Police Violence Exposure Among Young Adults in a U.S. National Probability Sample

Sampling and Design

The National Survey of Poly-Victimization and Suicide Risk was an online cross-sectional survey administered in January 2021 to a national probability sample of U.S. young adults ages 18 to 29 (Fedina et al., 2023). Participants were recruited through the National Opinion Research Center (NORC) AmeriSpeak Panel—a probability-based survey panel of 43,000 households designed to represent the noninstitutionalized U.S. population. NORC used a mixed-modal recruitment strategy including address-based sampling and random digit dialing (landline and cell), which allowed for recruitment of participants without internet access. NORC provided compensation for all participants in their ongoing panel, supplemented with an additional $3 cash incentive for completing this specific survey. The final sample included 1,077 young adults, representing a survey completion response rate of 15.7% and a weighted household recruitment rate of 22.7%, accounting for the panel recruitment rate, panel retention rate, and survey participation rate. Weights were developed for sampling frames to adjust for differential probability of selection and to adjust for eligibility and nonresponse. Institutional Review Boards at the University of Michigan and University of Chicago approved all study procedures.

Measures

Data Analyses

Past-year prevalence of police violence exposure was calculated for the entire sample and within each subgroup of interest. Correlations between binary exposures to each police violence sub-type were assessed using the phi (φ) coefficient. Chi-square tests were used to examine unadjusted associations between each demographic category (e.g., race/ethnicity, age group) and each police violence variable (e.g., physical, sexual). Unadjusted analyses were considered significant at p < .05. Logistic regression was used to test for independent adjusted associations between sociodemographic factors and each police violence outcome. Sociodemographic factors associated with any of the four police violence variables in the unadjusted chi-square analyses were selected for inclusion as predictor variables in the logistic regression analyses. Analyses with missing data were conducted using listwise deletion, due to the low frequency of missing data (<1% of cases for all analyses). Adjusted associations between individual predictor variables and each police violence outcome were reported as odds ratios (OR) with 95% confidence intervals (CI). All analyses were considered significant based on a two-tailed alpha = .05.

In addition to the primary logistic regression analyses, we conducted additional exploratory analyses to inform future work. While potentially under-powered, we tested the associations between race/ethnicity and police violence for effect modification by gender identity, using nested chi square analyses, as a preliminary step toward quantitative intersectional analyses of police violence in the United States.

Main Findings

The national debate in the United States regarding the causes, pervasiveness, and impact, of police violence has suffered from a lack of national data on rates of exposure and its sociodemographic correlates. To our knowledge, this study presents the first data on the past-year prevalence of nonfatal police violence among a national probability sample of young adults. We found that police violence exposure is, when considering the potential severity of the exposure, not rare or unusual among the general population of young adults in the United States, with 4% reporting past-year exposure to physical violence from police officers and 3.3% reporting past-year exposure to sexual violence. Less severe exposures, including psychological violence (e.g., threatening or discriminatory language) and neglectful violence (i.e., failing to prevent violence when expected to do so) were more common, reported by 8.5% and 7.5% of respondents, respectively.

Twelve-month prevalence rates in this study are comparable to those in the Survey of Police Public Encounters, collected in 2017 among adults (age 18+, no upper limit) in New York City and Baltimore using an earlier variation of the same measure, finding similar but somewhat higher rates compared to the present study (DeVylder et al., 2018). The difference may reflect the setting if rates are particularly high in those selected cities compared to the United States as a whole. Further, data from the present study were collected during the COVID-19 pandemic, when social distancing measures were in effect and there may have been reduced time spent in public (and therefore less incidental contact with the police) on average across the population. The data collection period also coincided with widespread protests against police violence following the death of George Floyd, which may have impacted rates of exposure to police violence, with some evidence suggesting that rates declined in areas with widespread protest (Campbell, 2024). While the rates in the present study were lower than those of the Survey of Police-Public Encounters studies, they were higher than those found in the Bureau of Justice Statistics Police Public Contact Survey, which likely reflects our focus on past-year exposures rather than the respondent’s most recent contact with police, regardless of timeframe (Tapp & Davis, 2024). Despite this variability in prevalence rates across studies, even the most conservative estimates translate to a significant population-level impact, cumulating in a substantially greater risk for negative health (Alang et al., 2021, 2023; Quinn et al., 2024; Thompson et al., 2021), mental health (Bowleg et al., 2020; DeVylder et al., 2017, 2018), and related outcomes among those directly exposed and those exposed indirectly through friends, family, neighbors, their community, and other sources (Green et al., 2024; Haile et al., 2023; Kim et al., 2024; Kincade & Fox, 2022; Quinn et al., 2024; Read et al., 2023).

Sociodemographic Correlates

Police violence exposure was broadly associated with race and ethnicity, in that all sub-types of violence were disproportionately more likely to be reported by non-Latine Black respondents and, to a lesser extent (and excluding neglect), by Latine respondents. Ethnoracial disparities have been repeatedly found across prior studies of police violence in the United States, with consistently elevated rates of police violence exposure among non-Latine Black respondents in both Survey of Police-Public Encounter studies (DeVylder et al., 2017, 2018), the Bureau of Justice Statistics Police Public Contact Survey (Tapp & Davis, 2024), and other data sources (Fryer, 2019). All subtypes were also associated with at least one indicator of lower SES, income or educational attainment. Male respondents were more likely to report psychological violence and transgender/nonbinary respondents more likely to report neglect but, surprisingly, no other forms of police violence. Notably, in our preliminary intersectional analyses (which should be replicated in more adequately powered samples), ethnoracial variation in likelihood of reporting police violence appears to be consistent between males and females (excepting neglect, which was nonsignificant in females). Nonsignificant effects for transgender respondents are notably not congruent with the limited past research on this topic, which have broadly reported greater rates of police violence exposure among transgender compared to cisgender individuals (DeVylder et al., 2017, 2018; Stenersen et al., 2022), and may reflect the small sub-sample of this group (n = 25). Data from two earlier prevalence studies, in contrast, found significantly elevated reporting of all subtypes of police violence among transgender respondents (DeVylder et al., 2017, 2018). These two prior survey studies (which were drawn from selected Northeastern cities rather than national samples) likewise found elevated rates of physical and sexual violence exposure among male compared to female respondents, which were not replicated here. Given the small subsample of transgender respondents, null results in the present should not negate these previous findings of elevated rates of police violence exposure among transgender adults, and this question should be explored further in future studies with targeted over-sampling. Statistical power may also explain the nonsignificant results by sexual orientation which, as with any potentially under-powered null result, should be interpreted with caution.

While the prevalence of police violence exposure varied widely by demographic group, it is notable that no group was immune. This is consistent with a prior latent class analysis of data from four Northeastern cities that failed to identify a “no exposure” group (Salas-Hernández et al., 2022). Rates were relatively stable across regions, other than for psychological and neglectful violence which peaked in the south and west (unadjusted analyses only), respectively, and between census-designated metropolitan and nonmetropolitan areas. When using self-reported residence data, neglectful police violence was most common in urban settings in fully adjusted analyses, although no other exposure varied by urbanicity. Taken together, we should consider police violence to be an issue that disproportionately affects ethnoracially minoritized groups and people with lower SES across the country, and one that will likely require federal policy solutions, supplemented by local efforts to address police neglect (i.e., failure to prevent violence when expected to do so) targeted toward urban settings.

Race and Ethnicity: Key Associations on a National Level

The role of police violence in U.S. society has been widely contested. Perspectives generally considered liberal or progressive argue that police violence is either a consequence of societal racism or a deliberate mechanism to reinforce structural inequalities; in contrast, views typically framed as conservative maintain that police violence reflects an equitable response to societal variation in rates of crime perpetration—views that are also highly variable by race in the United States (Cooley et al., 2019; Drakulich et al., 2023; Litterer, 2024; Vitro et al., 2022). Despite these diverging perspectives, police violence is a notable issue that drastically sets the United States apart from other economically advanced nations; for example, cross-national comparison study of 14 countries, the rate of fatal police shootings in the United States was approximately 5× greater than that of Canada and 13× greater than that of Netherlands, the highest-ranking European nation in the study (Hirschfield, 2023). Several countries in that study (i.e., Denmark, Finland, Spain, and Norway) reported no police killings, relative to 1,021 in the United States. Further, police violence in the United States is significantly over-represented among ethnoracially marginalized groups, particularly among non-Latine Black Americans (GBD 2019 Police Violence U.S. Subnational Collaborators, 2021). These long-standing disparities are well documented and further supported here by national probability sample data and adjusted analyses, which indicate race as the factor most consistently associated with all types of police violence, even when considering SES and other marginalized statuses. This carries significant structural implications, raising critical questions about (1) the presumed safety of policing given its continued operation as a site of racialized violence, and (2) the responsibility of researchers in advocating for systemic change to address the structural racism underlying this violence.

Racial disparities in exposure to police violence may arise through structural racism operating through multiple pathways, including residential segregation, concentrated disadvantage, and racially stratified patterns of police surveillance that contributed to an increased probability of, and frequency of, police contact among Black and other marginalized populations (DeVylder et al., 2022; Peterson & Krivo, 2005). As such, police violence functions as a population-level stressor with both direct and vicarious exposure pathways, consistent with fundamental cause frameworks and contributing to racial disparities in mental and physical health outcomes (Phelan & Link, 2015; Williams & Mohammed, 2009). Prior epidemiologic evidence demonstrates that these structurally patterned exposures persist independent of individual-level behavior or criminal involvement, supporting the interpretation of race as a proxy for differential exposure to institutionalized risk rather than a causal attribute (Alang et al., 2017; F. Edwards et al., 2019; Ross, 2015).

Clinical and Policy Implications

Previous studies have documented the extensive clinical correlates of direct and indirect police violence exposure (for review, see DeVylder et al., 2020, 2022; Haile et al., 2023; Herd, 2025; Motley et al., 2024), although there are no current guidelines informing how to address this issue in clinical practice. This is especially of concern for young adults, as police violence exposure may shape their sense of safety during a critical developmental period. At the minimum, the high prevalence rate and notable ethnoracial disparities suggest the need to screen for this exposure and offer appropriate health and mental health support for marginalized groups, possibly through primary care or emergency departments. Such screening should focus on sexual and physical violence, which has previously been shown to be particularly impactful in terms of its impact on mental health (DeVylder et al., 2017, 2018), although significant psychological or neglectful violence (e.g., if the injury was due to inadequate police intervention) can also have an impact and may likewise benefit from clinical identification and response. The development of responsive and targeted interventions is warranted given the relatively high frequency of exposure to police violence in the United States, but as of now no specific interventions exist to the best of our knowledge.

The centrality of structural racism as a key mechanism underlying disparities in police violence exposure suggests that public health or policy interventions to address or prevent systemic and structural racism may in turn reduce rates of police violence exposure and related harms in the United States. Further, policy change may address other related factors that contribute to police violence (e.g., qualified immunity policies, firearm proliferation), which have been highlighted in recent theoretical work (DeVylder et al., 2022; Haile et al., 2023) but have generally not been empirically examined in relation to this issue. Supporting efforts in this direction, there is evidence that police violence perpetration is also impactful on police officers themselves by increasing the likelihood of PTSD symptoms (DeVylder et al., 2019). Thus, it can be considered in the best interests of both civilians and police departments to address this pertinent issue.

Considerations Regarding Measurement

The prevalence rates found in these data may be perceived as unexpectedly high and, given the paucity of existing data on this topic, difficult to put into context. This necessitates additional comment about the measures used to assess police violence exposure. The Police Practices Inventory, developed by our research team due to the lack (at the time) of a validated self-report measure of police violence exposure for use in survey studies, has been previously shown to have acceptable construct validity and test-retest validity in the Survey of Police-Public Encounters studies (DeVylder et al., 2017), with some evidence that positive responses (i.e., “yes” to survey items) correspond with supporting open text descriptions (i.e., write-in responses) of actual incidents (T. Edwards et al., 2023).

That said, validity and reliability does not necessarily inform us about the internal and subjective cut-off that respondents have in mind when determining whether to endorse “yes” or “no” in response to particular items. Therefore, while the probability sample design does conceptually allow us to extrapolate these prevalence rates to the general U.S. young adult household population, this should be done with caution due to the possibility that the threshold for what is being considered physical/sexual violence here may be low, potentially leading to over-estimates or an over-representation of milder incidents of physical/sexual contact. That said, a high prevalence of negative police encounters is feasible given prior evidence that police stops drastically outnumber police arrests, and that each stop has some potential for contact that may be reported as excessive or violent (https://www.ppic.org/publication/racial-disparities-in-law-enforcement-stops/). Future research should continue to refine measures of police violence and aim toward replicating the prevalence rates presented here.

Limitations

Although the AmeriSpeak platform of NORC is a highly regarded data source, the response rates for selection into the panel and into our sample are low, posing a challenge to external validity, although, the demographic distribution of the data does correspond closely to the Current Population Survey. Nevertheless, our prevalence results would benefit from replication in different data sources. Further, recent analyses of more than 20 years of survey data have shown that response rates (which have stably declined over that same period, across survey modalities; Jabkowski & Cichocki, 2024) have negligible effects on significance or effect sizes of tests of association (Beehr et al., 2024), the primary focus of our study. Another potential limitation is the use of cross-sectional data, although considering that most of our exposure variables are either unambiguously present prior to the outcome (e.g., race/ethnicity), unlikely to be impacted by recall bias (e.g., educational attainment), or drawn from census data (e.g., residence in a metropolitan area), the cross-sectional design likely yielded similar associations as would be found in a prospective cohort design. Longitudinal designs may also allow us to understand the role of contemporaneous contextual effects (e.g., the widely publicized death of George Floyd shortly before this survey was administered), which may affect external generalizability, particularly in the context of shifting public attitudes.

The measures of police violence used in the present study are more detailed than in some prior studies (e.g., the National Survey of American Life used a single item from the Major Discrimination Scale; Oh et al., 2017), but still may not capture the nuances of interactions between the police and the general public. While the PPI is a validated measure of police violence exposure, it relies on the subjective experience of violence exposure and does not attempt to differentiate between unnecessary force and that which may be considered justified by the situation (e.g., threat toward an officer). The PPI has been validated in the general adult population, but more testing is needed to examine its measurement invariance across key demographic groups, including whether items are interpreted similarly across ethnoracial and income groups, and between urban and rural areas, all of which may influence respondent perceptions of what qualifies as normative behavior or excessive force by the police. The version of the PPI used in this study also did not distinguish between physical violence with or without a weapon, as in the original measure, due to space constraints in the survey. As noted above, it also uses relatively broad definitions of each subtype of police violence (e.g., including both strip searches and sexual assault as “sexual violence”), which could be improved in future studies through more detailed quantitative measures or through validation with qualitative interviews. We also were not adequately powered to separately examine singular and multiple past-year exposures to police violence. In addition, alternative measures may more directly capture discrimination- or racism-related aspects of police violence (e.g., see measure development studies by Motley, Roh, et al., 2025 and English et al., 2017). Finally, some of the demographic items (e.g., education) may have benefited from a more inclusive list of response options (e.g., levels of college education).