behavior.
| Statistic | What it says | Source |
|---|---|---|
| ρ = −.66, k = 5, N = 1,111; 80% CV −.66/−.66, SDρ = .00 | Strongest association in the abusive-supervision matrix: interpersonal justice collapse. | Mackey et al. (2017) |
| ρ = .53, k = 14, N = 5,223; ρXP U.S. = .60 | Largest behavioral consequence: supervisor-directed retaliation. | Mackey et al. (2017) |
| Tepper scale r = .222 vs other instruments r = .339, Qb = 14.306, p < .001 | Two-grains divergence on turnover intention by measurement instrument. | Schyns & Schilling (2013) |
| standardized beta = −.37, p < .01; Δχ²(2) = 11.71 | Only abusive-supervision to actual turnover result, fully mediated by justice. | Tepper (2000) |
| ρ̂ = .44, 95% CI [.39, .50], k = 30, N = 10,180 | Leadership builds the psychological-safety state abuse destroys. | Frazier et al. (2017) |
| ρ̂ = .62, 95% CI [.51, .73], k = 15, N = 4,648 | Psychological safety to learning behavior, the largest safety outcome. | Frazier et al. (2017) |
| r = .36, k = 48, N = 115,783 (CS); r = .21, k = 22 longitudinal | Bullying to mental health, cross-sectional and prospective. | Verkuil et al. (2015) |
| OR 1.58, 95% CI 1.39-1.79, k = 10 | Bullying to subsequent sickness absence, robust to controls. | Nielsen et al. (2016) |
| 18 of 25 (72%) pairwise construct-outcome comparisons had overlapping CIs | Empirical redundancy of the aggression constructs. | Hershcovis (2011) |
| r = 0.431, k_articles = 38 | Densest independent field-wide triangulation point: counterproductive behavior. | metaBUS |