The eligibility criteria for this meta-analysis study include quantitative empirical studies that examine the relationships between unethical leadership concepts and employees’ UPB. Specifically, the analysis includes studies that report on concepts such as leader UPB, unethical leadership, authoritarian leadership, leader moral disengagement, abusive supervision, exploitative leadership, and leader other-oriented perfectionism, in relation to employees’ UPB. The review incorporates both dyadic data and self-report survey data. Studies are not restricted by location or geography and encompass all types of settings and contexts. Only published articles from 2014–2024 are considered eligible. Qualitative studies are explicitly excluded from this analysis.

A systematic literature search was conducted across five online databases subscribed to by the researchers' institution: Scopus, Emerald Insight, JSTOR, ProQuest, and ScienceDirect. The search utilized Boolean techniques with specific search terms, including “unethical leadership AND unethical pro-organizational behaviour,” “leadership AND unethical pro-organizational behaviour,” “supervision AND unethical pro-organizational behaviour,” and “leader AND unethical pro-organizational behaviour.” These online databases provide a vast repository of peer-reviewed literature, ensuring the meta-analysis is based on high-quality and relevant studies while maintaining the transparency and reproducibility of the literature search strategy. The literature search was conducted between 22 October 2024 and 20 November 2024. The reliance on online databases aligns with other meta-analyses, such as those by Cuijpers et al. (2023), Doǧru (2022), and Rogers et al. (2023), which also used online databases for their studies.

The study employed two coders for the selection of relevant studies. All data were independently screened and reviewed by two coders. Agreement was achieved for 86.5% of the extracted data, with initial discrepancies resolved through discussions between the coders to determine study inclusion. For studies with persistent disputes, both coders expanded their screening process to include full-text analysis to ensure thorough evaluation and consensus.

For the meta-analysis, study details were systematically coded following a codebook (see Harun, 2026) developed based on the recommendations of Card (2012). Extracted data included information on sample country, sample setting (e.g., industry), sample size, measures for independent and dependent variables, means and standard deviations for both variables, effect sizes, and standard errors. The data were initially coded in Microsoft Excel file and later transferred to SPSS software Version 29 for meta-analysis. The primary coder independently extracted data after conducting a systematic literature search, ensuring thorough documentation and accuracy. A secondary coder reviewed the extracted data, achieving an agreement rate of 86.5% on the identified information. All initial discrepancies were addressed through detailed discussion and consensus between the coders, ensuring the reliability and validity of the data extraction process.

One meta-analytic study (Luan et al., 2023) was included because it offers a comprehensive quantitative synthesis of antecedents to employees’ UPB, providing effect size estimates unavailable in the primary literature. To prevent double-counting and circular reasoning, we applied a conservative coding protocol. The meta-analysis was treated as a single analytic unit, with only aggregated effect sizes extracted, and its primary studies were excluded from separate coding. We also cross-checked its primary study list against our database to ensure no dataset was duplicated. This approach maintained statistical independence while incorporating meta-analytic evidence that strengthened the robustness and generalizability of our findings.

The methodological quality of the included studies was assessed using the Quality Assessment Tool for Quantitative Studies (QATQS, National Collaborating Centre for Methods and Tools, 2010), which evaluates six key domains: selection bias, study design, confounders, blinding, data collection methods, and withdrawals/dropouts. Each study was rated as strong, moderate, or weak in each domain, with an overall global rating determined based on these domains. QATQS also highlights the discrepancies between reviewers and final decision after reconciliation. Table B2, Appendix B summarizes the quality assessment results based on QATQS criteria.

Of the 15 studies identified during systematic literature search, 80% (n = 12) were rated as strong, 7% (n = 1) as moderate, and the remaining 13% (n = 2) rated as weak. Strong studies typically demonstrated robust sampling strategies, appropriate control for confounding variables, and reliable data collection methods and analysis. Studies rated as moderate generally employed non-representative sampling techniques and lacked detailed descriptions of their data analysis procedures. Weak studies commonly exhibited high risks of selection bias, inadequate adjustment for confounders, or insufficient reporting of statistical analysis. Consequently, the two studies rated as weak were omitted from the subsequent analysis.

To assess the impact of study quality on the meta-analytic results, a sensitivity analysis was conducted by excluding studies rated as moderate quality. The exclusion of this study did not significantly alter the pooled effect size for all studies, Ź = .282 (Z = 7.741, p < .001, 95% CI = .211 to .353), compared to after exclusion, Ź = .286 (Z = 7.524, p < .001, 95% CI = .211 to .360). This indicates that the results are not substantially influenced by studies of moderate quality. A sensitivity analysis was not feasible for weak-quality studies due to insufficient reported information (e.g., effect size, variance, standard error).

The present study conducted a univariate meta‐analysis using a random‐effects model, as it provides more conservative effect size estimates compared to fixed‐effects models. A random‐effects model was applied to account for both within‐study sampling error and between‐study heterogeneity, thereby enabling generalization beyond the included studies (Borenstein et al., 2010; Luan et al., 2023). Moderator or subgroup analyses were not conducted due to insufficient numbers of effect sizes per category, which can yield unstable estimates and increase the risk of spurious findings (Viechtbauer, 2010). Effect sizes were calculated using Fisher’s Z coefficient. For studies that did not report effect sizes, they were computed from the provided raw correlation (r), standard deviation (SD), and sample size. For studies that reported standardized beta weights, they were transformed into r scores using the formula proposed by Peterson and Brown (2005) and subsequently converted to Fisher’s Z following the guidelines by Borenstein et al. (2009). For studies that report multiple independent effect sizes (e.g., Fehr et al., 2019; Liu et al., 2021; Luan et al., 2023; Schuh et al., 2021; Yan et al., 2024), all effect sizes were included. To assess heterogeneity, the present study observed the Q-statistic and I²-statistic, which evaluate variability across effect sizes beyond what is expected by chance. Additionally, publication bias was examined through a combination of visual inspection of funnel plots, Egger’s regression test, and Duval and Tweedie’s trim-and-fill method to account for potential bias in the included studies.

The data underlying this meta-analysis are openly available in Open Science Framework at Harun (2026).

A total of 129 studies were identified from five electronic databases subscribed to by the researchers’ institution (i.e., Scopus, Emerald Insight, JSTOR, Proquest, & Science Direct). After title and abstract screening, 66 studies were selected for full-text evaluation, and 13 studies met the eligibility criteria. However, one study was excluded due to an outlier, resulting in a final sample of 12 studies were included in the meta-analysis (see Figure 1 for meta-analytic flow diagram, Table 1 for individual study characteristics).

The meta-analysis included 12 studies, with six studies reporting multiple effect sizes, resulting in a total of 20 effect sizes. All included studies examined the relationship between unethical leadership concepts and employees’ UPB. The analyses encompassed 6,892 participants across five countries, four from Asia (i.e., China, Vietnam, Saudi Arabia, and Iran) and the United States. The studies were conducted between 2019 and 2024. Most of the studies focused on leader UPB, authoritarian leadership, and leader moral disengagement, while other independent variables included abusive supervision, exploitative leadership, leader other-oriented perfectionism, and unethical leadership. The sample sizes ranged from 175 to 1,043 participants, with an average of 391. The studies primarily utilized time-lagged surveys, vignette-based experiments, and dyadic survey data matching leaders/supervisors with their subordinates.

A total of 20 effect sizes from 12 studies were included, involving 6,892 participants across five countries, four from Asia (i.e., China, Vietnam, Saudi Arabia, and Iran) and the United States

Effect sizes assessing the effects of unethical leadership on employees’ UPB ranged from Ź = -.79 to .54 across 20 samples. The overall mean effect size was significant and positive (Ź = .238, p < .001, 95% CI .116 to .360). Observation on the Forest plot identified one outlier study and was removed from the subsequent analysis. The overall effect size after the deletion has improved (k = 20, Ź = .282, p < .001, 95% CI .211 to .353). This overall effect size is equivalent to approximately r = .275. This is a modest in magnitude, carries meaningful implications for organizational practice. In applied settings, this result suggests that unethical leadership explains a notable proportion of variance in UPB which is enough to have tangible consequences when scaled across an entire workforce. Even a moderate association means that changes in leadership ethical conduct can meaningfully influence the prevalence of UPB carried out by the employees.

Heterogeneity analyses indicated substantial between-sample variability, suggesting that the observed variation in effect sizes reflects genuine differences between studies rather than random sampling error (Q = 292.85, p < .01, I² = 92; see Table 2 and Figure 2). This high heterogeneity implies that the strength of the relationships between unethical leadership and UPB is not uniform across contexts. Differences in leadership style operationalization, industry and cultural settings, sample composition, and measurement tools may all contribute to this variability. Consequently, the pooled effect size should be interpreted as an average across diverse conditions rather than a precise universal estimate. This highlights that practical applications should be sensitive to contextual factors rather than assuming a one-size-fits-all effect.

To assess the publication bias, a funnel plot was generated (Figure C1, Appendix C). Visual inspection revealed an asymmetrical distribution of effect sizes, suggesting that smaller studies exhibited greater variability and a tendency toward larger effect sizes. Egger’s regression test for funnel plot asymmetry was conducted to statistically evaluate this observation. The test revealed significant asymmetry, t = 4.676, p < .001, 95% CI .297 to .783, indicating a potential presence of publication bias, probably due to smaller studies with nonsignificant or smaller effect sizes being underrepresented.

To further explore this, Duval and Tweedie’s trim-and-fill method was applied. However, this analysis estimated zero potentially missing studies, and the adjusted overall effect size remained identical to the original analysis. This suggests that while Egger’s test detected asymmetry, the trim-and-fill method did not identify sufficient evidence of missing studies to adjust the effect size.

A key strength of this study is its rigorous methodological design. The meta-analysis adhered to the PRISMA 2020 Statement (Page et al., 2021), using multiple high-quality databases, explicit inclusion criteria, and fully documented search strings (Table A1, Appendix A). Study quality was appraised via the QATQS, ensuring that only moderate- to high-quality empirical evidence informed the synthesis. A random-effects model with Fisher’s Z transformation was applied to accommodate substantial heterogeneity and permit generalisation beyond the sampled studies (Borenstein et al., 2009; Schmidt & Hunter, 2015). Methodological transparency was further demonstrated through publication bias assessment using funnel plots, Egger’s regression, and Duval and Tweedie’s trim-and-fill procedure (Sterne et al., 2011), enhancing the credibility and robustness of the findings.

Despite its methodological rigour, this meta-analysis has limitations. Substantial variation in effect sizes suggests that differences in sample characteristics, cultural contexts, and measurement tools may have influenced results. Although heterogeneity is common in meta-analyses, the univariate approach used here limits the ability to account for multiple sources of variability simultaneously (Riley et al., 2007), such as study design, cultural norms, and industry-specific factors. Cultural variability may further complicate comparability, as behaviours deemed unethical in one society may be viewed as normative in another, reflecting divergent value systems, leadership expectations, and ethical standards (Hofstede, 2001; House et al., 2004; Resick et al., 2006). Such differences can affect both the measurement and interpretation of unethical leadership and UPB, hindering direct cross-cultural comparisons (Dorfman et al., 2012; Tsui et al., 2007). Future meta-analyses should systematically code cultural background variables, such as Hofstede’s dimensions or GLOBE leadership scales, and test them as moderators to clarify the role of cultural context in shaping this relationship.

Secondly, the assessment of publication bias in this meta-analysis revealed potential concerns regarding the completeness and representativeness of the included studies. Both the funnel plot and Egger’s regression test indicated asymmetry, suggesting a possible underrepresentation of smaller studies with nonsignificant or weaker effects (Borenstein et al., 2009; Sterne et al., 2011). Although the trim-and-fill method estimated no missing studies, the observed asymmetry raises concerns that the overall effect size may be inflated. This pattern is consistent with the well-documented file drawer problem, wherein journals disproportionately publish studies with statistically significant results while null findings often remain unpublished (Rosenthal, 1979; Rothstein et al., 2005). As Rosenthal (1979) noted, such selective publication can distort the cumulative evidence base and lead to overestimated effect sizes. This bias is particularly relevant in the context of unethical leadership and UPB, where strong associations may be more appealing to publish, thereby shaping an incomplete or skewed understanding of the phenomenon (Franco et al., 2014).

The exclusion of unpublished studies further limits the comprehensiveness of this meta-analysis. Restricting the search to published literature may have overlooked valuable insights from grey literature such as dissertations, working papers, or internal organizational reports, which often contain nonsignificant results or employ diverse methodological approaches that could challenge or refine the conclusions (Hopewell et al., 2007; Song et al., 2010). Moreover, the search strategy relied primarily on electronic databases accessible through the authors’ institutional subscriptions and, to ensure methodological rigour, included only peer-reviewed journal articles. While this approach enhanced the quality and reliability of the evidence base, it inevitably excluded potentially relevant works indexed in other databases, such as PsycNET, and in grey literature repositories that may contain studies on unethical leadership and employees’ UPB. As Song et al. (2010) emphasised, the exclusion of unpublished and non-English studies can result in inflated estimates of effect sizes, thereby reducing the generalisability of findings.

Future research should take proactive measures to address the above methodological limitations. Key strategies include systematically incorporating unpublished and grey literature to balance the evidence base, expanding database coverage to additional disciplinary repositories, and applying statistical corrections for publication bias, such as trim-and-fill procedures (Duval & Tweedie, 2000) or selection models (Vevea & Hedges, 1995). Greater transparency can also be achieved through pre-registration of study protocols (Nosek et al., 2018), open sharing of data and analysis scripts (Munafò et al., 2017), and journal policies that encourage the publication of null or mixed results. Replication studies conducted across diverse organizational and cultural contexts will be especially important for validating observed relationships and refining effect size estimates (Open Science Collaboration, 2015). Collectively, these practices can mitigate publication bias, broaden the evidence base, and strengthen the robustness of conclusions in the unethical leadership–UPB literature.

Beyond methodological improvements, future research should also deepen theoretical understanding by examining the moderating and mediating factors that shape the unethical leadership–UPB relationship. Unpacking mechanisms such as psychological safety, organizational norms, and individual personality traits can clarify the processes through which unethical leadership promotes UPB. At the same time, exploring moderators such as team dynamics, organizational size, and leadership tenure can highlight the contextual conditions that amplify or buffer these effects. Such analyses not only advance theory but also generate actionable insights for designing interventions that reduce the likelihood of UPB under unethical leadership.

Expanding the geographical and industrial scope of research is crucial to enhance the generalizability of findings. Cross-cultural studies are particularly valuable, as cultural norms and values shape perceptions of leadership and ethical behaviour (Zhu, 2007). Exploring underrepresented regions, such as ASEAN, Africa, South America, or the Middle East, can fill critical gaps and highlight cultural variations in the unethical leadership-UPB relationship. Similarly, future studies should investigate this dynamic across diverse industries, particularly those less represented in the literature, such as creative sectors, non-profit organizations, or small businesses. Understanding the influence of industry-specific norms and pressures on this relationship can help tailor ethical guidelines and interventions to different organizational contexts.

Additionally, employing a variety of research designs can address current methodological limitations and strengthen the robustness of findings. Longitudinal studies, for instance, can establish causal relationships by capturing how unethical leadership evolves and impacts employee behaviours over time. Experimental designs can offer controlled settings to test specific variables, while meta-analyses with broader leadership constructs can provide a more comprehensive understanding of the diverse ways in which leadership styles affect UPB. Investigating constructs such as servant leadership, transformational leadership, or emerging leadership styles can enrich the literature and uncover alternative pathways to mitigating unethical behaviours in organizations. By broadening theoretical and methodological approaches, future research can significantly advance the understanding of unethical leadership and its implications for workplace behaviour.

Organizations must take proactive measures to address the impact of unethical leadership on employee behaviours, beginning with comprehensive leadership development programs. These initiatives should focus on cultivating ethical leadership practices that actively discourage behaviours like abusive supervision and authoritarian leadership. Such training can mitigate the risks of unethical leadership influencing employees to engage in UPB, which often reflects a reciprocal response to leaders' unethical conduct. Moreover, ethics-focused educational programs should highlight the dangers of UPB, emphasizing that while these behaviours may appear beneficial to organizational goals in the short term, they can have long-term negative consequences for both individual employees and the organization as a whole. By fostering an understanding of the harm caused by unethical behaviours, these programs can reinforce the importance of ethical decision-making across all organizational levels.

To maintain ethical practices, organizations should broaden their ethics monitoring systems to encompass leaders, ensuring accountability extends beyond the employee level. Establishing comprehensive ethical guidelines that explicitly address leadership behaviours can cultivate a culture of transparency and integrity. Regulatory bodies and policymakers can further support these efforts by advocating for more robust ethical governance frameworks, particularly in industries where UPB are prevalent. Additionally, policies should incorporate strong whistleblower protections, empowering employees to report unethical practices without fear of retaliation. These safeguards can foster transparency and trust, ultimately creating an organizational environment that prioritizes ethical conduct and mitigates the risks associated with unethical leadership and UPB. Collectively, these measures can establish a holistic framework for ethical governance and sustainable organizational success.