7

Racial Bias and Disparities in Proactive Policing

The high rates at which non-Whites are stopped, questioned, cited, arrested, or injured by the police present some of the most salient criminal justice policy phenomena in the United States (Kochel, Wilson, and Mastrofski, 2011; Lytle, 2014). Because these kinds of police contact are associated with at least some forms of what is known as proactive policing, recognition of this reality is an important starting point for this chapter. Additionally, because many proactive policing strategies by design increase the volume of interactions between police and the public, such strategies may increase the overall opportunity for problematic interactions that have disparate impacts.

Concerns about the interaction between race and policing are not new. For example, researchers have been studying differential stop and arrest rates across demographic groups—and more generally, racial disparities in criminal justice involvement, offending, and the likelihood of becoming a crime victim—for several decades (see, e.g., Sampson and Lauritsen, 1997; Tonry, 1995). Nonetheless, several recent high-profile incidents of police shootings and other police–citizen interactions caught on camera and viewed widely have made questions regarding basic fairness, racial discrimination, and the excessive use of force of all forms against non-Whites in the United States a pressing national issue.

In considering these incidents, it is important to note that the origins of policing in the United States are intimately interwoven with the country’s history of discrimination against non-White people, particularly toward Black people. From the tracking and kidnapping of enslaved Black people (Campbell, 2012) to the regulation of Black movement (Loewen, 2005) and

the criminalization of Black bodies for the purpose of economic exploitation (Lichtenstein, 1996), police officers have often been the enforcement arm of both explicitly racist and tacitly discriminatory norms and laws. Although some of the more egregious historical practices ended a long time ago, others ended later and within the living memory of many Americans—and all are remembered as part of the collective history shared by Black and other non-White communities. From this perspective, it is easy to see how the nation’s history is intrinsically linked to misgivings some non-White Americans continue to have about possible police animus and racial bias. And it is by no means clear that explicit animus-driven biases against non-Whites, or examples of racial animus by the police, are a thing of the past. There are certainly many examples of such problems in specific police departments, and some police agencies, as we noted in Chapter 3, have entered consent decrees to address U.S. Department of Justice findings of racial disparities in outcomes and racial bias in police practices.

The purpose of this chapter is to explore whether and to what extent proactive policing policies are deployed in a racially disparate way, if racial differences in implementation are due to racially biased behavior, and if so, what the motivation is for the bias. But before examining the evidence on these questions, we begin by defining and discussing the terminology used throughout the chapter.

Racial Disparity Racial disparities refer to objective differences that exist in the real world. The report uses the term racial disparity to denote outcomes that differ by race or ethnicity. For example, if in a certain community, Black people experience greater levels of poverty than White people and per capita, Black people are arrested more frequently for violent crime than White people, then these would be racial disparities in poverty and in arrest rates for violent crime. A critical point is that these differences can be discussed without assuming that race, per se, gives rise to the observed differences. For example, Black people may be arrested more frequently in part because they experience greater poverty.

Racially Biased Behavior As used in this report, the term racial bias refers to a difference in a person’s behavior that is attributable to the race or ethnicity of another person. For example, if a police officer decides to stop and frisk Person A (who is Black), but does not stop Person B (who is White), and if the officer bases that decision entirely or in part on race, that behavior would constitute racial bias. Racial profiling is a subset of racially biased behaviors, as defined by the committee (see Chapter 3).

To be clear, racial bias refers only to behavior and as used in this report is entirely agnostic as to the psychological motives or other causes that gave rise to that behavior. Potential causes include racial animus, statistical prediction, or other risk factors (see below).

Racial Animus The report will use the term racial animus to describe

negative attitudes toward a racial or ethnic group or toward members of such a group. For example, an officer may dislike Black or Latino people, and this attitude may lead to the racially biased behavior of an officer stopping Black or Latino people more frequently than White people with otherwise identical characteristics to those stopped. Racial animus thus refers to an internal, mental evaluation of individuals or groups based on race. Note that racial animus may give rise to racial bias in behavior, but it is certainly possible that an individual who harbors racial animus does not act on it. In such a case there would be racial animus but not racial bias in behavior. This report will use the phrase “racial animus” synonymously with “racial prejudice,” although social psychologists differentiate between the two dispositions.

Statistical Prediction The report will use the term statistical prediction to identify racially biased behavior that is due to individual or group predictions of behavioral outcomes. For example, statistical prediction occurs in the case where there is racial bias in the choice of individuals to stop on the street because of an assessment that Blacks and Latinos have different likelihoods of carrying weapons. Economists call such prediction “statistical discrimination.”

Situational Risk Factors for Racially Biased Behavior Racially biased behaviors may arise from racial animus, statistical prediction, or features of situations that facilitate differential treatment based on group membership. For instance, social psychologists argue, as we detail later in the chapter, that persons who are forced to act “quickly” may act consistent with their implicit biases, resulting in racially disparate behaviors.

With regard to motivations for racially biased behaviors, we want to briefly address the distinction between explicit and implicit attitudes. The term explicit is often used to describe attitudes and beliefs that are consciously and intentionally endorsed by the individual. By contrast, implicit is often used to describe subtle responses that are not necessarily consciously accessible to the individual and (if they are accessible) may not be endorsed. Racial animus and statistical prediction may both exist as conscious or “explicit” processes. An officer may knowingly dislike Latino people or may consciously believe that Black people commit more crime. But these processes may also operate without conscious awareness. For example, an officer may consciously espouse the idea that Black people are as good as White people and may sincerely believe that, for example, a Black and a White person in the same neighborhood wearing the same type of clothes are equally likely to be carrying weapons (meaning that there is no statistical prediction that would cause biased behavior). Still, that officer may encounter a young Black man on the street and experience a momentary negative “gut” reaction or somehow think that the individual looks suspicious. The psychology literature often refers to these kinds of

reactions as implicit biases; in the terminology adopted by this committee, they are examples of racial animus or prediction. By whatever name, they are attitudes that may affect behavior.

The chapter begins by discussing the challenges involved in measuring racially biased behavior and identifying its causes. It then acknowledges the high levels of police distrust found in some non-White communities—particularly Black communities. This distrust is discussed in the context of the historical relationship between police and Black communities, through which current perceptions of legitimacy and concerns about racial animus have been forged. The next section discusses reasons that proactive policing may be associated with racial disparities or racially biased behavior. The chapter then examines the relevance and possible implications for proactive policing of racially biased behavior driven by racial animus, identity threats, and other psychological processes. Finally, the chapter focuses on economic, statistical, and sociological studies, reviewing the research on racial profiling (a subset of biased behaviors that is of particular importance for policing) and outlining a framework for thinking about the potentially racially disparate consequences of proactive policing strategies.

In reviewing each of these issues, the concern of the committee is with evaluating available evidence on whether and how issues of race are intertwined with the policies and practices of proactive policing. In many cases there is little informative quantitative data on whether the use of proactive policing is influenced by the racial or ethnic identity of citizens in a causal sense. We call attention to this lack of data because it is especially troubling, given the importance of these issues in American society and the evidence of racially disparate policing in the ethnographic and descriptive literature.

MEASURING DISPARITIES, BIAS, AND THE MOTIVATIONS FOR BIAS: ISSUES AND CHALLENGES

Identifying the role of race in someone’s decision-making process is a complicated task, and determining the motive(s) behind another person’s observable action is even more complicated. For example, a police officer may decide to stop and question or frisk a Black citizen but may decide not to question a White citizen, creating a racial disparity in stops. Indeed, many police departments across the country collect data on officer-citizen interactions, including characteristics of the individual stopped such as race, ethnicity, gender, and age; consequences of the interaction whether it be a citation, arrest, or a warning; information on whether the individual was searched; and in some instances the consequences of searches. Using these incident-level data, it is easy to tabulate the racial composition of those stopped, searched, arrested, cited, and so on. Studies that do such tabulations tend to find that non-White people comprise a large share of stops.

For example, in Harcourt’s (2007) review of this research, the author noted studies of Maryland in the late 1990s where Blacks comprised 63 percent of all stops along Interstate 95, a study in Missouri in 2001 where Blacks and Latinos accounted for 75 percent of stops, and a study of Volusia County, Florida, where non-Whites accounted for 70 percent of stops and 80 percent of searches along Interstate 95. See also all the studies discussed in McMahon and colleagues (2002).

Based solely on measures of officer’s behavior, however, it is impossible to know whether this behavior was actually racially biased.¹ If the Black and White pedestrians, for instance, acted differently as the officer approached (e.g., nervous versus calm), or if the officer encountered them in different surroundings (at night in an alley versus at noon in the park), or if the officer was searching for a suspect described as Black, an objective observer might conclude that the officer was simply responding to the situation at hand—that is, the officer was not behaving in a different way toward each citizen because of their different races.

Given the various sources of racial disparities in police–citizen interactions, how does one assess whether a disparity in outcomes reflects racially biased behavior and then identify the motivation for a bias assessed to exist? The existing empirical research on racially biased behavior answers this question in one of three ways. First, researchers have identified situations where the only plausible difference in encounters is the race of the citizen. This can be accomplished through randomized trials in a laboratory setting or by using observational data in a regression analysis framework where researchers attempt to statistically adjust for other factors that are correlated with race and may also influence an officer’s decision. A second approach, benchmark analysis, involves comparisons of policing disparities to various population and/or behavioral benchmarks. Third, researchers attempt to disentangle biased behavior that is motivated by statistical prediction from

___________________

¹ The racial composition of stops tells us little about whether there is evidence of racial disparities. For example, the proportion of stops that are of Black people effectively reveals the likelihood that a person is Black among those who are stopped by the police. A more relevant statistic is the likelihood that Black people are stopped by the police and how this likelihood compares to those for members of other racial groups. Moving from what can be estimated with police stop data (e.g., the likelihood of being Black conditional on being stopped) to the more relevant stop rate (e.g., the likelihood of being stopped by police conditional on being Black) requires additional information. To be specific, suppose that the variable B equals 1 for individuals who are Black and 0 (zero) otherwise and that the variable S equals 1 when an individual is stopped by the police and 0 otherwise. The likelihood that a stopped individual is Black is thus given by the conditional probability P(B = 1|S = 1). We want to learn the likelihood that a Black person is stopped, given by the conditional probability P(S = 1|B = 1). To do so, we need additional information; specifically the unconditional probability of being stopped (P(S)) and the unconditional probability of being Black (P(B)). With this information, the likelihood that a Black person is stopped can be calculated via Bayes’s theorem.

other sources using an assessment of the “productivity” of police-citizen interactions, where productivity is measured by whether contraband or some other indication of illegal activity is uncovered.

Counterfactual-Based Measures of Bias

Studies of behavior in a simulated laboratory environments offer the benefit of studying how people make decisions in situations where, by construction, the only variable that differs across encounters is the race of the subject. By manipulating the images and information given to subjects, social psychologists are potentially able to separate different motives for biased behavior. This is rarely possible in nonlaboratory regression-based analyses, which attempt to quantify the role of other, nonrace-based factors in an officer’s decision and use the magnitude of those estimated relationships to extract an estimate of the influence of race.

That said, laboratory experiments suffer from problems of external validity. Even in the best, most immersive video simulation of an interaction with a hostile suspect, officers do not fire actual bullets at an actual human being, and no suspect really fires back. Further, officers are aware of the fact that they are in a simulation, and they know they are being monitored, so their behavior in an experimental study may differ from their behavior on the street.

Benchmark Measures of Bias

Benchmark studies effectively compare interracial differences in the likelihood of being stopped, employing data on the underlying racial composition of the population at risk of being stopped in conjunction with police stop data to estimate these conditional probabilities. Of course, there is much debate regarding what constitutes the appropriate population benchmark, with broad benchmarks subject to the criticism that the researcher is not properly identifying the population at risk and overly narrow benchmarks subject to the criticism that the definition of who is at risk may itself be a function of racial animus.

For example, suppose the outcome of interest was arrest-related deaths. One might argue that those at risk were those who were arrested by the police, making the race of the arrested population the relevant benchmark. However, differential arrest rates across races may reflect geographic differences in enforcement or racial bias in arrest decisions of law enforcement. To the extent that different enforcement practices contribute to racial disparities in arrest rates, using the racial composition of arrests to benchmark deaths in custody would understate the racial disparities in the risk of dying while being arrested, independent of any racial disparity in offending.

If one were interested in inequality more broadly, inclusive of how differences in poverty, educational resources, geographic segregation and isolation, and the nation’s social history contribute to disparities in adverse outcomes, one might prefer the broader benchmark based on representation of the general population. However, this broader benchmark would be inappropriate for isolating the degree to which racial disproportionality in the outcomes of interest is due to disparate treatment by the police, since the societal and environmental factors included in the broader, general population benchmark will contribute to racial disparities in offending that are independent of policing practice.

Outcome-Based Measures of Bias

The third common methodological approach to studying biased behavior tests for differences in the outcomes of police–citizen interactions by the race of the citizen. Researchers performing outcome-based tests reason that, to the extent that the productivity of stopping, and perhaps searching, an identified demographic group differs from the productivity associated with other groups, the police would enhance public safety by either diverting resources away from the group in question (if the productivity rate for the group is below the average for all stops) or policing this group more intensively (if the productivity rate is above the average).

For example, suppose that an analysis of searches conducted for a given municipality reveals that drugs or other contraband are discovered in 10 percent of searches in which the citizen is Hispanic but 20 percent of searches in which the citizen is (non-Hispanic) White. Using the productivity argument outlined above, one might infer that the police are searching too many Hispanic and not enough White people to “maximize productivity”; reallocating enforcement resources from Hispanic stops to White stops would uncover more contraband. An implication of this line of reasoning is that if the rate at which contraband is discovered is equal across the two groups, then the police are not “overpolicing” one group relative to the other. In other words, the outcome test offers an empirical prediction under the null hypothesis of no bias: hit rates should be equal across racial groups. In this context, an unequal hit rate is frequently interpreted as evidence of animus-driven biased behavior; the assumption here is that statistical prediction (by the police) would generate disparities in treatment that improve the allocation of police resources, whereas racially biased behavior that is not driven by statistical prediction is likely to be driven by animus (although, as noted above, there are other causes of racially biased behavior).

Of course, this line of reasoning is too simplistic because it glosses over the potential illegality of such a biased enforcement scheme and masks a key methodological weakness of the outcome test. Regarding the first point,

police practices that generate equal hit rates may prove to be unconstitutional to the extent that they are based on a strategy that targets members of a specific racial group for enhanced scrutiny. For example, suppose that young Black men are objectively more likely to carry prohibited drugs than young White men. Suppose further that young men of all races are less likely to carry drugs when the likelihood of being stopped by the police is high. Local police may rationally choose to stop Black men at a higher rate than White men and to do so until the “hit rate” (the rate at which stops and searches uncover contraband) is equal across these two groups. Doing so would minimize the amount of drug carrying, and perhaps distribution, for a given level of enforcement. However, this policy ultimately would generate a higher probability that young Black men who are not carrying are stopped and searched by the police, with the disparate treatment of Black men having a measurable disparate impact. Moreover, to the extent that officers act on an a priori assessment that Black males are more likely to be carrying, this may engender differential and perhaps less respectful and even abusive treatment of the group that is presumed guilty. The interviews of poor Black teenagers in St. Louis conducted by Brunson (2007) revealed that these young people perceived that police officers often assumed they were up to something, regularly stopped and searched them without probable cause, and often used unjustifiable violent force. In other words, even if profiling to equalize hit rates increases efficiency in terms of making arrests that generate contraband discoveries, it may certainly generate unconstitutional searches and abuse in the process, and with disparate impact.

The methodological problem associated with the empirical prediction under the null hypothesis of no bias is what some economists have deemed the “infra-marginality” problem. The outcome test for bias in treatment is essentially a test of whether the contraband discovery rate differs across groups. If the hit rate for stops made of Black people is lower than that for stops made of White people, then the common interpretation is that Black people are being held to a less stringent or lower evidentiary standard by police officers when making decisions whether to stop and search an individual. However, it is easy to generate a simple hypothetical example where suspects differ in their observable signals of culpability, where there are differences across race in the proportions that display these signals, where officers hold one group to a differential evidentiary standard, yet where the hit rates are equal across groups (see Box 7-1).

Further, if the average propensity to offend is higher among one racial group and the distributions of the offending propensity vary across racial groups, one cannot predict a priori what hit rates would be in the presence of racial animus (see Box 7-2). For example, if Black people are more likely to offend than White people, animus-driven, racially biased treatment by the police may generate higher hit rates for Blacks, equal hit rates across

racial groups, or lower hit rates for searches of Black people.² With this in

___________________

² An early and seminal contribution to this literature by Knowles, Persico, and Todd (2001) addressed the infra-marginality problem by offering a theoretical model whereby the search rates and carrying propensities are endogenously determined by strategic interaction between police and potential suspects. A key aspect of their theoretical model is that they assume that officers can always increase the likelihood of detection to 100 percent by focusing enforcement efforts on a specific group, a condition that forces both officers and suspects to play “mixed strategies” whereby officers search suspects at a given rate and individuals carry contraband with a given probability. In this model, the likelihood of carrying is ultimately determined by the cost to officers of searching a suspect while the likelihood of being searched is determined by the relative benefits of carrying contraband for the group in question. Bias is introduced into the model via a lower cost to officers of searching Black suspects and, in turn, a lower equilibrium carrying rate for Black citizens. Dharmapala and Ross (2004) extended the model of Knowles, Persico, and Todd (2001) to allow for imperfect observability of citizens by the

mind, it is likely that any comparison of hit rates across groups provides a somewhat asymmetric test for animus against Blacks. Specifically, lower hit rates imply disparate treatment. However, equal or even higher hit rates for Blacks cannot rule out bias or animus (see Dharmapala and Ross, 2004, for an alternative development of this argument).

___________________

police. The innovation here is to admit the possibility that for most individuals, the police cannot increase the likelihood of detection to 100 percent given resource constraints. With this extension, there will be some individuals for whom the relative benefits of carrying are high and for whom the pure strategy of always carrying dominates. To the extent that the fraction of the population for whom this pure strategy is optimal differs across racial groups, the infra-marginality problem reemerges and the strong predictions from Knowles, Persico, and Todd (2001) regarding the consequences of bias for relative hit rates disappears.

HISTORICAL BACKGROUND ON RACIAL DISPARITIES, BIAS, AND ANIMUS IN POLICING

There is a large and persistent gap in the level of trust that non-White people have in law enforcement as compared to White people, a longstanding phenomenon that is a function of the history reviewed below. This difference is highlighted in a recent study from the Cato Institute (Ekins, 2016). That study found that 68 percent of White respondents viewed the police favorably, while 40 percent of Black respondents reported favorable views. Black respondents (73%) were more likely to say that the police are too quick to use force than were White respondents (35%), and Black respondents were more likely to say that police tactics are generally too harsh (56% versus 26% for White respondents). Similarly, 43 percent of Black respondents, but 62 percent of White respondents, say the police are courteous; 31 percent of Black respondents, but 64 percent of White respondents, believe that the police treat everyone equally; and about 4 in 10 Black respondents rate the police highly in terms of enforcing the law, protecting them from crime, and responding quickly to calls for help, as opposed to approximately 6 in 10 White respondents.

In recent years, gaps have also developed between White people and Hispanic people; in 2015, there was a 5 percentage-point difference in the fraction of White respondents (57%) and Hispanic respondents (52%) who placed “a great deal” or “quite a lot” of confidence in the police (Jones, 2015). The Cato Institute study discussed above finds even larger gaps between White people and Hispanic people in the views held about the police (Ekins, 2016). We note that the more frequently measured racial gap between White people and Black people in views about the police is generally unchanged over recent decades and that overall trust in the police as measured by national polls, such as the Gallup Poll, has remained more or less constant over the past 30 years, with between 50 and 60 percent of adult Americans expressing trust in the police (Balz and Clement, 2014; Jones, 2015).

A representative survey of police officers in the United States revealed similar differences in how White and Black police officers viewed treatment of non-White people (Weisburd and Greenspan, 2000). More than half of Black officers, compared with just 17 percent of White officers, agreed or strongly agreed that Whites received “better treatment” than Blacks. A recent evaluation of video-recorded traffic stops made in Oakland, California, during April 2014 provides empirical support for the Black officers’ self-reports (Voigt et al., 2017). That study found that officers were more likely to speak to Black drivers in informal familiar language (e.g., “man” versus “sir”), while also using more harsh legal terms (e.g., arrest versus

registration), and fewer explanatory terms (e.g., “the reason . . .” or “. . . because . . .”) (Voigt et al., 2017).

These differences extend to how officers feel about the need for police reform, particularly with respect to how the police interact with people in non-White communities. For example, a recent survey of police officers conducted by the Pew Research Center (Morin et al., 2017) found that 27 percent of White officers, but 69 percent of Black officers, said that the protests that followed fatal encounters between police and Black people were motivated to at least some extent by a genuine desire to hold police accountable, whereas 92 percent of White officers, but only 29 percent of Black officers, said that the country has made the changes needed to ensure equal rights for Black people.

The response of people from different communities to a particular police incident or policing strategy is a function not only of the contemporary actions of the law enforcement but also of the historical relationship between those communities. The historical record provides the framework for how those contemporary actions are viewed. In this section, we begin with a historical context for thinking about how changes in policing practices may be viewed in different communities, particularly non-White communities, in the United States.

This section largely focuses on the relationship between Blacks (and Black communities) and the police. Blacks have been the largest non-White group for most of American history, whereas the Hispanic population in the United States was relatively small for most of the 20th century. It was only in the mid-1980s that the Hispanic population began to grow at the pace typical of recent years. In 1970, Blacks represented 11 percent of the population of the United States, while Latino/Hispanics represented 4.6 percent. By 2013, Blacks comprised 12.3 percent of the population and Latino/Hispanics represented 17.1 percent (National Academies of Sciences, Engineering, and Medicine, 2016). Though there are some portions of the Hispanic community that have long histories with the police, such as Mexican Americans in the Southwest and Puerto Ricans in New York City, there is relatively less historical analysis of these policies. A notable exception is Garcia’s (1980) documentation of the extent of cooperation between federal immigration officials and local police departments during the 1950s; Perea and colleagues (2014) also include considerable material on the police and Latinos.

Before we review key moments in the history of race and policing in the United States, three important points of clarification are necessary. First, the purpose of this short summary is not to document the specific history of each of the roughly 18,000 law enforcement agencies in the United States. Rather, we will describe policies, both federal and more local, that potentially influenced all local agencies because of their national character or

that were commonplace in areas where most Black people lived (the “Black Belt” South and northern urban centers) (see North, 1981; Hagan, Shedd, and Payne, 2005; Walker, Spohn, and DeLone, 2007).

Second, this is not a history of racial animus or bias in the formation and application of proactive policing strategies in particular (about which relatively little is known). Rather, the discussion focuses on the role of racial bias in general, and animus in particular, in police practices in general and the resultant disparate impacts on members of specific racial and ethnic groups. Current proactive policing policies have developed and have been implemented in an era in which the police are widely credited with having made considerable improvement in their behavior when dealing with Black Americans and with predominantly Black communities (National Research Council, 2004). But the form, impact, and perception of these policies are certainly affected by this history.

Third, in this account the committee has sought to highlight how the structure of criminal justice policies, both in the past and today, can have the effect of both creating and perpetuating racial inequality in the absence of explicit racial animus or racially biased behavior. Thus, while some policies might plausibly be considered to be grounded in “race neutral” reasons, it is critical to understand that these policies can have important negative economic, social, and health impacts on non-White people in society. As discussed in Chapter 3, this makes systematic bias in policing policies particularly difficult to identify and potentially address in both a legal and social science sense.

Racial Animus in Federal, State, and Local Policies

Police are charged with enforcing laws and ordinances in the communities they serve. For the majority of people living in the United States, the police are the visible face of the government and certainly of the criminal justice system. This means that police officers historically, in the course of their jobs, were tasked with enforcing rules that in some instances, as in the case of vagrancy laws, explicitly disadvantaged non-White people (Goluboff, 2016; Douglas, 1960).

There are plenty of examples from American history of policies that were explicitly the product of racial animus and deliberately disadvantaged Black people. These policies often relied on local police for the force of law. For example, the Fugitive Slave Act of 1850 specifically instructed local judges to issue warrants for the arrest of people who had fled across state boundaries despite being “held to service.” The act required those warrants to be acted upon by local law enforcement and also introduced criminal penalties for interfering with the apprehension of slaves who had escaped to free territory.

Even after the passage of the Fourteenth Amendment, local governments continued to pass laws that specifically required police officers to arrest Black people or people who engaged in certain economic transactions with Black people. In cities throughout the country, Black people were prohibited from otherwise public spaces that were reserved for White people, including but not limited to swimming pools, lunch counters, restrooms, water fountains, and public schools. People who entered these spaces in violation of local ordinances were arrested by the police (Hinton, 2016a, 2016b; Goluboff, 2016; Branch, 1998).

Prior to 1948, communities could freely adopt racially restrictive covenants that criminalized the leasing of property to non-White people or to members of certain religious groups (Brooks and Rose, 2013). Many of these restrictive covenants were adopted in what became known as “sundown towns”: municipalities that had either formal or informal policies that regulated when Black people could move freely in the town. Specifically, Black people were permitted to travel to town, to work, and to shop, but they had to leave “before the sun went down” (Loewen, 2005). Punishment for disobeying these rules could be violent and was often delivered by police officers or sheriff’s deputies. Because these sundown-town policies often were informal, it is difficult to determine when the practice ended—or whether it completely has. In his archival and ethnographic research, Loewen (2005) found formal signs demarking sundown towns as late as the 1970s, with officials in some jurisdictions still admitting to informal enforcement as recently as 2001.

Racial Disparities in Federal, State, and Local Policies

In addition to the criminal justice system being used for explicitly racist purposes, U.S. history is replete with examples of changes to the criminal justice system that have, at minimum, disparate impacts on non-White communities. These examples are particularly relevant, as modern proactive policies are often implemented in a manner that can be justified legally in race-neutral terms (although important concerns about the extent to which proactive policies interact with enforcement of the Fourteenth Amendment are discussed in Chapter 3). For the purpose of providing historical context, we highlight some particularly salient examples of policy changes that, despite having as a touchstone the seemingly objective goal of crime reduction, are now commonly understood both to have had serious negative implications for non-White people and to have enhanced, rather than mitigated, racial inequality in the United States (Alexander, 2012; Garland, 2001b; Hinton, 2016a, 2016b; Pager, 2003; Holzer, Raphael, and Stoll, 2006; Raphael and Stoll, 2013). A direct implication of this history is that, in the absence of clear evidence to the contrary, the historical record does

not support the assertion that changes in policing policies should be “given the benefit of the doubt” in terms of the relative harm they cause to White and non-White people.

The Thirteenth Amendment to the U.S. Constitution prohibited slavery “except as punishment for crime.” In other words, forced labor was still allowed, as long as the laborer had been convicted of a crime. As documented by Ransom and Sutch (1977), Foner (1988), Lichtenstein (1993, 1996), Mancini (1996), Ingram (2014), and LeFlouria (2015), after the ratification of the Thirteenth Amendment, southern state and county governments quickly enacted a series of laws known collectively as the “Black Codes” (Ransom and Sutch, 1977). While technically race neutral, these laws in practice were applied almost exclusively to Black people and were actively used to control and limit the newly freed population. Georgia, for example, criminalized hunting on Sundays and letting cattle roam free, but only in counties with large Black populations (Foner, 1988). A key feature of all of these laws was a vagrancy statute requiring individuals to prove, on demand and in writing, that they were employed. Failure to do so would result in arrest and fine, incarceration, or both (Ransom and Sutch, 1977). In spite of the fact that late 19th century employment rates appeared to be higher among Black adults than White adults (Higgs, 1977), vagrancy laws were almost exclusively used to arrest non-White people and continued to be used to arrest “uppity” non-White people and “radicals” through the 1960s (Lichtenstein, 1993; Goluboff, 2016).

Once arrested and convicted (Black Codes also tended to prohibit non-White people from serving on juries), Black people were no longer protected by the Thirteenth Amendment. In fact, state penitentiaries and county jails would charge private citizens for the privilege of “leasing” the labor of incarcerated individuals. The practice was lucrative for states, with several states drawing more than 50 percent of their annual state income from such revenue at the so-called “nadir” of American race relations in 1901 (Lichtenstein, 1996). Although the Panic of 1907 made convict leasing less profitable, the Federal Aid Road Act of 1916 allowed states to use the value of convict labor as part of their “match” for federal highway funding, reinvigorating the process (Schoenfeld, 2014). This produced a perverse incentive for states to arrest individuals from vulnerable populations—overwhelmingly Black people—to serve as revenue engines for the state (Larsen, 2016; Lichtenstein, 1993). This practice, although formally abolished by Alabama (the last state) in 1928, persisted in less formal ways until it was finally outlawed in 1941, a mere 75 years ago (Blackmon, 2009).

Law Enforcement Resistance to the Civil Rights Movement

Specific actions of individual police officers, along with policies and practices ordered by mayors or governors, created a situation in which law enforcement was used to deny or limit the social, economic, and legal rights of Black people during the Civil Rights Movement. Central components of modern U.S. history are the images of state and local law enforcement officers using force against Black men, women, and children engaged in either acts of peaceful protest (as in the Selma to Montgomery March led by the Student Nonviolent Coordinating Committee on March 7, 1965) or exerting their federally protected rights (as in the state-controlled Arkansas National Guard prohibiting nine Black students from entering Little Rock’s Central High School during September 5–20, 1957). As suggested by the events that led up to and followed the trial leading to the Supreme Court’s decision in United States v. Price (383 U.S. 787 [1966])—better known as the Mississippi Burning case—local law enforcement sometimes interfered even more violently with civil rights efforts. In that incident, members of the Neshoba County Sheriff’s office and a Philadelphia, Mississippi, police officer were eventually convicted by federal prosecutors for conspiring with members of the Ku Klux Klan and others to murder three civil rights workers who had gone to Philadelphia to register Black voters in the summer of 1964.

Racial Disparities in Criminal Justice Contact Driven by Federal Policy

This active resistance by law enforcement to the Civil Rights Movement coincided with federal government policies that expanded the physical presence of officers in places where many Black people lived, resulting in greater entanglement of Black people in the criminal justice systems of states and cities. Two weeks after President Johnson signed the Civil Rights Act of 1964 into law, the killing of an unarmed Black 15-year-old boy by New York City police sparked the Harlem uprising and a wave of other disturbances that summer—prompting President Johnson to call for a “War on Crime.” One of the major components of the War on Crime was the enactment of the Law Enforcement Assistance Act of 1965 and the Omnibus Crime Control and Safe Streets Act of 1968. Both bills created new federal grant programs that directed federal funds to local law enforcement agencies through the Justice Assistance Grant (JAG) Program. This program increased the level of policing in areas that recorded more violent crimes, which in many areas has led to greater policing of poorer and/or more predominantly Black communities.

The “War on Drugs” that developed during the presidency of Richard Nixon provides another example of the criminal justice system being used

in a way that disproportionately impacted non-White communities. As the federal government expanded the War on Drugs through the 1980s, the disparity in implementation manifested itself in two notable ways. First, Black people were arrested for drug use in proportions, as a percentage of group population, that were much higher than survey data on drug use would predict. By the early 1990s, Black adults made up only 13 percent of drug users (according to survey data) but constituted 40 percent of those arrested for drug violations (Langan, 1995).

Second, federal penalties for drug violations put in place by the Anti-Drug Abuse Act of 1986 were substantially more severe for drugs that Black people were more likely to use. In 1993, the National Household Survey on Drug Abuse found that 12 percent of White people over the age of 12 reported any lifetime use of cocaine, compared with 9.5 percent of Black and Latino people. The same survey found that cocaine in its processed, “crack” form was more popular among Black and Hispanic people than among White people: 3.4 percent of Black and 2.0 percent of Hispanic people over the age of 12 admitted to ever using crack cocaine, compared to 1.6 percent of White people. With little debate, the Anti-Drug Abuse Act imposed a 5-year mandatory minimum sentence for possession of only 5 grams of crack cocaine, while it imposed the same mandatory minimum punishment for possession of 500 grams of powder cocaine. Although the social settings of typical transactions for crack and powder cocaine are different and the faster absorption of processed crack cocaine leads to different consumption experiences (Nestler, 2005; National Institute on Drug Abuse, 2016), the 100:1 weight ratio in thresholds for punishment for crack and powder put in place in 1986 is difficult to justify using objective measures of social harm (Bobo and Johnson, 2004; Tonry and Melewski, 2008). Despite repeated recommendations by the U.S. Sentencing Commission to revise the crack quantity thresholds upward, to lessen the disparity between the sentencing scheme for the two forms of the drug, and several legislative efforts to mitigate the difference, Congress did not act to reduce the sentencing disparity between crack and powder cocaine until the Fair Sentencing Act of 2010.

Finally, the Violent Crime Control and Safe Streets Act of 1994, which itself contained no explicit mention of race, was enacted in the context of fears of “super-predators”: understood to be young people with “absolutely no respect for human life and no sense of the future,” who were overwhelmingly from “[B]lack inner-city neighborhoods” (DiLulio, 1995). This law implemented mandatory life sentences for people convicted of serious federal crimes if they had previously been convicted of drug or violent offenses at the federal or state level. While race-neutral on their face, such repeat-offender sanctions and “three strikes” laws, which impose longer punishments on people who have had more frequent interactions with the

criminal justice system, have led to racial disparities in punishment. In part supported by federal funding programs designed to increase local policing, police departments serving cities with more than 250,000 people have more police per capita (around 22 per 1,000 people in 2014) than departments in other places (about 16 per 1,000 people in 2014) (U.S. Department of Justice, 2014). To the extent that the Black population of the United States is more likely to live in larger cities than the White population,³ Black people are more likely to come in contact with the police, all other things being equal, simply because they live in places where there are more police officers.

Pointing out the instances in which criminal justice policies, and policing in particular, have differentially impacted Black people does not invalidate the observation that Black and other non-White communities have benefited from some policies that have brought greater focus on crime within their communities. Policing, like clean water, good street lighting, and public transportation, is a public good designed in principle to help victims of crime and, more generally, support community members in achieving their goals and projects. During the same period that birthed these social policies and proactive policing, many Black communities expressed a desire for both harsher punishments of criminal behavior and more responsive policing (Fortner, 2015; Kennedy, 1997). In fact, a parallel (if less prominent) critique of police and race in the United States is that Black neighborhoods have historically suffered from under-policing (Kennedy, 1997; Forman, 2017).

There is a growing body of evidence that exposure to violence and crime is an important component of persistent poverty (Sharkey and Torrats-Espinosa, 2017; McCoy et al., 2015; Sharkey et al., 2014), and scholars have also found that segregated Black neighborhoods have historically suffered from under-policing (Kennedy, 1997; Meares, 1998). Taken together, these findings imply (although they do not prove) that reductions in crime may reduce some forms of racial inequality. It is certainly the case that the spectacular declines in U.S. crime rates since their peak in the early 1990s were disproportionately concentrated in the nation’s largest cities (and in particular, in the central cities of metropolitan areas), in which the population is disproportionately non-White (Kneebone and Raphael, 2011). These declines greatly reduced the disparity in crime rates relative to suburban cities. Within cities, there is also evidence suggesting that these declines in crime accrued disproportionately to non-White neighborhoods (Lofstrom and Raphael, 2016).

While this link has not been conclusively and explicitly proven, the

___________________

³ Based on the 2014 American Community Survey, 20 percent of Black people versus 7.6 percent of White people live in cities with more than 250,000 residents (Ruggles et al., 2015).

crime reductions associated with proactive policies may place downward pressure on one component of inequality in the United States. Over the past 40 years, police departments across the country have become more diverse (Sklansky, 2005), a phenomenon that has been credibly shown to reduce one potential measure of bias in policing—the relative arrest rates of Black and White adults—without affecting official crime rates (McCrary, 2007). Table 7-1 presents tabulations from various sources on the racial composition of police and other law enforcement entities. Panel A shows data on local sworn officers for the period 1987 through 2013 from the Law Enforcement and Administrative Statistics Survey (LEMAS). Panel B shows similar estimates from the Decennial Census for the period 1950 through 2015 on the racial composition of individuals who self-report their occupation as “Police Officers and Detectives.” Panel C presents similar tabulations from the Decennial Census for those who self-report as “Sheriffs, Bailiffs, Correctional Officers, and Jailers.”

While there are some discrepancies between these different measures, the underlying trends are quite clear. In the 1950s, law enforcement officers were largely White. By 2015, over a quarter of police officers and detectives were from non-White groups, as were roughly a third of sheriff’s deputies and correctional officers. This finding is consistent with data from LEMAS according to which law enforcement agencies overall have become more diverse since 1987 and departments serving larger jurisdictions have become even more diverse (Reaves, 2015; see also Equal Employment Opportunity Commission, 2016, p. 13). However, a study analyzing police personnel data for 269 police departments in jurisdictions with more than 100,000 residents also found that (1) there are still substantial gaps between the representation of non-Whites within law enforcement agencies and their demographic representation in communities, and (2) the representation of non-Whites in police departments has not kept pace with changing demographics (Governing, 2015; see also Equal Employment Opportunity Commission, 2016, p. 13).

In summary, while there have been important changes in the scope for racial bias and animus in policing, with respect to the impact of proactive policing on racial bias and disparate outcomes, law enforcement in the United States does not start with a clean slate. As noted by Chief Terrence M. Cunningham in his presidential address to the International Association of Chiefs of Police, “this dark side of our shared history has created a generational—almost inherited—mistrust between many non-White communities and the law enforcement agencies that serve them” (Jackman, 2016).

This perspective is echoed in the discussions that the committee had with representatives from Black Lives Matter (see Appendix A). The committee was urged repeatedly by community advocates to avoid a narrow

TABLE 7-1 Racial/Ethnic Composition of Law Enforcement in the United States

SOURCES: Panel A data from Reaves (2015) and Panels B and C data from Integrated Public Use Microdata Samples.

framing of proactive policing and to frame the ways in which communities experience policing. For instance, Brittany Packnett, a leader in the Black Lives Matter movement and a national figure in the wake of the Ferguson, Missouri, protests, advised the committee to include the full context that shapes community experiences, from history to broader social constructions of race (see Appendix A). She elaborated:

The committee views this perspective as important to keep in mind throughout the remaining sections of the chapter. Because the focus of our study task is on policing, we do not examine the broader social forces that lead to the outcomes that police address. As Ms. Packnett suggested to the committee, an alternative approach would be to employ social rather than policing interventions to address crime problems. Although such approaches have been suggested to address, for example, crime hot spots (see Weisburd, Telep, and Lawton, 2014), they have not been developed broadly as crime prevention approaches in recent years. The choice of policing as a response to crime problems is in itself a policy decision that has implications for communities. This issue is beyond the scope of the committee’s deliberations but one that we think is important for policy makers to consider as they explore crime prevention approaches.

The historical and contemporary social context of policing plays a role in how communities perceive police behavior, and one cannot assess the motivation for or effect of disparate treatment or racially biased behavior without recognizing that proactive policing is nested in the more general historical context of racial disparities and bias in policing. This context must include attention to the experiences of Black and other non-White Americans, with due consideration to the impacts on Black and other non-White communities.

___________________

⁴ We should note that Ms. Packnett may not have been using “bias” in the way in which the committee has defined the term for the purposes of this report.

POTENTIAL REASONS WHY MODERN PROACTIVE POLICING MAY BE ASSOCIATED WITH DISPARITIES AND BIAS

The committee identified four components of many proactive policing strategies that are plausibly related to an increase in racially disparate criminal justice outcomes. These features of proactive policing may lead to racial bias in department policy in general, as well as in individual officers’ decisions about stops and arrests. As discussed in later sections, only a limited body of research has empirically tested the causal connections between these potential mechanisms and changes in racial disparities and racial bias.

First, if non-White people are more likely to commit criminal offences, racial disparities in police-citizen interactions are likely to occur. Earlier reviews of the empirical literature did indeed document relatively higher offending rates among Black people in the United States (Sampson and Lauritsen, 1997; Tonry, 1995), rates that were likely influenced by a range of factors known to increase crime, including differences in income, education, social networks, discrimination, neighborhood characteristics, and many others. More recently, O’Flaherty (2015, Chapter 11) reviewed empirical trends from homicide statistics and victimization surveys, which revealed a higher offending rate among Black people for homicide and robbery. Hence, a proactive effort to combat robbery may generate a racial disparity in arrest rates to the extent that members of one group commit this offense at a higher rate than the comparison group.

Disparities in the historical relationship between law enforcement and residents of difference races and ethnicities can manifest not only as differential crime rates across demographic groups but also as different behaviors on the part of citizens interacting with police. Extensive research demonstrates that, compared to White people, Black people are more distrustful and nervous (even scared) when interacting with a police officer (e.g., see Najdowski, Bottoms, and Goff, 2015). To the officer, this nervousness may appear suspicious (Najdowski, 2011), and the officer may stop and question the Black person based, in part, on this “suspicious” behavior. As a second example, U.S. neighborhoods are generally still segregated by race, and areas with more Black residents tend to have lower income and more crime. Accordingly, officers are statistically more likely to encounter Black citizens in high-crime neighborhoods. An officer may decide to question a Black suspect partly because the encounter occurs in a high-crime area (Terrill and Reisig, 2003). In both of these cases, a dimension that rightfully influences an officer’s decision (the citizen’s nervous behavior or the perception of greater danger at the location) is related to the suspect’s race.

Second, police departments may prioritize enforcing ordinances and laws that non-Whites are more likely to violate. For example, if Black and Latino drug users are more likely to purchase narcotics on street corners or

in open air markets, whereas White drug users are more likely to conduct transactions in private spaces, local drug enforcement that prioritizes “buy busts” will generate racial disparities in arrests. Disparities may also result from explicit or implicit priorities communicated by police leadership to officers on patrol. Certainly a proactive departmentwide policy of targeting, say, young Black men would be unconstitutional. Nonetheless, in a nation with nearly 18,000 independent law enforcement agencies, it is possible that such policies articulated from the top govern law enforcement practice in some jurisdictions.

Third, policing efforts may be geographically concentrated. This may result naturally from the geographic concentration of calls for services and calls to report a crime that initiate investigations. Alternatively, disparities may result from proactive strategies that target specific neighborhoods. To the extent that high-crime areas that disproportionately generate calls for service are more likely to become the focus of proactive strategies and more likely to be located in non-White neighborhoods, racial disparities in the incidence of police–citizen interactions will result. More focused geographic programs, such as hot spots policing, may reduce overall police intrusion in larger neighborhoods by focusing on the small number of streets that have high-crime rates (D. Weisburd, 2016). Nonetheless, non-White people may in turn be overrepresented on the targeted streets.

Finally, proactive policies that target high-activity offenders or those with more extensive criminal histories are likely to involve non-White suspects disproportionately. As previously discussed, non-White, and particularly Black, Americans are more likely to live in areas with more police per capita. This will likely result in more contact with the police, and thus an increased probability of being identified as a high-activity offender for Black people relative to otherwise identical White people.

EVIDENCE FROM PSYCHOLOGICAL SCIENCE ON RACIAL BIAS IN POLICING

This section attempts to evaluate the contribution contemporary psychological science can make to understanding the role of racial bias in proactive policing. A central focus of the relevant research is disaggregating the causes of biased behavior into two factors: “dispositional factors,” which are individual differences that tend to predict an individual’s relative tendency to engage in biased behaviors, and “situational factors” that tend to provoke such behaviors. Further, when psychologists discuss dispositional factors in the literature, they tend to refer to individual differences in “stereotyping” and “prejudice” (Stangor and Crandall, 2013). While not identical, these terms are similar to “statistical prediction” and “animus” as defined by the committee, and in the interest of simplicity we will

use the committee’s terms in reviewing the literature. A broad term some social psychologists use for situations that can tend to facilitate disparate treatment—regardless of the presence or absence of dispositional factors associated with bias—is “identity traps” (Goff, 2013, 2016; Goff and Godsil, 2017).

The Psychological Science of Bias

Studying racial animus and statistical prediction has been a central concern of American psychology since nearly the start of the 20th century. During that era in American history, negative racial attitudes were more openly expressed than is typical today. Overt, non-anonymous expressions of racial animus have declined markedly in American society because social norms have evolved to prohibit them (Dovidio, 2001; Dovidio and Gaertner, 2004). For example, self-reports of so-called “old-fashioned racism” in which people took pride in their anti-Black feelings, have decreased over time, and Whites’ stated support for integration and interracial marriage has increased. Similarly, evaluations of the police have suggested that there have been important gains in professionalism, leading to less racial animus among law enforcement officers (National Research Council, 2004).

Although overt expressions of biased behavior have declined in society and among police, racial animus has not disappeared. Rather, it has evolved.⁵ Researchers have developed a variety of tasks that measure animus and statistical prediction indirectly or implicitly. While this research has had little direct translation to the field of policing, it does establish basic findings about factors and situations where people are more likely to express, and act upon, negative racial attitudes. These findings have potential implications for how the actions of individual police officers, and policing policies more generally, may be shaped by attitudes such as racial animus or prediction.

For example, in one influential study, participants sat at a computer and completed a series of trials on which a face appears briefly and is quickly followed by a word (Fazio et al., 1995). Regardless of the face, participants are asked to classify each word as either “positive” or “negative.” But participants are typically influenced by the face. When a Black face precedes the word, participants are often faster to identify a negative word and slower to identify a positive word; when a White face precedes

___________________

⁵ It is important to note that social psychology has largely focused on studying Black/White racial biases. Although several researchers have acknowledged this limitation, such acknowledgment does little to counteract the limits on the field’s ability to generalize across interracial groups. Consequently, much of the literature reviewed here will necessarily be limited to a discussion of Black/White biases.

the word, they are often faster to classify a positive word and slower to classify a negative word. The evidence from this task (and many others like it) suggests that participants associate Black people (more than White people) with negativity. Similar work suggests that Blacks (more than Whites) are associated with weapons and with the general concept of danger (Payne, 2001). These implicit measures are potentially valuable tools in exploring racial attitudes because they have the potential to uncover attitudes that participants are unwilling to report when asked directly. These measures may even show associations of which the participants, themselves, are unaware. Critically, these forms of animus may influence real-world behavior, including interracial interactions (Dovidio, Kawakami, and Gaertner, 2002; McConnell and Leibold, 2001).⁶

There is, to our knowledge, no peer-reviewed work in psychology examining how any motivating factors, implicit or explicit, held by police influence their behavior toward subjects in the real world. There are, however, several studies examining the effects of different causes of racially biased behavior on interpersonal interactions more generally. For example, McConnell and Leibold (2001) asked White participants to complete an implicit measure of negative attitudes toward Black people (akin to racial animus). The participants also had brief interactions with one White experimenter and one Black experimenter. These interactions were videotaped and later coded. The results suggest that participants with greater animus had more awkward and uncomfortable interactions with the Black experimenter. They demonstrated more speech hesitation, more speech errors, and were less likely to smile when interacting with the Black experimenter. Moreover, the Black experimenter felt worse about interactions with these same participants.

Dovidio, Kawakami, and Gaertner (2002) measured attitudes toward Black people using both an explicit measure (a questionnaire) and an implicit measure. They found that explicitly measured attitudes were related to participant’s deliberate behavior: participants who expressed less positive attitudes on an explicit measure tended to say fewer nice things to a Black interaction partner. But implicit measures were related to nonverbal behavior, leading to more blinking and less direct eye contact. Finally, Richeson and Shelton (2003) showed that, when interacting with a Black person, participants with higher levels of implicitly measured negative at-

___________________

⁶ Implicit measures of animus include skin conductance, essentially a physical measure of sweating (Rankin and Campbell, 1955), other physiological measures (e.g., measuring facial muscle movements with EMG), brain imaging (EEG, fMRI) (e.g., Phelps et al., 2000; Hart et al., 2000), indirect self-reports (e.g., completing a word with blank letter-spaces with negative instead of positive words), subliminal priming, sequential priming (e.g., Fazio, 1995), implicit association tests (Greenwald, McGhee, and Schwartz, 1998), and the go/no-go association task (Nosek and Banaji, 2001).

titudes suffered cognitively—as if they were working harder to control their behavior. Though these studies do not involve police or law enforcement scenarios, they clearly have relevance for officers interacting with members of different racial groups, and they raise a host of questions: Are officers inadvertently communicating a sense of discomfort when interacting with Black or Latino citizens? Does an initial sense of discomfort alter the course of the interaction, potentially evoking greater hostility from the citizen? These questions have direct relevance for how proactive policing may influence racial disparities.

It is worth noting that one implicit measure of social cognition, the Implicit Association Test (IAT), has sparked vigorous debate. Questions have been raised about whether certain aspects of the measurement procedure (e.g., the cognitive demands of the task or the differential salience of categories, like Black and White), can create an illusion of biased behavior, racial animus, or the unconscious process psychologists refer to as “implicit bias” on the IAT, even when the participant does not harbor negative views toward Black people. Questions have also been raised about the reliability and validity of the measure. The task typically demonstrates good internal reliability (if a person completes the task and researchers compute a score based on the odd-number trials, it correlates well with a score based on the even-number trials), but the task does not show acceptable test-retest reliability: that is, if a person takes it today, he or she may get one measure of their racial attitudes; if that person does the task again tomorrow, he or she may get a very different measure. While some have argued that this nullifies the utility of the IAT-based measure, others hypothesize that the nature of racial attitudes is simply more labile than previously imagined (Cunningham, Preacher, and Banaji, 2001). Finally, in terms of construct validity, there is disagreement about what the IAT actually measures. Arguments persist regarding how well it predicts behavior (Oswald et al., 2013, 2015; Nosek, Greenwald, and Banaji, 2007), why there are relatively low correlations between different implicit measures of animus (Ito et al., 2015), and whether or not the attitudes being studied are conscious (Fiedler, Messner, and Bluemke, 2006; Hahn et al., 2014). Each side of these debates is supported by experimental laboratory research. Still, there is relative consensus that implicit measures of attitudes—including but not limited to the IAT—predict racially biased behaviors under some range of conditions (Greenwald et al., 2009; Oswald et al., 2013). Under which conditions, how well, why, and for which people, is less clear, given the current state of the research.

Evidence from Studies of Racial Bias in Law Enforcement

There are a number of studies that directly examine the question of racial bias in law enforcement, using evidence collected in both the real world and in the laboratory. This review is organized topically, focusing first on domains that seem most relevant to the question of proactive policing, then discussing a few domains that are broadly related to the justice system even if they do not bear directly on proactive policing.

Perceptions of Suspicion

Eberhardt and colleagues (2004) conducted a series of studies with both undergraduates and police officers in which they subliminally presented images related to crime (this preliminary exposure is called “priming”). They then presented a pair of faces: one White and one Black. When thinking about crime, where would people look? The study showed that students and officers both paid greater attention to Black faces. In a follow-up study, the researchers again tested police officers. Some were primed with the idea of crime, while others were not. Then the officers were presented with a picture of a suspect, who was either Black or White. After viewing the suspect, officers were given a surprise memory task that involved identifying the suspect from a lineup of other suspects of the same race. In both the Black-suspect lineup and the White-suspect lineup, some of the suspects had more stereotypically Black or Afrocentric features, whereas others had more sterootypically White or Eurocentric features. In other words, even though all the faces were persons of the same race (e.g., Black), some of the individuals had facial features that were more typical of Black people in general (i.e., similar to the stereotype of a Black face, e.g., darker skin tone), whereas others had facial features that were less typical of Black people and more typical of White people (i.e., similar to the stereotype of a White face, e.g., lighter skin tone). Crime-primed officers who viewed a Black suspect systematically misremembered the suspect: in the lineup, they typically identified a suspect with more stereotypical Black features. By contrast, the crime-primed officers who saw a White suspect showed a weak tendency to identify a less stereotypical White suspect. In other words, when the officers were thinking about issues of crime and criminality, they tended to ascribe the crime to suspects with more stereotypical Black or Afrocentric (less Eurocentric) faces.

Racial Bias in the Use of Force

One real-world investigation of the role of race in an officer’s decision to use force comes from Terrill, Mastrofski, and their colleagues (Terrill and

Mastrofski, 2002; Terrill and Reisig, 2003). Researchers analyzed reports from observers who rode with police officers in two cities for almost 6,000 hours, coding interactions with roughly 12,000 citizens. Among other variables, the observers recorded the gender, race, and age of the citizen; the citizen’s level of disrespect and resistance toward the officer; whether the citizen had a weapon, and critically, the officer’s use of force along a continuum from none to verbal commands to physical restraint to physical strikes with the body or “external mechanisms.”

In their analysis, the researchers sought to statistically control for variables that a reasonable police officer should be expected to use when making decisions about how much force to employ. For example, officers might be expected to use more force with someone who resisted arrest or someone who was engaged in conflict when the officer arrived. To the extent that their statistical models, which included these variables, captured the actual decision rules of the officers, the researchers could also examine whether a citizen’s race (or gender or apparent wealth) could explain additional variation in use of force. The data suggest that, under the statistical modeling of the role played by an extensive set of factors, police used greater force with non-White suspects (Terrill and Mastrofski, 2002).

In a follow-up analysis, Terrill and Reisig (2003) suggested that much of the apparent effect of a suspect’s race on police use of force can be explained by the neighborhood in which the encounter took place. This result suggests that discrepancies in behavior (racial bias) may be driven by statistical prediction on the basis of the environment. The authors suggested that the tendency to use greater force with non-White suspects may be driven by the fact that, in certain parts of town, officers expected greater levels of danger or reduced accountability. This suggestion is consistent with laboratory-based evidence on when people are most likely to be behave in a way that is consistent with racial animus.

Additionally, Kahn and colleagues (2016) examined individual variation in suspects’ faces, particularly whether racial stereotypicality is related to police use of force. In a random sample of booking photos, Kahn found that, controlling for type of arrest, reported level of resistance, and the presence of drugs and alcohol in a suspect’s system, ratings of White suspects’ phenotypic stereotypicality were negatively associated with likelihood and severity of force. That is, the more stereotypically “White” a suspect looked, the less force was used (statistically controlling for all the other variables). Interestingly there was no relationship between the stereotypicality of Black suspects’ faces and use of force in the same sample.

Racial Bias in the Decision to Shoot

Concern about the impact of suspect race on officer-involved shootings has sparked a flurry of work, using both real-world data and laboratory simulations. In what is probably the most sophisticated of the real-world analyses, similar conclusions have been drawn by the vast majority (but not all) of the work on this domain.

In laboratory experiments, several researchers have used video game simulations to explore racial bias in the decision to shoot. These simulations present a target (usually male) who is either White or Black. On some trials the target is armed, and on some trials the target is unarmed. The participant is instructed to shoot the armed targets but not to shoot the unarmed targets; not shooting in these experiments usually requires a response distinct from inaction—for example, pressing a button labeled “don’t shoot.”

In studies of undergraduates, these studies reveal consistent evidence of racial bias in both response time and accuracy (Correll et al., 2002, 2007; Greenwald et al., 2003; Plant, Peruche, and Butz, 2005; see Payne, 2001, for similar results from a somewhat different task). When responding to an unarmed target, the student participants are typically slower to indicate “don’t shoot” and more likely to incorrectly shoot if the target is Black (rather than White); when responding to an armed target, participants are typically slower to shoot and more likely to incorrectly choose “don’t shoot” if the target is White (rather than Black).

Results with police officers in similar experiments are somewhat mixed. Plant and Peruche (2005) found that officers showed bias in their behavior (at least initially). By contrast, Correll and colleagues (2007) found that police outperformed lay people—though they still showed bias in the speed of their responses (e.g., shooting armed Black targets more quickly), they showed no appreciable evidence of bias in their actual decisions (e.g., they were no more likely to shoot an unarmed Black target than to shoot an unarmed White target). After a number of additional studies with police, these researchers suggested that, when faced with a potentially hostile suspect, police probably activate some potential source of bias, but that police (unlike undergraduates) are able to overcome those potential sources and use diagnostic information (is the target armed?) to formulate their ultimate response (e.g., see Sim, Correll, and Sadler, 2013).

More recently, James and colleagues sought to develop a more realistic and immersive simulation (James, James, and Vila, 2016; James, Klinger, and Vila, 2014; James, Vila, and Daratha, 2013). These researchers used a deadly-force simulator, similar to those used by law enforcement agencies for training. The results were counterintuitive. Participants (both officers and lay people) were slower to shoot armed Black suspects than armed

White suspects, and they were less likely to shoot unarmed Black suspects than unarmed White suspects. While no explanation of the results is evident from the data, the researchers emphasized in explaining the data that all simulations have liabilities. Participants were typically aware that the research involved race, and in the work by James and colleagues, there was reason to suspect that officers and lay people responded strategically, intentionally attempting to act without racial bias (Wegener and Petty, 1997). This concern is compounded because, in these studies, participants had several seconds to respond. Given sufficient time, the desire to respond in an egalitarian fashion can override factors like racial animus or statistical prediction when individuals are aware that race may influence behavior (Fazio et al., 1995; Gaertner and Dovidio, 1986; Neely, 1977; Payne, Lambert, and Jacoby, 2002).

Several laboratory studies have also examined the effect of variability in racial stereotypicality on decisions to shoot—for example, does it matter if Black targets have lighter versus darker skin? Targets with greater racial stereotypicality seem to induce greater bias in the decision to shoot in computer simulations, and this pattern holds for both undergraduates and police (Kahn and Davies, 2011; Ma and Correll, 2011).

Risk and Protective Factors for Bias in Proactive Policing

In addition to innovations regarding the measurement and effect of dispositional factors that reflect internal characteristics of the individual, there is a sizable psychological literature identifying situations that tend to provoke racially biased behaviors, even absent negative racial attitudes, as well as situations that reduce the likelihood that racial animus will lead to biased behavior or (racially) disparate outcomes. Associations between race and concepts such as danger, crime, and negativity are notoriously difficult to “undo” in a permanent sense (Lai et al., 2014; Plant, Devine, and Peruche, 2010). In turn, statistical prediction that takes into account ethnic or racial characteristics may be accurate at times (e.g., in identifying the likelihood of individuals being involved in terrorist activities), although given the country’s normative commitment to equality, the legal standard for allowable statistical prediction is quite high. Recognizing the challenges of permanently changing attitudes, researchers have increasingly focused on identifying features that affect expression of attitudes as biased behavior; that is, they have focused on factors that increase or decrease the likelihood that any existing animus will result in biased behavior. It is particularly important to consider these factors in light of the nature of proactive policing, which may increase the frequency and extent of an officer’s contact with citizens and which may also rely heavily on officer discretion. To the extent that some proactive policies may increase the scope

for officer discretion, they may increase the potential for cues such as race to affect behavior. We will also consider protective factors such as training and policy interventions.

Risk Factors for Biased Behavior

In a series of studies by Fein and Spencer (1997), high-performing students were brought into a laboratory and given a challenging academic test. The students then received either false-positive or false-negative feedback on their test performance. Finally, students were presented with a woman whom they were led to believe was either Jewish (a stigmatized group in this context) or Italian (a nonstigmatized group in this context) and asked to rate her in terms of her overall personality and qualifications for a job. Students who received false-positive feedback rated the Italian and Jewish targets the same. However, students who received false-negative feedback rated the Jewish target much worse than the Italian one.

Across the literature, what social psychologists define as “threats to self-concept” tend to produce biased responses when: (1) the threat is in a domain that is important (Branscombe et al., 1999; Fein and Spencer, 1997); (2) a stigmatized individual is an appropriate target for negative behavior (Frantz et al., 2004; Goff, Steele, and Davies, 2008; Richeson and Sommers, 2016); and (3) negative behavior does not violate anti-racist norms (Dovidio, 2001; Dovidio, Gaertner, and Abad-Merino, 2017). For instance, in experimental laboratory studies, researchers have demonstrated that the mere presence of a stigmatized group member often causes White participants to experience concerns with appearing racist (Goff, Steele, and Davies, 2008; Richeson and Shelton, 2007, 2013; Shelton and Richeson, 2005, 2006, 2015; Shelton, Richeson, and Vorauer, 2006; Trawalter, Richeson, and Shelton, 2009; Vorauer et al., 2000; Vorauer and Kumhyr, 2001; Vorauer, Main, and O’Connell, 1998). Subsequently, this concern predicts social distancing behaviors (Goff, Steele, and Davies, 2008), negative evaluations of interracial interactions (Vorauer, Main, and O’Connell, 1998), and even negative evaluations of stigmatized group members (Frantz et al., 2004).

More broadly, psychology has identified robust sets of dispositional (individual characteristics) and situational (related to the environment) factors that are associated with higher levels of racially biased behavior. A dispositional risk factor for bias is a relatively enduring personal trait that puts an individual at risk of biased behavior. For example, social dominance orientation⁷ measures individuals’ support for hierarchies that disadvantage

___________________

⁷ While many forms of explicit animus have declined over time, social dominance orientation continues to exert an influence on behavior without the expression of this attitude having been diminished by prevailing social norms.

members of lower social strata; individuals who self-report higher levels of social dominance orientation are more likely to engage in biased behavior (Sidanius and Pratto, 1999). One study revealed that social-dominance orientation tends to be relatively higher in police officers compared to members of the general public, college students, and also public defenders, even after controlling for differences in gender, income, social class, age, education, and ethnicity (Sidanius et al., 1994).

Another dispositional risk factor identified in this literature is aversive racism. In this case, egalitarian values are explicitly stated, but unacknowledged negative racial attitudes and/or negative affective responses to members of stigmatized groups are found (Dovidio and Gaertner, 2004; Gaertner and Dovidio, 1986). To the extent that proactive policing places officers and residents in more frequent interactions that feature asymmetric power, it may be a risk factor for contemporary forms of behavioral bias, such as social dominance orientation or aversive racism (Gaertner and Dovidio, 1986; Sidanius and Pratto, 2001). However, this hypothesis has not been directly tested.

A third dispositional risk factor identified by social psychologists is executive function, which is a kind of flexible cognitive capacity that allows people to pay attention, follow rules, and use different intentional strategies. Unlike social dominance orientation and aversive racism, executive function is a protective factor for biased behavior. A variety of studies clearly demonstrate that participants with greater executive function may harbor animus, but the animus is less likely to influence their behavior (Amodio et al., 2008; Ito et al., 2015; Payne, 2005).

A situational risk factor for biased behavior is an aspect of a person’s physical or social surroundings that serves as a cue, making biased behavior more likely (Goff, 2016). One situational risk factor that may be relevant to policing is task complexity. When faced with a complex situation requiring difficult decisions, people often rely on superficial cues, hunches, and intuitions, which increases the likelihood of biased behaviors (e.g., Macrae et al., 1994; Miller, Bland, and Quinton, 2000; Robinson, Schmeichel, and Inzlicht, 2010). This influence is plainly reflected in much of the work reviewed above (e.g., Bodenhausen and Lichtenstein, 1987; Richeson and Shelton, 2005; Sommers and Ellsworth, 2000). The liability associated with task complexity is often exacerbated by time pressure, fatigue, and emotions such as fear or anger. These factors are endemic to policing and can undermine the ability to engage in complex or controlled thought. For example, a program of research by Payne and colleagues found that participants are more likely to mistakenly identify nongun objects as guns when the objects are paired with images of Black (rather than White) faces, but only when given a limited amount of time to respond (Payne, 2001; Payne, Lambert, and Jacoby, 2002).

A second (probably related) situational risk factor relevant in policing may be environmental threat, or the sense of danger created by the location in which an encounter occurs. In both studies of real-world police behavior and in laboratory simulations, poorer and more dangerous neighborhoods were associated with greater use of force (Terrill and Reisig, 2003; Correll et al., 2011).

A third situational factor relevant to policing may be stereotype threat, the concern with confirming or being negatively evaluated in terms of stereotypes about one’s own group (Steele, 1992). While this phenomenon is usually studied with academically stigmatized groups in educational settings, there is a growing body of research suggesting that those concerned with being seen as criminal will respond with behaviors that objectively attract police suspicion (Najdowski, 2011; Najdowski, Bottoms, and Goff, 2015). Conversely, those concerned with being stereotyped as motivated by racial animus can respond negatively to individuals who evoke that concern (Frantz et al., 2004; Goff, Steele, and Davies, 2008)—a finding of potential concern for police officers in the current climate.

Fourth and finally, concerns with appearing racist, foolish, or otherwise being negatively evaluated can reduce executive functioning and increase intergroup negativity (Richeson and Shelton, 2007, 2013; Trawalter, Richeson, and Shelton, 2009).

In summary, while not yet specifically tested among police, social psychological research on risk factors for racially biased behavior suggest that police officers who are (1) working in an environment where they are interacting with non-Whites, (2) have plausible race-neutral reasons for the biased behavior, and (3) feel criticized or generally unappreciated, may be more likely to exhibit biased behavior against non-Whites. Further, there is evidence that people who hold relatively stronger racial animus toward non-Whites are more likely to behave differently toward non-Whites in both overt and subtle ways. In the policing context, this may inhibit deescalation of an event and influence the perceived legitimacy of a police-citizen encounter, hypotheses that should be tested directly in field studies. Importantly, although dispositional differences robustly predict behavioral differences, they are often strongly moderated by situational factors. While the available research does not allow us to draw conclusions regarding proactive policing, future studies should examine these risk factors more directly.

Protective (bias-reducing) Factors for Biased Behavior

Research suggests that expertise and training may reduce the likelihood of bias occurring in particular situations. In general, individuals with more practice tend to be more likely to complete tasks accurately (e.g., MacLeod,

1998) and are less likely to engage in biased behavior (e.g., Kawakami, Dovidio, and van Kamp, 2005). Not surprisingly, practice on weapon identification can reduce racially biased behavior. For example, Plant and colleagues (Plant and Peruche, 2005; Plant, Peruche, and Butz, 2005) gave participants (including both student and police) a choice to shoot or not shoot a target person who was viewed on-screen accompanied by either a gun or another object. Over the course of the 160-trial task, the participants made racially biased shooting errors at first, but by the second half of the task, bias had decreased dramatically. Similarly, Correll and colleagues (2007), as noted earlier, compared police officers to community members in a videogame simulation. They observed that community members showed bias in their decisions to shoot but the police officers did not. A subsequent study of undergraduates in the laboratory demonstrated one possible reason why police (who have more expertise) tend to exhibit less bias and greater accuracy: undergraduates who practiced the simulation showed similar improvements in performance (Correll et al., 2007).

A second class of protective factors might be agency policies that directly address specific risks or that provide clear guidelines for behavior. For example, research has found that study participants demonstrate greater racial bias when they are forced to respond quickly and when they feel threatened. Policies that help address these vulnerabilities (e.g., more strongly encouraging officers to seek cover or increasing two-officer patrols) may help mitigate against these risk factors. In addition, clear guidelines about when and how to interact with members of the community may have the potential to reduce biased behavior by reducing discretion. Guidelines that clearly specify, for example, when a speeding car should be stopped might reduce the likelihood that officers will use race to inform their decisions.

Many of the studies to date have been conducted in the laboratory rather than assessing police officers acting in field settings, a limitation that make it difficult to draw conclusions regarding the ways and extent to which racial animus, whether acknowledged or not, may shape police behavior in practice. However, some risk factors for biased behavior may also be common to features of proactive policing. This suggests the importance of more direct research on the dispositional and situational factors that may influence whether police officers make biased decisions. It is also important to identify training programs that improve officer performance. There are no experimental studies of the effectiveness of “implicit bias training” in policing, for example, and the committee encourages further research in this area.

EVIDENCE FROM CRIMINOLOGY, ECONOMICS, AND SOCIOLOGY ON RACIAL BIAS IN POLICING

Like psychology, empirical research in criminology, economics, and sociology recognizes that the existence of disparities is not in and of itself evidence of racially biased behavior among police officers, nor is it evidence for the source (e.g., animus or statistical prediction or something else) for that bias. This section describes common strategies that have been used in these fields to estimate how police officers are using race in their decision-making process, focusing on a specific type of racial bias commonly called “racial profiling.” Racial profiling usually refers to police decisions to engage in vehicle or pedestrian stops, searches, or arrests or to take other law enforcement actions based at least in part on a targeted individual’s race, outside of the context in which officers target an individual because he satisfies a specific description of a criminal suspect.

Most of the empirical research on racial profiling does not address proactive policing specifically, though there are a few notable exceptions. Nevertheless, the research does consider profiling in the context of police stops, searches, and arrests, which is relevant insofar as proactive strategies, especially deterrence-based strategies, often promote increasing stops, searches, or arrests as a means to prevent crime. Nearly all the research on racial profiling, including evaluations of racial profiling in proactive policing strategies, evaluates the role of bias in police behavior using either a benchmarking approach or an outcome test. In the discussion and analysis that follows, the committee selectively reviews these two bodies of research. We highlight principal findings from some of the key studies in this area and also highlight the methodological shortcomings inherent to each methodological approach to identifying racial disparities in police–citizen interactions.

Comparisons of Racial Composition of Police–Citizen Interactions to Alternative Population Benchmarks

There are a number of studies that employ census data, driver’s license data, or both to compare the racial composition of those stopped by the police to the racial composition of the local resident population, to the population of areas near roadways, or to the current population of drivers. A 2003 study of Minnesota compared the racial composition of police stops and searches to the racial composition of the driving-age population in 65 municipalities throughout the state and demonstrated that in nearly all localities, Black people were overrepresented and White people underrepresented among those stopped and searched by the police (Institute on

Race and Poverty, 2003).⁸ A 2005 study of Illinois by the Northwestern University Center for Public Safety compared the racial composition of traffic stops in various localities throughout Illinois, as well as traffic stops statewide, to estimates of the racial composition of drivers in each locality and statewide (Weiss and Grumet-Morris, 2005). The authors found that across all localities, non-White drivers were only slightly more likely to be stopped than White drivers and that, in some localities (Chicago in particular), non-White drivers were less likely to be stopped by police. As one final example, a 2000 study by the Texas Department of Public Safety compared the racial composition of those stopped, cited, warned, and searched by the police to the racial composition of state residents. The analysis found that Black and Hispanic drivers were underrepresented among stopped drivers, among those cited, and among those receiving written warnings, but overrepresented among vehicle searches (Texas Department of Public Safety, 2000).

A series of studies attempts to refine the benchmarks to focus more specifically on the distribution of driver’s licenses and on heterogeneity in miles driven, as well as the difference in the geographic distribution of driving behavior by race. Smith and colleagues (2004) analyzed stop data for the state of North Carolina and compared cross-city variation in the proportion of stops of Black drivers to three benchmarks: (1) the proportion of licensed drivers in the locality who are Black, (2) estimates from direct observations of those driving on streets and freeways of the proportion of these actual drivers who are Black, and (3) the proportion of auto accidents involving Black people. The findings showed higher stop rates relative to the benchmarks for Black drivers, with the disproportionality in stop rates increasing with the relative size of the Black driving population in each locality. Alpert, Dunham, and Smith (2007), in an analysis of Dade County, Florida, police, compared the racial composition of stops to the racial composition of observed traffic violators at key intersections and to the racial composition of the not-at-fault driver in two-vehicle crashes throughout the county. Whereas they found that Black drivers were only slightly overrepresented among those stopped, there were clearer and larger difference in post-stop treatment. Lamberth (1994) used traffic surveys conducted by the researcher to estimate the racial composition of drivers as well as the proportion of drivers that are exceeding the speed limit along the New Jersey Turnpike by at least 5 miles per hour. These estimates were then compared to the racial composition of those stopped by state troopers and to the racial composition of those arrested, using data culled from patrol and radio logs. The analysis found that Blacks were disproportionately rep-

___________________

⁸ This study also compared racial disparities in whether searches uncovered contraband, a methodological approach we review in the next section.

resented among those stopped and arrested along the New Jersey Turnpike, relative to either the estimates of the racial composition of drivers or the observed racial composition of speeders.

Grogger and Ridgeway (2006) proposed an alternative benchmarking strategy that exploited differences in the level of ambient lighting and the consequent effect on officer ability to identify drivers’ ethnicity/race. Using microdata on officer-initiated stops in Oakland, CA, they basically assessed whether Black drivers constituted a higher proportion of stops made during the day relative to those made during the night. The reasoning behind this comparison was that those stopped at night, “under the veil of darkness” and for whom officer could not assess race, provided an unbiased estimate of the racial composition of those violating traffic laws or being stopped for reasons independent of race. They found little evidence of a disparity and concluded that their test yielded little evidence of disparate treatment by the police.⁹

Stop, question, and frisk (SQF) was used in New York City as a proactive policing strategy in specific neighborhoods or to target crime hot spots (Gelman, Fagan, and Kiss, 2007; Weisburd, Telep, and Lawton, 2014). In a benchmarking study of this intervention, Ridgeway (2007) compared the composition of those stopped, questioned, and frisked by New York City police to the racial composition of crime suspect descriptions. This study also compared the behavior of individual officers to internal benchmarks constructed to account for differences in shift and patrol location. Ridgeway found that despite comprising the overwhelming share of stops (more than 80%), Black residents were underrepresented among stops relative to citizens’ crime-suspect descriptions. The study also identified a small set of officers who stopped non-White people at very high rates relative to the constructed internal benchmark; it also identified cross-borough heterogeneity in the degree to which non-White people were overrepresented among those stopped by the police.

Gelman, Fagan, and Kiss (2007) provided an additional benchmarking study of SQF in New York City that used prior criminal activity as a benchmark. The authors employed stop-and-frisk data for New York City for the late 1990s to estimate count models at the precinct level. These models were then used to assess the degree to which non-Whites were stopped and frisked at rates disproportionate to prior-year arrests. They also assessed whether there was heterogeneity that varied systematically with precinct-level characteristics. For example, they assessed whether White people were more likely to be stopped in predominantly non-White precincts and vice

___________________

⁹ One criticism of this research is that the location of the stop and the make and model of the vehicle may often be sufficient indicators of a non-White driver, given racial and ethnic income disparities and the racial residential patterns of Oakland, California, during the period studied.

versa. The findings indicate that Black people and Hispanic people were stopped at rates that were disproportionate to their arrest rates. There was some evidence that “being out of place” increases SQF for all races/ ethnicities. This study also found evidence of lower hit rates (likelihood of arrest, conditional on SQF) for Black people and Hispanic people relative to White people (Gelman, Fagan, and Kiss, 2007), the interpretation of which will be discussed in detail in the next section. Studies like these use broad-area statistics to develop benchmarks. However, other research in New York City suggests that SQFs were highly concentrated on specific streets within neighborhoods, which raises questions regarding the ability to rely on the comparisons made using broad-area benchmarks (Weisburd, Telep, and Lawton, 2014).

Beckett and colleagues (2005) and Beckett, Nyrop, and Pfingst (2006) employed surveys of chronic drug users to estimate the racial composition of those engaged in drug delivery. These two studies were primarily interested in explaining the high proportions of Black people among those arrested in Seattle for drug offenses overall (Beckett et al., 2005) and for drug delivery offenses (Beckett, Nyrop, and Pfingst, 2006). The authors noted the relatively small Black population of Seattle (roughly 7% of the population) and the high proportion of Black people among drug arrestees (more than 50%), which often resulted from strategic and concentrated enforcement of drug laws. They used several sources to measure the composition of serious drug users and those engaged in drug delivery. First, they conducted a survey of participants in needle exchanges, recording the race/ethnicity of users, the drugs in the needles being exchanged, and the perceived race/ ethnicity of the person from whom the exchanging person purchased or received drugs most recently. Second, they spent roughly 40 hours observing outdoor transactions in two prominent open air drug markets in Seattle, recording perceived race/ethnicity of buyers and sellers and the drugs being sold (inferred from the knowledge of key informants or in some instances from subjects attempting to sell the observer drugs). Third, the 2005 study incorporated information on the racial composition and specific substances used among those participating in publicly funded (either fully or partially) substance abuse treatment programs. Beckett and colleagues used these datasets to estimate the composition of users by substance (the 2005 study) and the racial composition of drug deliverers by substance (the 2006 study).

In these studies, White people constituted the overwhelming majority of heroin, meth, and MDMA users and deliverers in Seattle, while Blacks constituted the plurality of crack cocaine users and those involved in crack delivery. The authors estimated the proportion of street transactions involving each substance and found that the proportion of arrests involving crack cocaine exceeded by a fair amount the proportion of street transactions involving that drug. They also found that drug arrests involving crack

yielded less in terms of confiscated product, money, and weapons. Buy-bust arrests were also less productive along these dimensions, relative to arrests of individuals in private spaces involving warrants. The disproportionate representation of Blacks among drug arrestees appeared to be due in large part to a focus on crack cocaine and a focus of resources on buy-busts occurring outside. The authors noted anecdotal evidence from buy-busts of a local prioritization of enforcement targeted toward crack cocaine, with undercover officers explicitly looking to buy crack and in some instances not buying heroin or meth when offered. They attributed this prioritization to a racialized script regarding what it means to be a drug addict and a drug offender and the ensuing direction of policing resources toward drug offenses committed by Black people, despite their minority status in the city and among drug users and deliveries not involving crack cocaine.

The conclusion from these Seattle studies has been called into question by Engel, Smith, and Cullen (2012). These authors analyzed the geographic distribution of drug arrests in Seattle for a later period than that studied by Beckett and colleagues (2005) and Beckett, Nyrop, and Pfingst (2006) and assessed how the geography of drug arrests compares to the geography of drug-related calls for service through the city’s 911 system (as captured and recorded through the Seattle Police Department’s computer-aided dispatch system). Engel, Smith, and Cullen (2012) also coded the suspect’s race, as indicated from the narratives of calls for service, and compared the composition of described suspects to the composition of drug arrests. They found a tight association, across geographic units of varying levels of spatial disaggregation, between drug arrests, drug-related calls for services, and reported crimes. The authors also found close correspondence between the racial composition of drug arrests and the composition of suspect descriptions in 911 calls. Based on these findings, the authors contested the conclusion that the enforcement focus on crack cocaine or some other form of racialized framing of the local drug problem was driving racial disparities in arrests. Instead, these authors argued that the geography of police deployment driven by responses to calls for service explains enforcement priorities and that racial disproportionality was much less when benchmarked against suspect descriptions from the public. The authors left as an open question whether calls for service by the public are racially biased.

There are several notable differences between the Engel, Smith, and Cullen (2012) study and the research in Beckett et al. (2005) and Beckett, Nyrop, and Pfingst (2006) that make it difficult to draw head-to-head comparisons of the conclusions and findings. To start, the two sets of research teams seem to be asking different questions in their study designs. Beckett and colleagues essentially asked whether the racial composition of drug arrests matches the racial composition of those who violate drug law in Seattle (using rehab statistics, needle exchange surveys, and observation of

drug market activity as benchmarks). Given that they found heavy overrepresentation of Black people among drug arrests, they then asked what policy and practice factors may be behind these disparities. Engel, Smith, and Cullen (2012), on the other hand, asked whether responsive deployment of police resources (to calls for service and crime rates) can explain the geography of arrests and whether the racial composition of suspect description from the public better match the racial composition of arrests. It is entirely plausible that both statistical associations are valid; namely, that (1) Black drug users and deliverers face a higher risk of arrest relative to White drug users and deliverers, and (2) the racial composition of reported suspects matches the racial composition of arrests. Hence, it is not clear that this more recent research contradicts the earlier findings by Beckett and colleagues, as it essentially asks and answers fundamentally different questions.¹⁰

With the exception of the benchmarking study of SQF and the drug enforcement study in Seattle, there are very few benchmarking studies relevant to proactive policing that assess whether specific policing efforts have disparate impacts on non-White communities. Employing standard benchmarking strategies to assess the outcomes of geographically focused and person-focused policing strategies would likely face particular difficulties. For example, defining the racial composition of individuals congregating on specific street corners that would be the target of proactive SQF or the racial composition of individuals who are candidates for lever-pulling strategies or strategic preemptive call-ins is likely impossible using standard administrative data or local-area data collected and published by the U.S. Census Bureau. In fact, the extensive field work and original data collection by Becket and colleagues in Seattle was intended primarily to overcome these data limitations.

___________________

¹⁰ In her published reaction to Engel, Smith, and Cullen (2012), Beckett (2012) also raised several objections that are worth noting. First, the Seattle Police Department testified that calls for service were not used for deployment decisions during the time period under analysis in the earlier studies. Second, Engle and colleagues did not disaggregate arrests by drug type, and Beckett (2012) showed that the racial disparities in arrest rates can largely be attributed to the high proportion of arrests that were crack related (i.e., geography does not explain racial arrest disparities). Addressing the bivariate analysis showing the strong association between crime and drug arrest, Beckett (2012) raised concerns about possible reverse causality: to the extent that the police focus on specific areas to enforce drug law, they will also encounter and record more crime through their presence in the area. She even raised the concern that drug-related calls for service may be endogenous to enforcement policy, as the Seattle Police Department unveiled a drug offense monitoring system and actively encouraged the public to report drug activity into the CAD (computer-aided dispatch) system. Beckett (2012) also raised the possibility that the community outreach effort focused on neighborhoods where drug arrests were concentrated.

Outcome Tests for Racial Disparities in Treatment

As noted above, a second body of research employs outcome tests for unjustifiable racial disparities in police–citizen interactions. Outcome tests analyze the results of police–citizen interactions and infer racial bias from racial differences in the productivity of a police stop, usually measured by whether a search results in the detection of contraband. As noted earlier, the method of testing for biased behavior by testing for differences in outcomes has been extensively debated. Here, we review the key studies that have employed this strategy to study police practices in the United States. Again, while much of this research has focused on generating tests for racial profiling by the police in general, a few notable studies have applied the methods to study the fairness and efficiency of SQF as a proactive policing strategy.

Knowles, Persico, and Todd (2001) provided one of the earliest outcome test studies of police behavior. The authors employed data on police stops and searches by the Maryland State Police occurring along Interstate 95 during the 1990s and compared hit rates, using various definitions of contraband and guilt, by the race/ethnicity of the searched drivers. The principal findings from this study are that hit rates for searches involving Black drivers were equal to or higher than those involving White drivers (roughly 32% higher for the most inclusive definition). By contrast, hit rates for Hispanic drivers were very low: equal to roughly one-third those for Black and White drivers.

Sanga (2009) reanalyzed the dataset used in Knowles, Persico, and Todd (2001) by extending the analysis to all driver searches conducted by the Maryland State Police during the time period analyzed in the original study and extending the analysis through 2006. Note that the earlier study focused only on searches of drivers resulting from stops made along I-95, which constituted roughly 30 percent of all searches. Analysis of all searches yielded statistically significant differences in hit rates, with hit rates for Black searches 6 percentage points lower than those for Whites, during the period 1995–1999. Estimates using the extended period from 1995 through 2006 yielded even larger disparities, with Black hit rates 10 percentage points lower than White hit rates, suggesting that the overall disparity in hit rates increased in the latter part of the extended period. Similar to the original finding in Knowles, Persico, and Todd (2001), Sanga (2009) found very large Hispanic-White differentials in hit rates, with the Hispanic hit rate for the entire period approximately 30 percentage points lower than the hit rate resulting from searches of White drivers.

Ayres and Borowsky (2008) analyzed field development report data (completed after each motor vehicle stop or interaction with a citizen) for the Los Angeles Police Department for the period July 2003 through June 2004. The authors analyzed disparities in stop rates per 10,000 residents

and in differences in the likelihood of being searched, asked to exit the car, or frisked conditional on being stopped, as well as racial disparities in a series of outcomes that included arrest, citation, and the detection of contraband, weapons, or drugs. Using reporting districts as the unit of observations, the authors found that controlling for geographic variation in crime rates, demographics, unemployment rates, and poverty does not explain racial disparities in the stops per 10,000 local residents. Two divisions in particular (Central and Hollywood) had stop rates that exceeded 10,000 per 10,000 Black residents The authors also found that Black residents who were stopped were more likely to be arrested, less likely to be cited, more likely to be asked to exit the vehicle, more likely to be frisked, and had lower hit rates relative to White residents. Similar disparities relative to White residents were found for Hispanic residents. They also performed tests of whether officer race mediated these disparities and found evidence of less biased behavior against non-White citizens by non-White officers.

Anwar and Fang (2006) developed a model of police stops and searches whereby officers develop a posterior likelihood of a successful hit and search all motorists where the expected net benefits are positive. In their model, officers are permitted to behave in a non-monolithic manner, in the sense that an officer is permitted to determine differential suspicion thresholds by motorist race. The authors were particularly interested in the infra-marginality problem that plagues outcomes tests of racial profiling, and they used their model to generate a sufficient but not necessary condition for evidence of racially biased treatment that in turn provides a low-powered empirical test for racial profiling. To be specific, their model implies that when officers do not engage in biased behavior, particularly biased behavior motived by statistical prediction, differences across officers of different racial groups in the propensity to stop or search motorists (conditional on motorist race) does not depend on the race of the motorist. In other words, if White officers stop White drivers at a higher rate relative to Black officers, the satisfaction of this condition would imply that White officers also stop Black motorists at a higher rate.

The authors used this model to analyze data on highway stops in Florida. They found that White officers searched individuals of all races at higher rates, followed by Hispanic officers, with Black officers third. Similarly, cross-officer-race disparities in hit rates yielded rankings that do not depend on the race of the motorist. However, the authors did find higher search rates for Black people, followed by Hispanic people and then White people, and lower hit rates for Hispanic people and Black people relative to White people. White officers searched all groups at the highest rates, but the differences by motorist race were largest for White officers, as were the hit rate differentials.

Antonovics and Knight (2009) analyzed data on police stops by the

Boston police department, focusing specifically on differences in the propensity to search motorists of different races, according to the race of the officer. They developed a simple model whereby heterogeneity by officer race in the costs of searching yields differences in the propensity to stop and search motorists of different races. Similar to previous work in this area, they formalized “taste-based” bias (animus-related biased behavior, in this committee’s terminology) as a lower cost associated with stopping members of the biased-against group. The key prediction here is that to the extent that officer race interacts with motorist race in determining the likelihood that a stop results in a search, then there is a disparity in search costs that cannot be justified by statistical prediction in actual racial disparities in the propensity to offend. The authors’ analysis found evidence of such an interaction effect, with Black officers more likely to search White motorists relative to White officers, and White officers more likely to search Black motorist relative to Black officers.

Several studies of the use of SQF employed outcome tests of racial bias. Persico and Coviello (2015) analyzed New York City data for the period 2003 to 2012. The principal strategy in this study was to assess whether Black and Latino people who were stopped by the police were less likely to be arrested than White people who were stopped by the police. The proposition here is that to the extent that an arrest is a signal of (accurate predictor for) actual law breaking, lower arrest rates for Black or Latino stops relative to White stops would be evidence of disparate and biased treatment of non-White people. The authors found that Black stops were less likely to result in an arrest relative to White stops and that the difference in arrest rates is statistically significant. However, controlling for precinct of arrest led to a higher conditional arrest rate for Black people. To be sure, the authors’ conclusions are premised on the assumption that an arrest is a race-neutral gauge of suspect culpability. They offered evidence that, conditional on reported offense, Black people were only slightly more likely to be arrested than White people. However, to the extent that the reported offense is endogenous to the interaction between the officer and citizen (i.e., officers list more serious charges for people they wish to arrest), this evidence is of limited value. Moreover, given the enormous racial disparities in the frequency of coercive interactions with the police, one might expect that the SQF incidents involving Black and Latino people are more likely to be tense exchanges, with perhaps an elevated likelihood of escalation.

Goel, Rao, and Shroff (2016) analyzed racial disparities in the composition of police stops in New York City due to suspicion of criminal possession of a weapon. The authors used data on 760,000 such SQF incidents occurring between 2008 and 2012 to assess the productivity of such stops in terms of weapons recovered, racial disparities in productivity as measured by average hit rates, the degree to which these disparities are ex-

plained by localized (i.e., precinct level) variation in practice, and whether improvements in efficiency could reduce racial disparities in the incidents of these stops while still confiscating the majority of confiscated weapons.

The authors modeled the likelihood that an SQF incident results in a weapons seizure using the first 3 years of data. Their logistic model specification includes data on demographics (race, age, gender), physical characteristics (build, weight), time and day of the stop, precinct, a distance-weighted measure of the productivity of recent such stops, a complete set of precinct fixed effects, and full interactions between all terms. They estimated the parameters of their model using the 2008 through 2010 data and then used the trained model to estimate the ex ante likelihood of recovering a weapon for more recent years. The key innovation is that they used factors observed ex ante to the stop to predict the likelihood of discovering a weapon and used their trained model to estimate the complete distribution of the ex ante hit-rate probability across all stops occurring during the out-of-sample period.

Several findings resulted from this study. First, average hit rates were lower for Black people relative to White people, and there were large differences in the distribution of ex ante likelihood by race. Second, much of the racial disparity can be explained by variation in practice by precinct, though White hit rates were still higher within precincts. Third, the authors showed that the ex ante likelihood of recovering a weapon is very low for a large proportion of searches. They also showed that focusing on the subset of searches with high ex ante likelihood could greatly curtail the volume of searches while recovering nearly all of the weapons discovered through the less-focused SQF program (Goel, Rao, and Shroff, 2016).

A key issue left unanswered in this study concerns whether the hit rates are in part a function of the level of SQF activity. The authors estimated that the 6 percent of stops with the highest ex ante probabilities of uncovering weapons netted 50 percent of the weapons recovered. Similarly, 60 percent of stops generated 90 percent of the weapons recovered. These tabulations implicitly assume that the likelihood of carrying a weapon is not sensitive to the extent of SQF activity initiated by the police. To the extent that SQF deters the unlawful carrying of firearms, the ex ante probability distribution for weapons carrying estimated under conditions when SQF is used liberally may not match the comparable distribution in an environment when SQF is rarely used.

CONCLUSION

Concerns about racial bias, particularly bias driven by animus, loom large in discussions of policing today, and these concerns must be framed in the context of the history of the police as agents who sometimes en-

forced discriminatory laws and social norms, often through violence. The currently high levels of distrust in the police must be understood through this historical frame because that history continues to influence Black views about the police.

Having outlined this history, the committee also notes evidence that racial animus has become less central to American society and to policing relative to the 18th and 19th centuries. However, there are a number of indications that racial animus, and racially biased behavior in general, continue in more subtle forms, and some proactive strategies may have features that align with psychological risk factors for biased behavior by police officers. Because none of these areas in the psychological literature has tested how the observed laboratory effects may or not may not generalize to street-level policing decisions, the evidence from this field is not sufficient for the committee to draw any conclusions regarding its specific impacts on proactive policing. Nevertheless, the general evidence for the continuing existence of both negative racial attitudes, whether conscious or unacknowledged, and racially biased behavior support the importance of efforts to study the role of racial bias in contemporary police behavior generally and in proactive policing in particular.

Inferring the role of racial bias in contributing to disparate impacts is a challenging question for research. Existing research demonstrates that concentrated enforcement efforts in high-crime areas and on highly active individuals can lead to racial disparities in police–citizen interactions. One consequence of this geographic concentration is observable in the geographic concentrations of arrests, SQFs, and general police activity in non-White neighborhoods in many cities. From a statistical standpoint, this means that regression adjustment that accounts for local measured crime rates or that includes general spatial control variables—as in several studies that adjusted for precinct of arrest or patrol districts within jurisdiction—frequently generates findings of substantially reduced, or even eliminated, evidence of racial bias. But, as community advocates argued to the committee (see Appendix A), this does not address or respond to the fact that crime in these areas is itself an outcome of social forces that may have been generated by long-term inequalities and discrimination in non-White communities.

Studies that seek to benchmark citizen-police interactions against simple population counts or broad publicly available measures of criminal activity do not yield conclusive information regarding the potential for racial bias in proactive policing efforts. This statement is based on several facts. Proactive strategies tend to be geographically focused, sometimes person-focused, and often problem focused, where the definition of the problem is articulated a priori through some executive local analysis. Assessing disparate impact requires detailed information on the geography and nature

of the strategy, in order to identify an appropriate benchmark. In addition, assessing whether the disparate impact is indicative of racial bias further requires localized knowledge of the relative importance of the problem at hand. For example, should the Seattle police focus more on crack cocaine, which tends to be sold and used by Black people, or heroin, which is relatively more likely to be used and sold by White people?

Although research that tests for differential outcomes of police–citizen interactions is potentially informative regarding the disparate impacts of proactive policing efforts, research on this issue tends to be undertheorized or defines the objectives of police so narrowly as to limit the ability to draw broad inferences from existing empirical findings. For example, the infra-marginality problem detailed early in this chapter (refer to Box 7-1) makes it difficult to conclude that there is no bias from a finding of equal hit rates across racial groups. If one’s prior assumption is that non-White people tend to offend at a higher rate, a finding of lower hit rates for non-White searches might support the conclusion that bias, and potentially animus-driven bias, exists, though even this conclusion is conditional on several assumptions regarding the objectives pursued by the police and the responsiveness of individuals to the threat of being searched.

Some of the most compelling evidence on the potential impact of aggressive enforcement-based proactive policing and increased citizen–police contacts on racial outcomes relates to the use of SQF in New York City. This research seeks to model the probabilities that police suspicion of criminal possession of a weapon turns out to be justified, given the information available to officers when deciding whether to stop someone. This work has found substantial racial and ethnic disparities in the distribution of these probabilities, suggesting that police in New York City apply a lower threshold of suspicion to Black and Hispanic residents.

While this chapter has reviewed the psychological science related to police bias and detailed what researchers have discovered by examining stop rates and outcomes tests of police behaviors, there are a number of areas not engaged by this chapter. That is largely because scholarship of police behavior has disproportionately focused on crime and delinquency and less frequently on the racial causes and consequences of those behaviors. As a result, there simply is not much literature on the downstream consequences of police contact between communities and police. There is promising research on the negative consequences of police contact on institutional trust (Lerman and Weaver, 2014b; Jones, 2014; Weaver and Lerman, 2010) and racial identity development (Brunson, 2007; Brunson and Miller, 2005, 2006). But the evidence base is still thin regarding how police contact—much less proactive policing—may influence family dynamics, educational outcomes, employment, and housing decisions. Similarly, there are some studies exploring the link between racial residential segregation and police

size (Kent and Jacobs, 2005; Stults and Baumer, 2007) and some work demonstrating the overrepresentation of Black faces in local news coverage of crime (even compared to actual rates of arrest) (Dixon and Linz, 2000a, 2000b) and how that coverage influences public opinion on what crime policy should be (Dixon, 2006, 2008a, 2008b; Dixon and Azocar, 2007). However, there is not a robust empirical literature exploring how racial attitudes or politics may causally influence police policy or officer decision making.

That the scholarship does not exist should not be taken to suggest either the presence or absence of racial disparities, racial bias, or racial animus. It should also not be taken to mean that the topics are unimportant; indeed, our review of ethnographic evidence and consultations with community groups (see Appendix A) suggests that these topics are central to fully quantifying the net social costs and benefits of proactive policing policies. However, the committee notes that even basic measures of the impact of police policies and practices on communities are undeveloped or nonexistent, relative to estimates of the impact of policing on crime. For example, there is no component of the Uniform Crime Reports that records officer-initiated stops or incidents of force, in aggregate or broken out by race. The National Incident Based Reporting System (NIBRS), which contains significantly more detail about police actions, crime, victims, and arrestees, currently does not cover police agencies operating in places where roughly 70 percent of the U.S. population lives, and it covers none of the nation’s largest urban areas.¹¹ Further, while there are a number of well-developed estimates of the cost of crime (Heaton, 2010), the committee is unaware of any estimates of the costs of racial disparities in criminal justice contact, the costs of racial profiling or other racially biased behaviors by police, or the cost of racial animus in policing. For example, if place-based policing reduces the probability that most non-White people are subject to racial profiling but increases the rate at which non-White people in certain crime hot spots have unwanted police interactions that are perceived to be racially biased, does this increase or decrease social welfare? The committee is currently unaware of any quantitative metric for this issue. Until these basic measurement issues are addressed, any attempt to quantify the net social cost or benefit of policing policies will be necessarily inaccurate.

The committee calls for scholars to produce research that will expand the conceptual map of what causes police behavior; what the consequences of those behaviors are; and how race, gender, class, and other vulnerable identities may play a role in both. Ethnographic and qualitative evidence

___________________

¹¹ As of 2015, only one agency with a jurisdiction of more than 1 million people reported NIBRS data to the Federal Bureau of Investigation (the Fairfax County, Virginia, police department), and only 32 agencies with jurisdiction of more than 250,000 did so.

on community and citizen responses to policing should be used to develop hypotheses that can be rigorously tested with quantitative data. The current lack of such scholarship severely limits the ways in which evidence-based approaches to policing can address community concerns about group-based disparities.

The committee has drawn the following main conclusions regarding racial disparities in proactive policing:

Taken in its totality, this chapter suggests the need for more systematic research in specific areas, including more research involving police officers acting in field settings (see Chapter 8). There are large disparities in all manner of interactions and outcomes with the police by race and ethnicity. Given the current state of the empirical evidence, it is difficult to infer the meaning of these disparities. But given the long history of policing bias against non-Whites, and especially Black Americans, it is clearly time for this question to be more carefully examined.

This page intentionally left blank.