Traffic stop profile measures confirmed the implementation of the intervention strategy. Both the relative percent of safety stops and the absolute number of safety stops completed marked increased in Fayetteville in comparison to the measure-specific synthetic control agencies. This increase in the percent of safety stops was matched with a corresponding relative reduction in economic and investigatory stops.
Motor vehicle crash outcomes were all reduced, though confidence intervals were relatively wider. Measures of traffic stop disparities were also reduced, suggesting a focus on safety stops (and relative de-prioritization of investigatory and economic stops) was a viable strategy to reduce Black non-Hispanic disparities in their traffic stop program.
Neither index crimes nor violent crimes changed appreciably during the intervention relative to the synthetic control agencies: three measure point estimates saw small reductions and one saw a small increase, but these nominal changes were much smaller than their associated confidence intervals. The disagreement in direction of the small change violent crime counts (decrease) and rates (increase) demonstrates that the measure was largely unchanged; small variation in population denominators explain the metric direction disagreement and the intervention effect on violent crime was effectively indistinguishable. This study does not provide any evidence of a negative effect on crime for de-prioritizing investigatory and economic stops. However, a more detailed view of the trend of the reduction in the total number of stops during the transition into the intervention suggests the first half of the Ferguson Effect, a reduction in output by some officers in response to community outcry and public attention, may have occurred in the first intervention year. Staffing changes as agency culture changed may also have occurred during the intervention roll-out period, producing or contributing to this reduction in output before the subsequent large increase in safety stops.
These results suggest redesigning a traffic stop program for public health impact may reduce negative motor vehicle crash outcomes, simultaneously reduce some negative consequences of traffic stop programs (e.g. race-ethnic disparities, reduced economic stop burden on communities), and the relative de-prioritization may not have an significant impact on crime rates. Safety traffic stops, especially when directed at high crash areas using regular review and traffic stop GPS data for evaluation, may be a more effective public safety tool than economic or investigatory stops. If investigatory stops can be de-prioritized with little impact on crime, but carry with them negative consequences to community trust, those traffic programs may be de-emphasized even without a relative prioritization of safety stops.
However, these apparent public health wins can be fleeting, as transitions in administrators may bring entirely new or adjusted priorities. Since Chief Medlock’s retirement in 2016, the percent of safety-related stops has dropped and the percent of Black drivers stopped has increased (Open Data Policing 2019). Future analyses may explore whether these new changes are associated with increases, decreases, or neither in crash, injury, and crime measures. Adherence to consistent public health priorities, especially when those relative priorities and implicit logics are made explicit, may help administrators transition while keeping interventions consistent.
Negative consequences of traffic stops
This study posits a relationship between certain stop types and public health outcomes under a conventional framework. However, that conventional framework ignores or downplays the real, negative consequences of traffic stop enforcement in practice. Regulatory and equipment stops, and their associated fines, are a direct form of criminalizing individual and community economic poverty. Beyond the immediate impacts, the harm of economic stops creates a negative spiral operating within communities collectively and individuals specifically, extracting wealth and people’s bodies from low-income communities as the inability to pay mounting traffic tickets escalate into denied registration and warrants for arrest. The United State Justice Department Civil Rights Division cited this extreme and racialized extraction of wealth through traffic stops in its review of the Ferguson Police Department (US Department of Justice, Civil Rights Division 2015). When used unaccountably (e.g. recording no GPS data, as is the norm in NC), moving and safety violation stops can be enforced in an area with few motor vehicle crashes to justify them. Lastly, investigatory stops may have strikingly low contraband hit rates or racialized application (Baumgartner et al. 2018a, 2018b), which subject some to antagonistic law enforcement interactions over years (Peralta and Corley 2016) without contraband to show for the interaction.
Beyond the serious financial and carceral consequences, at their most severe, traffic stops can have fatal consequences for motorists, even when unarmed. Sandra Bland, an unarmed Black woman who died in jail after a routine traffic stop, had multiple other unpaid traffic tickets at the time of her arrest, including for operating a vehicle without a license and lack of insurance (Katy Smyser 2015). Walter Scott, an unarmed Black man, was shot to death, in the back, by a South Carolina police officer after a traffic stop for a non-functioning brake light (Blinder 2017). Philando Castile was pulled over forty times, for reasons including speeding, driving without a muffler and not wearing a seat belt, in the years running up to his fatal shooting during a traffic stop (Peralta and Corley 2016). An uncritical increase in traffic stop enforcement means increased interactions with law enforcement, creating more opportunities for escalated and fatal encounters that may disproportionately impact low-income people and people of color given structural disparities and both implicit and explicit bias. The associated loss of community trust has real public health consequences, including fewer calls for timely emergency services (Desmond et al. 2016). Beyond the negative consequences acknowledged to be more objective, public safety interventions driven by traffic stops should acknowledge the disparate, subjective, emotional experience drivers of color experience. Recent studies now document how these disparities in chronic stress get biologically embedded (i.e. “get under the skin”) and have measurable and negative consequences for individual health (Hertzman and Boyce 2010; Krieger et al. 2015; Nuru-Jeter et al. 2009), including specifically symptoms of post-traumatic stress disorder associated with increased interactions with police (Hirschtick et al. 2019).
Program effectiveness, program efficiency
Central to this discussion are questions of absolute and relative intervention efficacy and efficiency. In Fayetteville’s case, their safety stop program was likely more efficient because of its use of crash data to inform prioritization of intersections and the geocoded stop data to ensure intervention fidelity. However, safety related traffic stops are not the only method to reduce motor vehicle crash injuries. The efficacy of even maximally efficient traffic stop programs must be weighed against strategies from other sectors such as public education campaigns and built environment investments, which may be either or both more efficacious and cost-efficient (Centers for Disease Control and Prevention, National Center for Injury Prevention and Control 2019). Likewise, focusing on policing interventions for public safety in the absence of infrastructure improvements, given historical (e.g. redlining) and present disparities in those investments raise equity concerns (Rothstein 2017).
When considering equitable investment in communities, this intervention to reprioritize traffic stops may best be a stop gap response to immediately reduce disparities and promote traffic crash outcomes but is not an ultimate solution. Though the intervention reduced racial disparities in Fayetteville compared by 21% of what they could have been, Black drivers still experienced over twice the incidence of traffic stops per vehicle miles traveled as White non-Hispanic drivers at the end of the study period. If not considering alternative interventions that may be more efficient, efficacious, or equitable, an investment in traffic stop programs in isolation may be capable of reducing motor vehicle crashes further but may require a totalitarian police state model stopping nearly all drivers for every possible infraction. Intervention considerations should include not only comparison of the positive efficacy and financial cost of programs but should weigh the negative collateral or intentional damages done. Traffic stop programs may be intentionally phased out or scaled back alongside infrastructure investments and other interventions that carry fewer negative and inequitable consequences to remain in alignment with public safety needs.
The same principles are true when considering other public safety outcomes: though policing has seen large funding increases and expanding scope of practice (Hinton 2016), policing should not be seen as either a panacea overall or the most efficacious intervention for non-vehicular crime and injury specifically. Police do not replace mental health workers, social workers, or public health workers capable of implementing evidence-based programs at the individual and community level for substance misuse and violence-related outcomes. As law enforcement agencies are increasingly accountable to the efficacies and efficiencies of their programs, it is in their best interest to focus on programs, including carefully-designed traffic stop programs, that have fewer negative consequences, more equitable outcomes, improved efficacy, and efficient implementation when compared to interventions from other sectors.
Program priorities and the relative worth of life
In both law enforcement and public health, we implicitly and explicitly prioritize certain causes of disease, injury, and death over causes. Our prioritizations are revealed by our evidence and assumptions of efficacy and efficiency, by program funding and implementation, and ultimately by community investments enabled by political power. Even ignoring other sectors and intervention strategies besides traffic stops, police may compare the cost and efficacy of traffic stop programs in preventing injury and death by motor vehicle crash to preventing injury or death during a burglary, assault, homicide, or suicide. When considering who is targeted by interventions, public health recommends considering the burden of traffic stop preventable injuries, the exposure to traffic stops in the form of patrols patterns and priorities, and distributions of both exposure and outcome across population subgroups (Ward et al. 2019) alongside efficacy and cost. Because of unequal distribution of outcomes, exposure to interventions, differences in intervention effectiveness and efficiency, these priorities come to represent the relative value of lives by race-ethnicity and socio-economic position. As example, if community investment (including through law enforcement and traffic stop patrol programs) in preventing deaths by assault grossly outweighs investment in prevention of deaths by motor vehicle crashes, overdose, or heart disease, and especially when the underlying burden of assault injuries and mortality is comparably low, we implicitly prioritize the health and lives of populations seeking to prevent assault over other public health priorities and other populations.
These prioritization dynamics operate at multiple levels within and above agencies: within agencies as individual officer, patrol team, and precincts patterns; and above as clusters of agencies, statewide, nationwide, and between countries. At the national level we see these prioritizations in the focus on criminalizing drug use and addiction in urban, Black communities in the 1980s that lead to disproportionate incarceration of Black people at a level rarely seen anywhere else in the world (Hinton 2016). In contrast, the multiple phases of the opioid epidemic since 2000, hitting more (but not exclusively) rural and white communities, has been comparably treated as a public health crisis rather than a criminal justice one (Bailey et al. 2017; Netherland and Hansen 2017). Though this intervention analysis provided some contextual factors at the agency level, future research should not be limited to either implicit bias at the individual or policy effects at the agency level, but instead should continue to focus on questions or program priorities and implicit worth of human life at multiple and interacting levels.
Whether legally defensible or not, traffic stop programs may still be considered unjust and burdensome. They may ignore racial disparities in financial hardships, erode community trust, embody community stress, and trade injury and loss of life outcomes in some communities to promote or appear to promote the well-being of other communities. Even within the same community, for example, a seatbelt program that extracts large amounts of financial resources may cause serious harm to individual and community health and may outweigh the injury prevention benefit. Co-designing traffic stop programs along with impacted communities may alleviate some of these negative outcomes, though likely not all given the multiple underlying dynamics at play (Smith and Holmes 2014). It is precisely these implicit disparities in the value of people’s experiences, and ultimately their bodies and lives, that drives associated policy platforms calling for the end of criminalization and dehumanization of Black and low-income communities (The Movement for Black Lives 2019).
We argue that public health has a fundamental interest in detailed traffic stop data given associated public safety outcomes and equity considerations under both conventional and anti-racist frameworks (Ford and Airhihenbuwa 2010). However, not all states maintain active traffic stop databases like North Carolina’s. Further, most active traffic stop databases that do exist were started recently. When contrasted with many other public health surveillance systems, limited data on traffic stops suggest a relatively limited oversight of law enforcement activities. Public health has already acknowledged that data on deaths caused by officers are public health data that can and should be maintained (Feldman et al. 2019; Krieger et al. 2015), and that collecting law enforcement data in general is fundamental to accountability and trust (McGregor 2015). Data collection on traffic stops should also include some within-agency spatial component, as Fayetteville has elected to collect, such as spatial coordinates or an address or intersection that could be retroactively geocoded. Besides promoting accountability and transparency, such detailed data on traffic stop programs also benefits police agencies. Spatially-referenced traffic stop data can inform prediction and intervention models of public safety events like crashes and violent assaults and also ensure accountability within the agency and to community priorities. GPS tools for spatial referencing are increasingly low-cost, included in most cell phones, and retrospective geocoding are inexpensive. Recognizing the decreasing cost and increasing utility, the National Institute of Justice (NIJ) and the Bureau of Justice Assistance collaborated with the National Highway Traffic Safety Administration (NHTSA) to promote the Data-Driven Approaches to Crime and Traffic Safety (DDACTS) (Crime Mapping for DDACTS - Crime Mapping and Analysis NewsCrime Mapping and Analysis News n.d.) program. Agencies that capture detailed traffic stop data would be following these NIJ best practices.
As an example of the equity implications of public safety interventions, NHTSA put out a manual for state highway safety offices that included evidence of law enforcement traffic stop activities by types of traffic stop (Goodwin et al. 2015). This document informed updates of CDC guidelines around motor vehicle safety interventions (CDC Injury Center Motor Vehicle Safety 2019). Included as an evidence-based intervention are “a saturation patrol (also called a blanket patrol, ‘wolf pack,’ or dedicated DWI patrol)” (Goodwin et al. 2015). Likewise, movement from secondary to primary enforcement of seatbelt laws (e.g. allowing seatbelt ticketing when no other infraction is present) is associated with more seatbelt usage and reduced traffic crash fatalities. But when public health advocates for saturation approaches do not acknowledge and measure disparities, these approaches may disproportionately burden under-resourced communities with the negative consequences of traffic stops. And, without some within-jurisdiction accountability, agencies are free to use their discretion to distribute DWI and seatbelt patrols into neighborhoods for other reasons. Those neighborhoods may not have the political and economic capital to fight in court, may not equitably weather the negative effects of such saturation interventions, and may not have the associated needs or see the consequent benefits to their public health outcomes.
This study has multiple limitations. Since only one agency enacted the intervention, our findings are suggestive but limited by sample size in many ways. For instance, in Fig. 2, because placebo tests are limited to the control pool of 8 non-intervention agencies, permutation p-values could only be in multiples of 0.125. Moreover, the relatively small control pool was only selected to provide adequate comparison to Fayetteville, i.e. by ensuring a spread of most metric around Fayetteville. Therefore, in some cases, some placebo trends and related tests were unstable for some metrics when no linear combination of other control agencies could remotely match the placebo agency. As example, no linear combination (weights adding to 100%) of smaller agencies can effectively model Charlotte, the largest agency with twice the population, twice the traffic stops, and three times the index crime count; if Charlotte were the agency of interest, it would require a different control pool.
Even in the case of Fayetteville, though the control pool provided adequate coverage for most metrics, one metric (the percent of Black non-Hispanic traffic stops) was best represented by a 100% weighted match against a single city agency in Durham, North Carolina. This effectively reduces the more nuanced synthetic control method to a simpler difference-in-difference model comparing a single intervention city to a single control city. In this case, Durham may be well suited as a control city to Fayetteville on most metrics (see Table 2) in this case, including closely matching this metric (e.g. 57% of traffic stop drivers are black in both cities in the pre-intervention period). However, this single control city analysis is not as robust to city-specific variation. If a group of agencies were to adopt this prioritization formally or smaller variations in these metrics were considered in a national study, results may be more robust. If a group of agencies were to adopt this prioritization formally or smaller variations in these metrics were considered in a national study, results may be more robust.
We do hypothesize that the synthetic control method improved confounding control compared to a simpler difference-in-difference model. However, an approach that incorporated data on more agencies and more covariates under a more detailed confounding control scheme would likely produce more accurate results than our approach of matching on the pre-intervention period. In this case, because of both small numbers of units and a lack of clarity on whether potential covariates were mediators or confounders of the intervention effect on each specific measure, we did not additionally adjust for metric-specific known confounders beyond the confounding control that metric-specific matching on the pre-intervention period provides. For example, while local economic changes associated with changes in a given metric (say, crime) across multiple cities would be adjusted for by comparison to the synthetic control built from cities matched on that crime metric, if Fayetteville had city-specific economic changes unrelated to those otherwise matched cities this analysis would not detect it. However, including time-unvarying or time-varying covariates requires the synthetic control to attempt to match both pre-intervention trends and covariates simultaneously; in sparse models with small sample sizes, this effectively deprioritizes unknown confounder control for (supposedly) known confounder control, should those covariates truly be confounders (and not mediators, etc.). While we did not have that causal clarity on covariates (or sample size) here to make that trade-off, other synthetic control studies with sufficient sample size and covariate clarity should include carefully chosen covariates to better control for local confounding otherwise uncontrolled for by pre-intervention matching. That said, particularly when there is a scarcity of implementation sites and promising interventions, documentation of aspiring anti-racist interventions is worthwhile in the face of these limitations (Jones et al. 2019).
Further, the capture of race-ethnicity in administrative datasets has known limitations (Knox and Lowe 2019). Race-ethnicity is a powerful social construct associated with many associated health disparities (Tsai and Venkataramani 2016), so many we that require dedicated frameworks to harmonize them (Duran and Pérez-Stable 2019). Because of its social construction (Ford and Airhihenbuwa 2018), the meaning of race-ethnicity changes over place and time and can vary person to person even within the same time and place. Health research acknowledges that self-identification may differ from social-identification (Jones et al. 2008). Even in the same person, conceptions of race-ethnicity change over the life course (Mihoko Doyle and Kao 2007). Concretely in this study, the self-identification options in justice databases are limited and may not match driver’s self-identity. Stopping officers may not refer to driver-specified race-ethnicity, notably incomplete in NC driver’s license records (Richard Stradling 2018), but instead fill out form SBI-122 based on their own ascription of the race of the driver. Indeed, there is documentation that in some regions law enforcement officers may knowingly misidentify race-ethnicity in response to scrutiny under new racial profiling laws and accountability that databases would seek to provide (Friberg et al. 2015).