The link between use-of-force and burnout amongst police officers: a meta-analysis

1 Introduction

Police work uniquely combines chronic psychological strain with the institutional mandate to use force—a combination that places officers in situations requiring both sustained emotional regulation and the capacity to act decisively in high-stakes encounters with the public. The profession is routinely characterized by high demands, moral ambiguity, and frequent exposure to dangerous and potentially traumatic events (Alves et al., 2023; Ben Salah et al., 2025). Such experiences generate stress, and they also challenge the emotional regulation and empathic control essential for professional and proportionate conflict management, including use of force (UoF). The authority to use force, while central to the police role, is simultaneously a ubiquitous and highly regulated part of police work, constrained by ethical, legal, and moral norms and subject to considerable public scrutiny (Maguire et al., 2024). At the same time, chronic exposure to such strain has repeatedly been linked to adverse outcomes, including depression, anxiety, and burnout (Syed et al., 2020).

Burnout, defined as a response to prolonged and poorly managed occupational stress, encompasses emotional exhaustion, depersonalization or cynicism, and reduced professional efficacy (Schaufeli et al., 2009). Burnout is a multidimensional construct, and its components may exert divergent effects on behavior that are dependent on a multitude of factors, ranging from individual differences such as personality, to contextual factors and what outcomes are measured (Civilotti et al., 2022; Swider and Zimmerman, 2010). The direction of this relationship is also theoretically ambiguous. Burnout may influence UoF through reduced emotional regulation and altered decision-making, while repeated exposure to force-related incidents may contribute to burnout, particularly depersonalization, through mechanisms such as emotional numbing or moral strain. In addition, both burnout and UoF may arise from shared organizational stressors. Taken together, this suggests a multidirectional relationship rather than a simple linear association. Empirical findings are correspondingly mixed, with studies reporting weak or heterogeneous associations between burnout and behavioral outcomes in policing contexts (Euwema et al., 2004; Kop and Euwema, 2001; Swider and Zimmerman, 2010).

Despite extensive research on police stress and mental health (Regehr et al., 2021; Syed et al., 2020; Ugwu and Idemudia, 2025), no systematic synthesis has yet examined the relationship between burnout and use of force. Given the conceptual complexity and empirical inconsistency of the existing literature, a meta-analytic approach is needed to clarify whether an association exists, and under what conditions it may emerge. Examining burnout dimensions separately, as well as potential moderators such as measurement type and study context, may help explain divergent findings across studies. Such an understanding is essential both for theory and practice—informing prevention, training, and organizational strategies aimed at mitigating burnout and promoting proportionate, ethical policing. Against this background, the present meta-analysis examines the association between police officers' burnout and their use of force with specific focus on disentangling subdimension-specific effects and sources of heterogeneity. Specifically, it addresses the following questions:

2 What is the association between police officers' use of force and their different levels of burnout?

2.1 Which factors moderate this association?

The remainder of the paper proceeds as follows. The next section outlines the conceptual background linking burnout to police use of force, followed by a description of the systematic search, screening, and analytic procedures. We then present pooled and subgroup results, including moderator analyses, before discussing their implications for understanding how psychological strain may shape officers' use-of-force tendencies.

2.2 Background

As previously mentioned, police work has been hailed one of the most stressful occupations that one can have, and several studies have examined the prevalence of burnout. The negative effects of burnout on the individuals that are affected range from psychological disturbances such as depression, anxiety and elevated risk of suicide to physical symptoms such as metabolic syndromes (Bishopp and Boots, 2014; Correia et al., 2023). With regards to job performance and behavioral effects the results are more fragmented, not least for research on police officers. Some studies have found that burnout is associated with increased aggressiveness and interpersonal conflict, although findings are not consistent and the direction of these relationships remains unclear (Gomes et al., 2022; Queirós et al., 2013). In the following section the constructs of burnout and UoF will be described in further detail.

2.3 Burnout

The term burnout was first introduced by Herman Freudenberger in 1974 and later developed by Christina Maslach and colleagues into a multidimensional construct that has since become central to occupational health research (Schaufeli et al., 2009). While there are several different measures and definitions of burnout, the most commonly used sub-dimensions are emotional exhaustion (EE), depersonalization or cynicism (DP), and reduced professional efficacy or personal accomplishment (PA) (Maslach and Jackson, 1981a).

These three dimensions capture distinct psychological processes that may have different implications for behavior. EE reflects a depletion of emotional resources and is often associated with withdrawal and reduced engagement. DP involves emotional distancing and the development of cynical or detached attitudes toward others, which may reduce empathy in interpersonal interactions (Swider and Zimmerman, 2010). Reduced PA reflects feelings of ineffectiveness and may influence motivation and work-related functioning (Maslach and Jackson, 1981b). In policing contexts, these dimensions are particularly relevant given the interpersonal and high-stakes nature of the work. Officers are routinely required to make rapid decisions in situations involving conflict, uncertainty, and potential threat, and variations in emotional engagement and perceptions of others may influence how such encounters are managed. These behavioral implications are particularly relevant in situations where encounters may escalate and require decisions regarding the use of force. In later conceptual advancements of burnout, Leiter and Maslach (2016, pp. 89–90) argued that, despite the established idea that “… various patterns of burnout, which might change at different points in time […] the potential of these varying patterns has not been exploited very much…”, but instead, that there has been a shift toward treating burnout as a one-dimensional construct. These dimensions do not necessarily move in tandem but are instead best explored as different pathways (Leiter and Maslach, 2016). In empirically identifying a set of such pathways, Leiter and Maslach (2016) found five profiles: Burnout (high on all three dimensions, reflecting a fully developed syndrome with emotional depletion, detachment, and reduced efficacy), Engagement (low on all three, indicating high energy, involvement, and professional efficacy), Overextended (high emotional exhaustion only, suggesting strain and resource depletion without detachment), Disengaged (high depersonalization only, reflecting cynicism and emotional distancing without necessarily being exhausted), and Ineffective (high inefficacy only, indicating reduced sense of competence and accomplishment without pronounced exhaustion or detachment).

In terms of measurement, burnout has most commonly been measured using the Maslach Burnout Inventory (MBI), widely regarded as one of the most widely used instruments in the field (Schaufeli et al., 2020). The MBI assesses the three dimensions described above and has been extensively applied across occupational settings, including policing, where the MBI-HSS (Human Services Survey) has been used extensively (Alves et al., 2023; Ben Salah et al., 2025). Typically, high scores on EE and DP indicate symptoms of burnout, while items for PA are positively phrased and thus low scores are indicative of burnout (Maslach et al., 1997). While the original descriptions of burnout, and the MBI-GS (General Survey), were focused on occupations that involved interpersonal labor, later definitions have broadened the scope to include work-related stress regardless of occupation (Maslach and Jackson, 1981a; Schaufeli et al., 2020). Items in the MBI-HSS are therefore less focused on individuals as the source of exhaustion or cynicism and more geared toward work itself as the source.

Empirically, police work is characterized by several factors known to contribute to burnout, including exposure to potentially traumatic events, shift work, and frequent interpersonal conflict (Alves et al., 2023; Ben Salah et al., 2025; Gomes et al., 2022). These occupational demands may increase vulnerability to burnout across its subdimensions, and we can theoretically assume that all of Leiter and Maslach's (2016) profiles exist in the police occupation to different degrees. Empirical studies further indicate that burnout is prevalent among police officers, with a substantial proportion reporting elevated levels on key dimensions. For example, a Swedish study found that approximately 28% of officers reported high levels of EE and 56% high levels of DP (Backteman-Erlanson et al., 2013), while similar patterns have been observed in other contexts, including among Spanish officers (De la Fuente Solana et al., 2013). In a systematic review of psychosocial risk and protective factors for burnout among police, results showed mixed findings for many of the factors examined, including sociodemographic, personality and coping strategies, with organizational and operational factors being the most predominant factors for burnout (Alves et al., 2023; Ben Salah et al., 2025).

Taken together, existing research indicates that burnout is both conceptually complex and empirically widespread among police officers. At the same time, variation in how burnout is measured and in how its dimensions are expressed across contexts may contribute to the mixed findings observed in studies examining its behavioral correlates.

2.4 Use of force

One defining aspect of police work is the paradox of having the mandate to use force against citizens to complete their tasks, but also the need to avoid using it (Bittner, 1970; Todak and James, 2018). The mandate to use force comes with great responsibility and police officers are under constant scrutiny, not least when using force (Emsing et al., 2024).

Previous studies on UoF that have examined what factors that affect the amount and frequency of use of force have examined both officer and “perpetrator” as well as encounter and community (Bolger, 2015). As for officer characteristics, experience (Mangels et al., 2020; Paoline and Terrill, 2007), gender (Bolger, 2015) and educational attainment (Lim and Lee, 2015; Paoline and Terrill, 2007) have been shown to have a negative correlation with use of force, with more experienced, female officers with college degrees (or similar) appearing to be less prone to UoF.

There are however significant methodological challenges associated with studying police officer UoF (Emsing et al., 2024). UoF as a construct has varying definitions making comparisons difficult (Celestin and Kruschke, 2019; Emsing et al., 2024). Across studies, use of force may refer to a range of actions, including physical restraint, use of weapons, and in some cases broader coercive behaviors. Some studies rely on officially recorded incidents, while others assess self-reported behavior or attitudes toward UoF. Importantly, studies may therefore capture either actual use of force or attitudes toward or willingness to use force, which may reflect distinct underlying processes. This variability complicates direct comparisons across studies. In addition, despite the large number of interactions between police and citizens, the number of use of force incidents are proportionally low and rare events making statistical comparisons difficult (Bennell et al., 2021; Sunde et al., 2025). Further, the availability of data regarding UoF is often less than satisfactory, and the scientific value of the data is affected by for example different definitions of UoF and routines for when officers are required to report such incidents (Bennell et al., 2021). In the US and many other countries various forms of UoF continuums are often employed, both for training as well as policy-making and research. Such UoF continuums however, similarly do the definitions of UoF differ and range from mere presence of police to using physical force (Emsing et al., 2024).

Despite being rare events, UoF incidents that are even perceived to be unjust or unproportional can have significantly negative effects on public trust in the police and can damage the relationship between the police and citizens (van de Klomp et al., 2014). Furthermore, the risk of injury for both police officers as well as citizens have been shown to increase when communication fails or is not possible, and UoF is necessary (Smith et al., 2010). Minimizing the UoF among police officers is therefore of utmost importance.

2.5 Conceptual framework

Based on the considerations outlined above and building on to the idea that burnout is comprised of different pathways (Leiter and Maslach, 2016), we assume that the association between burnout and UoF can be understood as a multidimensional and potentially bidirectional process. While the original body of burnout research do not itself hypothesize how burnout is linked to UoF, we here draw together a body of research, both empirical and theoretical, that can point us in a few directions. This is not an exhaustive conceptual framework but rather a crude analytical framework in which we start operating in.

DP, characterized by emotional distancing and reduced empathy, may affect how officers perceive and respond to citizens during encounters, potentially shaping decisions related to force. This profile sits within the “disengaged” category of people (Leiter and Maslach, 2016). In that sense, DP can be understood as a form of emotional distancing that reduces empathic engagement in interpersonal encounters. Empathy typically functions as a key inhibitory mechanism, constraining harmful or coercive actions by making the suffering of others psychologically salient. When empathic engagement is reduced, this inhibitory function is weakened (Espejo-Siles et al., 2020). In such conditions, individuals may more readily engage in processes of moral disengagement, that is, cognitive mechanisms that allow harmful actions to be reframed as acceptable without violating one's moral standards. These include, for example, moral justification, dehumanization of others, and attribution of blame. Empirical research on policing has shown that such mechanisms are actively used to legitimize the use of force, for instance by construing civilians as adversaries, emphasizing the necessity of maintaining public order, or shifting responsibility to higher authorities (e.g., Soares et al., 2018). We should thus expect a positive correlation between DP and UoF.

Different theoretical pathways are possible for the EE construct. Corresponding to the “overextended” profile (Leiter and Maslach, 2016), one possibility is that EE reflects a depletion of cognitive and emotional resources that are necessary to manage conflict in an appropriate manner. This kind of sustained emotional labor that police officers inevitably need to engage in (Hopkins et al., 2023) could, potentially, lead to poor self-regulation and thus more UoF. However, we theorize that the opposite is more likely. Instead of escalating conflicts, EE may lead to withdrawal or reduced engagement, which could alter officers' exposure to situations where force is required. In their original work, Maslach and Jackson (1981a) showed that police officers with high EE exhibited signs of withdrawal more than their counterparts without high EE. Theoretically, emotional fatigue could also be linked with how dominant police officers are in such situations; Euwema et al. (2004) have shown that less dominance is linked to more “effective” conflict situations, defined in terms of reduced escalation and higher satisfaction among both officers and civilians. We could thus expect a negative association, or a null finding.

In contrast to DP, reduced PA does not necessarily operate through diminished empathy or moral disengagement, but through perceptions of competence and effectiveness in interaction. Research suggests that officers' sense of accomplishment may be linked to the perceived effectiveness of their interactional strategies, including the use of authority and dominance (Euwema et al., 2004). In line with this, studies on emotional labor in policing indicate that the use of coercive or dominant emotional expressions can increase officers' sense of personal accomplishment when such strategies are successful in achieving compliance or control (Schaible and Six, 2016). From this perspective, lower levels of PA may be associated with reduced confidence in the effectiveness of such strategies, potentially resulting in less assertive or less dominant interaction styles. Within the framework of Leiter and Maslach's (2016) burnout profiles, this corresponds to the ineffective profile, where diminished efficacy reflects a lack of perceived competence rather than emotional detachment. This suggests that PA may shape how officers engage in interactions, particularly their willingness to assert control, rather than whether they disengage or morally reframe the situation.

Taken together, these dimension-specific mechanisms provide analytically distinct pathways, but they do not account for the interdependence between burnout dimensions or their co-occurrence in more complex profiles (Leiter and Maslach, 2016). Empirical work suggests that these dimensions may be correlated and dynamically related rather than independent (e.g., Hawkins, 2001), indicating that their behavioral implications are likely shaped by their combined configuration rather than by any single dimension in isolation.

3 Method

This study is a systematic review and meta-analysis conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines (Page et al., 2020). The protocol was developed a priori and follows best practices for evidence synthesis. The study protocol was registered at the Open Science Framework (https://osf.io/j5t9z)¹

3.1 Search strategy

Extensive preliminary searches of the literature were conducted before establishing the final search string. Ultimately, we developed a search string with three blocks: (police^* OR “police officer^*” OR “law enforcement officer^*” OR “police personnel^*”) AND (“mental health^*” OR “mental illness^*” OR burnout^* OR depersonalization^* OR cynicism^* OR “personal accomplishment^*” OR “emotional exhaustion^*” OR stress^* OR anxiety) ^* AND (“use of force^*” OR “use-of-force^*” OR “excessive force^*”). We searched the title, abstracts, and keywords where applicable, and searched, in total, six electronic databases selected for their suitability, including PsycINFO, PsycARTICLES, Web of Science, Scopus, Criminal Justice Database, and PubMed. We allowed for gray literature in this meta-analysis due to the low number of articles on UoF and burnout. Back- and forward citation searches were also performed on relevant works.

3.2 Selection procedure

This meta-analysis was structured using the PEO framework (Munn et al., 2018), in which the population is active-duty police officers, the exposure is UoF, and the outcome is burnout. The review is non-interventional and focuses on synthesizing quantitative evidence on the association between these two constructs rather than evaluating the effects of a treatment or training program. UoF is conceptualized as an occupational exposure that may contribute to psychological strain over time, while burnout is treated as the primary outcome of interest. However, given the cross-sectional design of all the studies in this review, it is not possible to test the direction between UoF and burnout. In other words, we cannot say from this meta-analysis alone whether UoF leads to burnout, whether burnout leads to UoF, or whether there is a reciprocal relationship. By applying the PEO model, the review is analytically anchored in an exposure–outcome logic that is appropriate for observational research and allows for the systematic identification, screening, and synthesis of studies examining the relationship between burnout and UoF across policing contexts. We did not put a time-limit on the inclusion criteria, and in addition to peer-reviewed scientific articles, we allowed for gray literature (PhD dissertations and peer-reviewed conference papers). This decision was made based on a substantial pre-search of the literature, where it was clear that we needed to widen the inclusion criteria in order to have enough articles to make the analysis meaningful. All records were retrieved and imported into the open web application RAYYAN (https://new.rayyan.ai/), and duplicates were removed in the first phase. In the second phase, abstracts were screened. If the abstracts clearly made mention of police officers and mental health indicators, they were forwarded to the full-text screening stage. The second author was responsible for abstract screening, however, to ensure a sound methodological procedure, the first and second author retrieved 100 reports that were screened independent of one another. Interrater reliability was good with Cohen's K = 0.91, indicating nearly perfect agreement. In the full-text screening stage, both the first- and second author read the texts in full, with potential disagreements being discussed and, if need be, solved by the third author. However, no such disagreements ever arose.

In the selection procedure we encountered some cases where the same sample had been used. To not double count, and thus inflate the meta-analysis results, we chose what we believed was the most relevant article. Firstly, Allen's (2022) PhD thesis was published as a peer-reviewed scientific article—we therefore choose the peer-reviewed article because of its conventional higher standards due to peer-review. Secondly, (Kop and Euwema 2001) have published extensively on the same sample of police officers. To not double count, we chose the latest article using the relevant measures for this meta-analysis. The full procedure is visualized in Figure 1.

3.3 Data collection

Relevant articles were collected in an Excel spread sheet with columns regarding authorship, year, country, theory, methodology, main results, sample size, gender distribution, age, work experience, police officer rank, location (urban, countryside), measurements, Cronbach's a test on measurements (if any), and relevant statistical output. In the case of missing data where measures had been implemented but not reported, mainly because of statistical insignificance, we contacted the authors to retrieve the missing data.

3.4 Data synthesis

Meta-analysis and narrative synthesis were used to analyze and integrate the included studies. All analyses were conducted in R using the meta (Schwarzer, 2007) and metafor packages.

Overall effect sizes were estimated using random-effects models. This choice was made a priori, given the substantial conceptual and contextual variation across policing environments, which makes it implausible to assume a single common effect size (Borenstein et al., 2021). Effect sizes were synthesized using correlation coefficients, as this was the most frequently reported metric. When correlations were not available, they were derived from reported regression coefficients using the procedure proposed by Peterson and Brown (2005). When multiple effect sizes were reported from the same sample (e.g., multiple UoF outcomes), they were combined using Fisher's z transformation and averaged prior to back-transformation to Pearson's r, to avoid statistical dependence while preserving each study's contribution (Borenstein et al., 2021). As a robustness test, we also tested three-level-nested meta-analyses where multiple UoF outcomes were detected, using Van den Noortgate et al.'s (2013) approach. The results did not meaningfully change, and since Van den Noortgate et al. (2013) suggest that these nested models are used when there are enough studies to make reasonable estimations (a minimum of 30), we retained our original meta-analytical approach. These tests are found in Appendix A. Statistical heterogeneity was assessed using Cochran's Q, τ², and I². Cochran's Q tests whether the dispersion in observed effect sizes exceeds what would be expected by sampling error alone, while I² and τ² quantify the proportion and absolute magnitude of between-study variance, respectively (Borenstein et al., 2021). Potential publication bias was examined using funnel plots, Egger's regression test, Begg's rank correlation test, and trim-and-fill procedures (Borenstein et al., 2021).

Based on the results of the primary analyses, moderator analyses were conducted for the DP subdimension, as this was the only outcome showing a statistically significant pooled association. This resulted in a small analytic sample (k = 8), and not all studies reported the same study-level characteristics, which constrained the set of moderators that could be examined. Given this limited evidence base, moderator analyses were treated as exploratory and descriptive. We therefore examined moderators separately, using subgroup analyses for categorical variables and single-moderator meta-regressions for continuous variables. This strategy was chosen to map how heterogeneity was distributed across key study characteristics, without conditioning on other variables that were structurally related within the small set of studies. In this context, a fully adjusted multivariable meta-regression would be highly parameterized relative to the available data and vulnerable to leverage by individual studies. Rather than treating such a model as a primary source of inference, we prioritized simpler, one-moderator analyses to preserve interpretability and reduce the risk of overadjustment. Results are therefore presented as patterns of variation across study characteristics rather than as independent moderator effects.

3.5 Critical appraisal

A critical appraisal of the included studies was conducted using criteria informed by the Appraisal tool for Cross-Sectional Studies (AXIS) guidelines (Downes et al., 2016). It consists of 20 items covering key domains including study design, reporting quality, and potential sources of bias. Each item is evaluated using a categorical response (e.g., yes, no, or unclear), allowing reviewers to systematically examine aspects such as sampling methods, measurement validity, statistical analysis, and transparency of reporting. The AXIS tool does not generate an overall numerical score; instead, it supports a qualitative judgement of study quality based on the pattern of responses across items. It is widely used in evidence synthesis to inform the interpretation of cross-sectional evidence. The second and third author jointly conducted the critical appraisal. Potential disagreements were discussed with the first author. The full results are presented in Table 1.

Table 1. Critical appraisal of the included studies.

Most studies were rated as moderate quality (n = 5), with one high-quality and two low-quality studies, reflecting considerable variability in methodological rigor. Only a small number of studies reported formal power analyses, and several relied on relatively small samples, raising concerns about statistical power and the stability of estimates reported.

A key methodological issue concerns the lack of uniform definitions and measurement of UoF across studies, as reflected in the AXIS appraisal. For example, one study (Allen and Jacquin, 2022) operationalizes UoF using supervisor-reported complaints of excessive force, whereas others employ broader use-of-force continua ranging from verbal commands to lethal force. This variation is evident across the included studies, which range from attitudinal measures (e.g., Burke and Mikkelsen, 2005) to complaint-based outcomes (Allen and Jacquin, 2022) and behavioral measures. Studies utilizing administrative or observational data (e.g., Mastracci and Adams, 2020; Sack, 2009; Kop and Euwema, 2001) represent a potential methodological strength, as they reduce reliance on self-report and associated recall and social desirability bias. However, differences in reporting thresholds varies across departments. Specifically, what level of force requires documentation limits comparability across studies. Table 2 reports on the detailed use of different UoF and burnout measures.

Table 2. UoF and burnout measurement types in the included studies.

Kop and Euwema (2001).

Schaible and Six (2016).

Use of firearm excluded in analysis.

Paoline and Terrill (2011).

UoF is generally a low-frequency event, as noted across several of the included studies, which poses challenges for statistical analysis, particularly in studies with small samples or short observation periods. As identified in the appraisal, several studies relied on self-reported measures of use-of-force, which are susceptible to recall and social desirability bias, whereas a smaller number of studies employed administrative or observational data.

Non-response bias represents a significant concern, particularly in studies examining work-related health outcomes as reflected in the moderate to low response rates and lack of non-response analysis in several studies. Officers experiencing higher levels of burnout may be less likely to participate in survey-based research, potentially biasing results. This issue is compounded by the widespread use of anonymous survey methods and the limited availability of detailed population data from police departments, which restricts the ability to assess representativeness.

4 Results

The ten studies included in the meta-analysis examine the relationship between burnout and UoF among police officers across a range of national context, all using cross-sectional designs. The sample spans five countries (United States, Netherlands, Norway, Switzerland/Belgium/France, and multinational conference data), with publication years ranging from 2001 to 2024. Five studies are peer-reviewed journal articles (Burke and Mikkelsen, 2005; Kop and Euwema, 2001; Manzoni and Eisner, 2006; Ellrich, 2016; Mastracci and Adams, 2020), while four are doctoral dissertations (Sack, 2009; Fiske, 2021; Devan, 2020; Allen and Jacquin, 2022), and one is a (peer-reviewed) conference paper (Telle et al., 2024).

As an exploratory step, we computed an overall meta-analysis combining all available studies reporting an aggregate burnout score (i.e., pooled across subdimensions). While this approach is conceptually problematic—given the multidimensional nature of burnout and potential differences in subscale meaning—it is presented here for transparency and as a descriptive overview.

The random-effects model (k = 10, n = 4,916) produced a pooled correlation of r = 0.05, 95% CI [−0.14, 0.25], p = 0.60, indicating no statistically significant association between UoF and aggregate burnout. Between-study heterogeneity was extremely high (I² = 97.8%; τ² = 0.0989; Q(df = 9) = 401.03, p < 0.0001). Given the substantial heterogeneity and the theoretical limitations of aggregating burnout dimensions, this result should be considered only as a preliminary descriptive summary. Subsequent analyses therefore focus on specific burnout dimensions (see Figure 2).

4.1 Emotional exhaustion

A random-effects meta-analysis of k = 7 studies (N = 1,840) examined the association between UoF and EE. The pooled correlation was r = −0.06, 95% CI [−0.19, 0.08], p = 0.39, indicating no statistically significant association. Heterogeneity was substantial (I² = 87.6%; τ² = 0.0283; Q(df = 6) = 48.24, p < 0.0001) (see Figure 3).

4.2 Depersonalization

A random-effects meta-analysis of k = 8 studies (N = 4,483) examined the association between UoF and DP. The pooled correlation was r = 0.21, 95% CI [0.02, 0.38], p = 0.03, indicating a small but statistically significant positive association. Heterogeneity was considerable (I² = 97.3%; τ² = 0.0725; Q(df = 7) = 258.56, p < 0.0001) (see Figure 4).

4.3 Personal accomplishment

A random-effects meta-analysis of k = 5 studies (N = 1,579) examined the association between UoF and sense of PA. The pooled correlation was r = −0.04, 95% CI [−0.11, 0.03], p = 0.25, indicating no statistically significant association. Heterogeneity was low to moderate (I² = 43.1%; τ² = 0.0025; Q(df = 4) = 7.03, p = 0.13) (see Figure 5).

4.4 Sensitivity analyses

Two sensitivity tests (Borenstein et al., 2021) were deployed on the full model, where we (1) excluded gray literature, and (2) the studies that required conversion of regression coefficients to correlations using the approach proposed by Peterson and Brown (2005). To assess the influence of individual studies on the pooled effect of our subdimensions, we conducted a leave-one-out analysis using the metainf function from the meta package in R. We deployed this on the subdimensions of emotional exhaustion and depersonalization, because of the low number of studies for personal accomplishment. This approach re-estimates the pooled effect size under a random-effects model after sequentially omitting one study at a time. Correlations were transformed to Fisher's z for analysis and back-transformed for presentation. For each iteration, we recorded the pooled effect size, 95% confidence interval, p-value, and between-study heterogeneity (I²) to identify studies exerting disproportionate influence on the overall results.

4.4.1 Full model

A sensitivity analysis excluding gray literature (k = 4 of 10 studies) resulted in a modest increase in the pooled effect size (r = 0.14, 95% CI [−0.12, 0.37]) compared to the full model (r = 0.05, 95% CI [−0.14, 0.25]). This shift appears to be driven primarily by the exclusion of one study reporting a relatively strong negative association. However, in both models, the pooled effect remained small, non-significant, and highly imprecise, with confidence intervals spanning zero and substantial heterogeneity (I² ≈ 98%). Three studies required conversion of regression coefficients to correlations using the approach proposed by Peterson and Brown (2005). A sensitivity analysis excluding these studies (k = 7) resulted in a smaller pooled effect size (r = 0.02, 95% CI [−0.06, 0.11]) compared to the full model (r = 0.05, 95% CI [−0.14, 0.25]). Notably, heterogeneity decreased substantially (I² = 76% vs. 98%), indicating that the converted effect sizes contributed disproportionately to between-study variability. However, in both models, the pooled effect remained small, non-significant, and close to zero.

4.4.2 Depersonalization

The leave-one-out analysis indicated that the pooled correlation ranged from 0.161 (95% CI: −0.0026–0.3169) to 0.256 (95% CI: 0.0685–0.4268) depending on which study was omitted. The direction of the effect never changed, but the p-value varied around the conventional significance threshold, ranging from 0.016 to 0.073. Between-study heterogeneity (I²) remained high in all models (91.9%−97.7%), with the largest reduction observed when excluding Ellrich (2016) (I² = 91.9% vs. 97.3% in the full model), suggesting that this study contributed disproportionately to between-study variability.

4.4.3 Emotional exhaustion

The random-effects pooled correlation was small and non-significant (r = −0.060; 95% CI: −0.244 to 0.128; I² = 87.6%). Leave-one-out analyses showed pooled estimates ranging from −0.100 to +0.003, with p-values spanning 0.264–0.964. Thus, the effect was consistently non-significant and generally negative; when omitting (Devan 2020) the estimate approached zero (r ≈ 0.003). Heterogeneity remained high across models (I² = 85.7%−89.6%), with the largest reduction observed when excluding Devan (2020) (I² = 76.6%; τ = 0.125 vs. 0.190 in the full model), indicating that this study contributed the most to between-study variability.

5 Publication bias

5.1 Depersonalization

Visual inspection of the funnel plot suggested some asymmetry. Egger's regression test indicated no statistically significant evidence of asymmetry (t = −1.86, df = 6, p = 0.113), and Begg's rank correlation test likewise yielded a non-significant result (z = −0.99, p = 0.322). Trim-and-fill analysis imputed five missing studies; however, contrary to what would be expected under classic publication bias, the adjusted pooled correlation increased from r = 0.214 to r = 0.453 (95% CI: 0.233–0.628). Because trim-and-fill typically reduces an inflated effect size by imputing missing small or non-significant studies, this pattern suggests that the observed funnel plot asymmetry is unlikely to be driven by selective non-publication. Instead, the increase in the pooled effect is more plausibly attributable to heterogeneity-driven asymmetry and sensitivity to influential studies, a known limitation of trim-and-fill under substantial between-study heterogeneity (see Peters et al., 2007; Terrin et al., 2003) (see Figure 6).

5.2 Emotional exhaustion

The funnel plot suggested potential asymmetry, which was supported by Egger's regression test (t(5) = −2.67, p = 0.044), but not by Begg's rank correlation test (z = −1.65, p = 0.099). Trim-and-fill analysis estimated three potentially missing studies on the right side of the plot. Adjusting for these imputed studies increased the pooled correlation from −0.060 (95% CI: −0.244 to 0.128) to 0.068 (95% CI: −0.138 to 0.268), but the association remained non-significant (also, keep in mind our previous brief note on trim-and-fill procedures that inflate the effect size). Heterogeneity remained high (I² = 91.4%), suggesting that potential publication bias does not fully explain the between-study variability (see Figure 7).

5.3 Sub-group analysis and meta-regression

5.3.1 Subgroup (Region)

The random-effects pooled correlation across all studies was r = 0.214 (95% CI: 0.027–0.386; p = 0.031), with very high heterogeneity (I² = 97.3%). Subgroup analyses indicated a moderate positive association in Europe (r = 0.349; 95% CI: 0.096 to 0.560) and a near-zero association in the US (r = 0.062; 95% CI: −0.219 to 0.333). The between-group difference was statistically significant (Q = 6.07, df = 1, p = 0.014). Heterogeneity remained high within both subgroups (Europe I² = 96.6%; US I² = 92.1%) (see Figure 8).

5.4 Subgroup measurement type (attitude vs. real UoF)

In our initial operationalization, we calculated the average pearson's r if studies reported both attitudes toward UoF and actual UoF (i.e., measurements of UoF). To properly assess whether these measurements differed, we disaggregated two studies that had aggregated scores. Here, to avoid dependence between samples in the analysis, we chose to go with the attitude effect size. This choice was made since attitudinal studies were fewer (k = 2). To improve the sample size and make a meaningful comparison, we thus extracted the attitude effect size. We also checked the direction and pattern of the ‘actual' UoF measure—all in all, this was in line with the other actual UoF measures, thus reinforcing the robustness of this choice. The random-effects pooled correlation across all studies was r = 0.227 (95% CI: 0.035–0.403; p = 0.027), with very high heterogeneity (I² = 97.3%). Subgroup analyses indicated a moderate positive association for studies using attitude-based measures (r = 0.362; 95% CI: 0.118–0.564) and a near-zero association for studies using real use-of-force data (r = 0.075; 95% CI: −0.246–0.382). The between-group difference was statistically significant (Q between = 5.33, df = 1, p = 0.021). Heterogeneity remained high within both subgroups (Attitude I² = 96.3%; Real UoF I² = 94.1%) (see Figure 9).

5.5 Meta-regression on gender distribution

An exploratory meta-regression was conducted to examine whether study-level gender composition was associated with variability in the DP effect sizes (k = 8). The moderator was mean-centered so that the intercept reflected the predicted effect at the average gender composition across studies (mean proportion male ≈84%), rather than an extrapolated value outside the observed data range (71–94% men). The model indicated a negative association between gender composition and the UoF–DP association (β = −3.06 in Fisher's z units, SE = 0.65, F(1,6) = 22.46, p = 0.003), suggesting that studies with higher proportions of men tended to report weaker UoF–DP associations. At the average gender composition, the predicted association was r ≈ 0.21. The model reduced between-study heterogeneity relative to the intercept-only model (R² = 78.7%), although substantial residual heterogeneity remained (I² = 79.9%). Given the small number of studies, and the restricted range of the moderator, this pattern should be interpreted cautiously. This pattern is illustrated in Figure 10, which shows that studies with higher proportions of men tended to report weaker between UoF and DP (see Figure 10).

6 Discussion

The aim of the current meta-analytic review was to examine the association between burnout and its different subdimensions and UoF among police officers. The findings provide evidence for a significant, albeit small association between depersonalization and UoF. This result suggests that individuals experiencing higher levels of DP, characterized by emotional distancing, cynicism, and reduced empathy, also engage in violent or aggressive behaviors. This finding aligns with the underlying theoretical model of burnout, that posits that emotional exhaustion and detachment can reduce self-regulation and empathy, thereby increasing the risk of violent responses, especially in high-stress professional contexts such as policing. However, we are not able to test this direct causal pathway, we are merely able to detect the statistical significance of this particular link. Previous research has also emphasized the potential role of depersonalization in shaping police behavior. For example, Kop and Euwema (2001) found that officers with higher levels of emotional exhaustion and depersonalization were more likely to use force, whereas subsequent work suggested that moderate levels of burnout may be associated with less dominant or conflict-oriented behavior (Euwema et al., 2004). Our review, to some degree, support this interpretation, that not all types of burnout will lead to the same uniform consequence. While avoidant behaviors could, in some contexts, reduce the likelihood of force, policing situations often require engagement in conflict, and the manner in which such encounters are managed may be influenced by levels of depersonalization.

Two findings warrant particular attention for future research. First, the moderator analysis indicates that the proportion of men in study samples explains a substantial share of the between-study variance in the DP–UoF association. Specifically, studies with higher proportions of men reported weaker correlations. This is a study-level finding and should not be interpreted as evidence of individual-level gender effects. Rather, it points to potential compositional or contextual mechanisms. One tentative explanation is that gender composition may proxy underlying organizational or cultural differences within policing environments. Alternatively, selective attrition processes, such as the “healthy worker effect”, may shape observed relationships if individuals with higher burnout levels are more likely to leave the profession. Given evidence that women exit policing earlier and at higher rates (e.g., Alecu and Bringsrud Fekjær, 2020), such dynamics could influence aggregate patterns. However, these interpretations remain speculative and require direct empirical testing.

Second, the association between burnout and UoF was consistently stronger for attitudinal measures than for behavioral measures. This divergence is methodologically and substantively important. It aligns with longstanding challenges in measuring actual use of force, which is often underreported, context- dependent, and difficult to standardize across studies (e.g., Celestin and Kruschke, 2019; Emsing et al., 2024). Attitudes toward UoF may therefore capture latent dispositions more effectively than observed behavior. At the same time, policing contexts impose situational constraints: officers cannot always avoid conflict, and the decision to use force is shaped by both individual and situational factors. As such, burnout, particularly DP, may influence the quality and proportionality of force rather than its frequency alone.

Despite these important insights, the current meta-analysis is subject to several limitations. Firstly, all studies employed cross-sectional designs, which restrict the ability to draw causal inferences about the directionality of the relationship. It is simply not possible from this meta-analysis alone to understand how burnout and UoF relate to one another over time. Secondly, and most notably, the number of studies meeting inclusion criteria was limited, which constrains the robustness and generalizability of our findings. According to conventional meta-analytical standards (Borenstein et al., 2021), all analyses were underpowered to various degrees and likely unfit to detect smaller effect sizes. While we did detect effects with regards to DP, both EE and PA remained non-significant. DP was the sub-dimension with the most studies, and largest total sample size emanating from those studies. We thus urge readers to interpret the absence of significance from the EE and PA subdimensions with great caution, as both dimensions were underpowered and unlikely to detect small(er) effect sizes that could, in fact, exist in the population of interest. In this context, it is also important to note that the high levels of heterogeneity in the EE dimension further complicate this matter. We were only able to disentangle a smaller proportion of this heterogeneity in the gender analysis of DP but constrained in our analysis of the EE dimension's heterogeneity. Additional robustness tests of all relevant moderators yielded highly non-significant results in the EE analysis.

Taken together, the findings highlight the need for more robust empirical designs. Longitudinal and experimental studies are necessary to disentangle causal pathways between burnout and UoF. Future research would also benefit from larger and more diverse samples, as well as more standardized and behaviorally valid measures of use of force. Importantly, studies should explicitly distinguish between attitudinal and behavioral outcomes, as these appear to capture different aspects of the relationship. We also suggest that future burnout – UoF research should be conceptualized more clearly along the lines of Leiter and Maslach's (2016) developed framework. Leiter and Maslach's (2016) framework provides a useful analytical frame that can disentangle how different subdimensions of burnout affect UoF, and how they are grouped into different distinct clusters and profiles among police officers.

In conclusion, this meta-analysis identifies a small but consistent association between DP and UoF, while underscoring the importance of measurement, study design, and sample composition in shaping observed effects. The findings suggest that depersonalization may be relevant for understanding variation in police UoF, but that this relationship is complex, context-dependent, and not yet fully understood.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Generative AI was only used to refine the language in the manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Salvatore Zaffina, Occupational Health, Italy

Reviewed by: Arun Sondhi, Leeds Trinity University, United Kingdom

Xizheng Xu, Hunan Police Academy, China

Footnotes

Minor adjustments have been made, such as the use of the AXIS protocol for critical appraisal. The AXIS protocol was deemed the most suitable instrument in the context of the design of our collected articles.