Criminal Justice, Arrests Data, and Structural Racism Measurement for Health Equity Research: Promises and Pitfalls
本文回顾了美国刑事司法系统中种族主义的历史背景，以联邦调查局统一犯罪报告中的逮捕数据为例，讨论了使用这些数据测量结构性种族主义的潜力和局限，并为健康公平研究提供了建议。
Involvement in the criminal justice system is a well-documented health and racial justice issue [1-3]. Racism in criminal justice can be traced back to the 1700s, when policing bodies assembled for the purpose of patrolling enslaved people. As the COVID-19 pandemic and racial justice movements of the 2020s reinforced, people racialized as Black are still disproportionately arrested, killed, incarcerated, and otherwise negatively impacted by the criminal justice system [3-10]. Research on those directly involved in criminal justice systems and their broader communities has grown in concert with literature on racism as a fundamental cause of health and healthcare inequities [11-28]. In this commentary, we provide a brief historical review of policing law, policy, and practice, as critical context on current criminal justice systems and data. We present the case of the Federal Bureau of Investigation (FBI) Uniform Crime Reports (UCR) Age, Sex, and Race (ASR) Arrest and Clearance data to illustrate key considerations when using the data for racism measurement. Finally, we offer recommendations for using federal criminal justice data in health equity research. Scholars across multiple disciplines have advanced theoretical and empirical work on racism and health, highlighting the need for better conceptualization and measurement in quantitative approaches [19, 22, 29, 30]. Lett et al. connect different levels of intervention points within the multiple levels of systemic racism, including within criminal justice systems, and operationalize the measurement of racism [31-33]. Criminal justice systems have been characterized as loci of institutionalized racism, and also critical to the production and maintenance of structural racism, which Dean and Thorpe (2022) define as the “totality of ways… systems and institutions interact to assert racist policies, practices, and beliefs.” [20, 25, 26, 31, 33]. Research on racism measurement has advanced rapidly in the last few years, with the development of institutionalized (single measure and single topic constructs) to structural (multi-dimensional) measures, classification, and use [19]. To capture the role of criminal justice systems, researchers have commonly used secondary data on policing and incarceration. However, much work remains on critically examining these data and measures for their use in health services and policy research. Research in public health has increasingly documented the adverse consequences of U.S. policing practices. Alang and colleagues outlined several mechanisms by which policing impacts Black community health, including injuries and fatalities, activation of adverse stress-response systems from direct and indirect encounters with police, arrests and incarcerations, financial consequences of legal and medical bills, and pervasive oppression and resultant distrust of institutions [34]. Research on individual-level contacts with the police has identified several negative health outcomes, including lower self-rated health [35], reduction in access to, and use of, harm reduction services for substance use disorders [36-38], and exacerbations of multiple mental health conditions [39]. Simckes et al. built upon Alang's work with expanded conceptualization of the impacts of law enforcement agencies at systems and community levels [40], to illustrate the connections between policing and institutional and structural racism. The growth of U.S. policing in the latter part of the 20th century partly arose from the War on Drugs, a political campaign launched in 1971 that introduced new laws criminalizing involvement with drugs and enhanced punitive sentences for all (not only drug) felonies [41]. From 1980 to 1997, incarceration for nonviolent drug offenses alone increased from 50,000 to 400,000; despite subsequent declines, over 350,000 people were still incarcerated for drug offenses in 2023 [42]. Following waves of anti-Black policing and criminalization in media [43, 44], people racialized as Black represented 49% of arrests for drug selling (trafficking) and 36% for drug possession, despite making up only 16% and 13% of actual sellers and users, respectively [45]. Furthermore, other policies institutionalized anti-Black racism in design and implementation, such as sentencing disparities between offenses for crack (widespread in Black communities) versus cocaine (predominant in white communities). As of 2001, an estimated 17% of adult Black men had served time in a state or federal prison, twice the rate of Hispanic men and nearly 6 times the rate of white men [46]. Following expansion of 4th Amendment search and seizure law in the 1960s [47, 48], law enforcement departments oversaw a multi-decade surge in proactive and increasingly violent policing, including through the implementation of pretextual stops [49, 50]. Pretextual stops, also called investigatory stops, differ from traffic stops in that an officer can stop someone to search for contraband without any “reasonable suspicion of criminal activity.” [49, 50] These stops operationalize racism in three ways: by allowing police perceptions of who and what appears to be criminal to determine who is stopped; then determine who is more likely to have contraband discovered, more likely to be arrested and detained for certain crimes; and by providing an opportunity for violence during prolonged stops [49, 50]. The legacy of both laws and policing practices resulted in individuals racialized as Black being more likely to be arrested and convicted for drug crimes despite comparable usage [2, 4, 6, 51]. The FBI UCR program spans over 18,000 agencies across cities, counties, states, tribal territories, and federal law enforcement entities [52]. The UCR Summary Reporting System (SRS) collected mostly monthly agency information counts for ten major offense categories based on the most serious offense in the reported incident, also known as the hierarchy rule (Table 1) [53, 54]. In 1982, the FBI created the National Incident-Based Reporting System (NIBRS), which collects more detailed crime information at the incident level, and reports up to 10 crimes per incident while not following a hierarchy rule [55, 56]. UCR fully transitioned to NIBRS in 2021 [57]. While all states are certified to report NIBRS, this encompasses only 80% of the U.S. population [58]. The UCR Age, Sex and Race Arrests and Clearances (ASR) data, which reports agency arrests data, is derived from SRS/RS/NIBRS and provides monthly data on the number of arrests from 1974 to 2020. In the early years of data collection, around 9000 agencies reported ASR data, but by 2020, approximately 15,000 out of 18,000 participating agencies reported data. The number of arrests is provided as total arrests for a given crime reported in that month and not as unique arrests [59]. Before 2013, the ASR race data categories were White, Black, American Indian or Alaska Native, and Asian or Other Pacific Islander; after 2013, the categories changed to the following: White, Black or African American, American Indian or Alaska Native, Asian, and Native Hawaiian or Other Pacific. Before 2013, ethnicity data were categorized as Hispanic and Non-Hispanic; after 2013, changed to Hispanic or Latino and Not Hispanic or Latino. Age data is reported with both sex and race data, albeit in different reporting categories; with sex, age is reported in a group range (i.e., 15–24), whereas with race and ethnicity data, age is only reported as “juvenile” (under 18), or “adult” (over 18). The ASR data offer an opportunity to construct measures that proxy for racism within policing. Comparisons of arrest data can reflect racially disparate treatment by the criminal justice system, particularly for crimes that rely on pretextual stops and police discretion, such as drug and weapons offenses. In addition to providing a descriptive portrait of policing activity, arrest data can also be indicative of other aspects of racist institutional practices and broader systems failures, such as the arrests of people for offenses involving drugs—which can be indicative of the presence of mental and behavioral health disorders—or for other offenses indicating potential socioeconomic instability like vagrancy and prostitution, in lieu of alternative response to emergency calls, and the wider provision of social services [60-63]. The somewhat standardized data collection on arrest offenses allows for the construction of national racism measures using ASR. Researchers can construct measures similar to other measures of institutional and structural racism used in health research, e.g., calculating ratios of the incidence of an outcome for the Black population to that of the white population [19]. With arrest data, researchers would calculate the arrest rate of each racialized group, for example the number of Black arrestees per Black resident within a geographic unit per year, and likewise for white residents [8, 19, 64]. A proxy measure of racism in the criminal justice system would thus be calculated as the ratios of these two arrest rates to capture the degree to which Black residents are arrested, compared to white residents, accounting for differential population sizes. Furthermore, researchers can capitalize on the longitudinal nature of ASR to document trends over time and across regions, providing opportunities to examine the enduring effects of major policies, including variations over time [65]. The case of the ASR data illustrates the potential for UCR data to advance research on racism and health services and policy research. However, researchers wishing to employ the ASR data should be mindful of limitations in using the data for analysis, particularly in measuring racism. Below, we outline key features and discuss critical considerations. One major limitation of the longitudinal ASR dataset is the underreporting of data from police agencies. As mentioned above, the number of reporting agencies has varied, albeit steadily increased over time, in 1974, only approximately 9000 agencies reported, with that number approaching 15,000 by 2020 (leaving some 3000 agencies without reporting). Additionally, some agencies may not report data on certain crimes without clarification on whether there were no reported arrests or the agency simply elected not to report, meaning that zero recorded data could mean an absence of data or no reported crimes. Furthermore, among those that report, 45%–55% of agencies report data for all 12 months, and some agencies choose to report their data in batches (i.e., grouping numbers by quarter), which can hamper research on monthly trends [59]. In addition to agency reporting practices, arrests are captured without individual identification and thus may represent the same person, and because of the hierarchy rule, it is also not possible to know if the arrestee had additional charges [59]. Concerning racism measurement, researchers must conduct careful exploratory analyses to assess for missingness patterns suggestive of underreporting- particularly if reporting behaviors constitute a contributing mechanism to state or local agency institutional racism. Historically, those aiming to appease political interests by showing “good” crime numbers and, more recently, agencies and officers seeking to lower their reported arrest rates of Black individuals all have incentives to underreport their total activities as well as their race and ethnicity data [66, 67]. Furthermore, although data missingness can suggest instances of agency-level underreporting, there may also be instances of under-reporting within types of charges or other aspects of within-agency underreporting that may not be discernible with the data. Conversely, there may also be circumstances of over-reporting due to local pressures [68] to support the continued growth in local law enforcement expenditures, staffing, and increasingly militarized equipment that has persisted following the initial War on Drugs. While ASR data does include race and ethnicity items, there are multiple limitations in using race data in analyses of several UCR datasets. There is no UCR dataset with consistent race and ethnicity information until the ASR begins in 1974, with the race categories of “American Indian,” “Asian,” “Black” and “White.” The ethnicity categories were “Hispanic” and “non-Hispanic,” but ethnicity has been severely underreported over time [59]. From 1974 to 1980, and 1987 to 2017, fewer than 1% of all agencies nationally reported ethnicity data to be included in ASR. In summary, potentially reflecting the lack of systematic investment and prioritization, the ASR race and ethnicity data are highly limited and inconsistent with federal data collection policies, and have been subject to substantial underreporting, particularly of ethnicity. With respect to racism measurements, race in arrests is derived from officer perceptions- not self-identified by the arrestee [59]. As explained by Lett et al., the assignment of race, instead of through self-identification, is not necessarily “more or less correct” on its own, given that the process of racialization occurs through “different outputs.” In this case, the assigning of race is potentially useful in as a proxy for understanding discrimination based law enforcement officer perception of individuals [32]. However, the absence of self-reported data can introduce potential inaccuracies, such that any related analyses, particularly those that would intend to use the data as a community measurement in analyses of outcomes beyond criminal justice and policing involvement, should include recognition of this limitation. The ASR dataset consists of data organized by state and reported at the agency-month level, where law enforcement agency jurisdictions do not necessarily correspond with other geographic administrative units. Throughout the 1990s, eight agencies nationally spanned more than one county. Conversely, multiple reporting agencies may operate within the same county, such as city police and county sheriff departments, with the abovementioned caveat that not all agencies report their data. From 1990 to 2000, many counties contained one or more reporting agencies, with mean 3.1 agencies per county (SD 4.4). Bureau of Justice Statistics published agency-county crosswalk files in 1996, 2000, 2005, and 2012—however, no other files have been generated for antecedent or subsequent years [69]. Thus, researchers should proceed with caution in attempting to construct area-level measures based on UCR agency-rows. The National Archive for Criminal Justice Data contains county-level imputed data, but prior researchers in criminology have demonstrated substantial conceptual and technical inaccuracies with this approach, such that use of these data for policy analyses is not recommended [59]. The most accurate, valid and defensible construct(s) may be those aggregated to the state level; however, many criminal legal policies, though passed at the state level, are implemented locally with considerable discretion by county. For example, in many states that continue to criminalize the recreational use of cannabis, local jurisdictions—particularly in larger metro areas—have elected not to prosecute cannabis crimes [70-72], and indeed, prosecutorial discretion remains a factor in decisions to enforce criminal laws. Such localized policy and decision making makes local data and analysis preferable, but due to the aforementioned limitations, ASR may be unsuitable for local analyses. Researchers interested in using criminal justice data should carefully consider the promise and the pitfalls, as they would with any other administrative dataset in health services. First, researchers must recognize that criminal justice data is a by-product of racialized systems. For example, by understanding the historical context of policing, data on arrests more closely reflect policing behavior, rather than crime. The data are a product of policies and practices, including where officers police, what they focus their attention on, who they police and how, reflecting individual and system discretion and decisions. Second, systems—from healthcare to criminal justice—also enact racism by their collection, storage, and strategic reporting (or lack thereof) of data. Researchers should attend to systematic sources of missingness and inaccuracies. Below, we discuss potential strategies to address UCR limitations. In 2021, the FBI UCR program retired SRS and transitioned to fully using NIBRS data collection system [57, 59], which collects more comprehensive data, including more data on different types of offenses, all the offenses within a specific incident instead of just the highest-ranked, most serious offense, and more specific location and demographic data about the person(s) involved and where the incident and arrest took place [53, 57]. The implementation of NIBRS may allow for better national-level empirical analysis over time going forward. Furthermore, the FBI could require systematic reporting data on pretextual stops and other encounters, not only arrests, to enhance transparency and ultimately, governance of criminal justice systems. For example, computer aided dispatch data reflects all incidents, including both officer-initiated and community requests. Although these data do not typically include race or ethnicity, researchers could use geocoded linkages to characterize neighborhoods or communities with relatively higher and lower proportions of Black residents. Researchers interested in earlier years and historical events should still look into requesting data from local municipalities and state level such as the Administrative Offices of the Courts. Submitting data requests to state and local governments can be a long and opaque process, but the efforts can often result in more data on policing contact, a major current gap in systematic data collection [39]. Local data may capture the major growth in pretextual stops, as discussed above, as well as other encounters that may have harmful individual and community health consequences that are not captured by arrests. Local data can also provide more precise demographic and geographic information that can allow for linkage to other data sets, along with being used to fill in known gaps in federal data reporting [64, 73, 74]. Finally, researchers interested in these topics should build collaborations with subject matter experts such as criminologists, attorneys and other systems actors, and people with lived experience. Similar to how health service and policy research often involves academic, government officials, clinical practitioners, administrators and patients, such collaborations in criminal justice can ensure understanding of concepts and technical definitions, and differentiations between written policy and practice/implementation in policing, detention, sentencing, and others [75]. New researchers to this area should also be mindful that there are many scholars—particularly scholars of color—both within health services research and in other fields such as sociology and feminist studies, who have been engaged in these topics for decades and produced an of literature on justice systems. services researchers should be in with and and the time to how criminal justice systems are in and empirical analysis on structural racism. national criminal justice provide an opportunity to operationalize structural racism for However, both conceptual and limitations on the and for the measurement of institutional and structural racism. to use criminal justice in the context of racism measurement, health policy and services researchers must to build and understanding of both the opportunities and limitations of with criminal justice data and of how justice data reflect the that institutional racism, along with how data to disparate health The would like to the for Research for at the of and and for the support of this research. We would also like to at the of for providing The no of