The Rise of Evidence-Based Policing: Targeting, Testing, and Tracking

Evidence-based policing is a method of making decisions about “what works” in policing: which practices and strategies accomplish police missions most cost-effectively. In contrast to basing decisions on theory, assumptions, tradition, or convention, an evidence-based approach continuously tests hypotheses with empirical research findings. While research on all aspects of policing grew substantially in the late twentieth century, the application of research to police practice intensified in the early twentyfirst century, especially for three tasks that make up the “triple-T” strategy of policing: targeting, testing, and tracking. Evidence-based targeting requires systematic ranking and comparison of levels of harm associated with various places, times, people, and situations that policing can lawfully address. Evidence-based testing helps assure that police neither increase crime nor waste money. Tracking whether police are doing what police leaders decide should be done may grow most rapidly in coming years by the use of GPS records of where police go and body-worn video records of what happens in encounters with citizens. Can research help improve democratic policing? For over four decades, numerous colleagues and I have struggled to answer that question. When we began, policing was done very differently from how it is Lawrence W. Sherman is Wolfson Professor of Criminology at Cambridge University, a director of the College of Policing (UK), Distinguished University Professor at the University of Maryland, and director of the Cambridge Police Executive Programme. He is grateful for comments on earlier drafts by Heather Strang, Peter Neyroud, Daniel Nagin, Stephen Mastrofski, John Eck, Alex Piquero, David Weisburd, Lorraine Mazerolle, Cynthia Lum, and Michael Tonry. 378 Lawrence W. Sherman done now. In the years to come, it will be done even more differently. Yet the trajectory of change is likely to remain consistent: from less research evidence to more and from lesser to greater use of the evidence available. The 50 years covered in this volume bear witness to the rise of evidence-based policing in both the quantity of evidence and its influence on police practices. This essay reports on my ongoing participant-observation study of these changes, for which I have tried to be a primary instigator. Policing in 1975 was largely delivered in a one-size-fits-all strategy, sometimes described as the “three Rs”: random patrol, rapid response, and reactive investigations (Berkow 2011). After World War II, random patrols in police cars were promoted on the theory that police “omnipresence” would deter crime (Wilson 1950). In the 1960s, the advent of three-digit emergency phone numbers turned random patrol into an airport-style “holding pattern” for rapid response, also based on a theory of deterrence, but producing a new theory of organizational action: Albert J. Reiss’s (1971) distinction between reactive and proactive actions. What police did when they reacted to a citizen call was not subject to much police agency direction or analysis. The main organizational requirement was to arrive, do something, and leave as quickly as possible. As Reiss reported, his 1966 Chicago study found patrol officers spending 14 percent of their time “reactively” answering such calls and investigating crime, about 1 percent of their time “proactively” stopping people at their own initiative, and 85 percent of their time in unstructured random patrolling. Reactive investigations, when required, were taken over by detectives, who in theory investigated all reported crimes but in practice closed most cases as unsolvable. Cases that were solved relied primarily on the evidence presented to them by crime victims and the patrol officers who arrived first on the scene (Greenwood and Petersilia 1975). By 1975, the three Rs had become the standard model of urban policing (Skogan and Frdyl 2004, p. 223) across the United States as well as in the United Kingdom and other predominantly Anglo-Celtic cultures (e.g., Canada, Australia, New Zealand). There was almost no targeting of patterns or predictions of crime or disorder, no testing of what worked best to prevent or solve crimes and problems, or much tracking and managing of what police were doing, where, when, and how, in relation to any specific objectives. Most police agencies lacked computers. The few computers in use in 1975 were mainframes deRise of Evidence-Based Policing 379 signed for dispatching police cars, not analyzing crime data. Police leaders rarely discussed crime at all unless a particularly gruesome crime received substantial publicity. By 2012, the three Rs were changing into what I describe in this essay as the “triple-T” of targeting, testing, and tracking. While the standard model is far from gone, its resources are increasingly guided by statistical evidence. In the emerging triple-T strategy, both patrol and detective managers had moved toward far greater proactive management of police resources. Although these changes are far more evident in some police agencies than in others, Anglo-Celtic policing is increasingly targeting scarce resources on evidence of large, predictable, and harmful statistical patterns rather than on isolated cases (Spelman and Eck 1989). Compared to 1975, these focused police strategies are far more elaborate and differentiated, choosing from a wider range of priorities and objectives on the basis of extensive data analysis. Police methods have also become far more subject to testing, with evaluation and debate over “what works,” the core idea of evidence-based policing (Sherman 1984, 1998). Yet the most evident use of new statistical evidence is the growth of tracking and managing what police were or were not doing in relation to the dynamic patterns of crime and public safety problems. Examples of the growth of evidence-based policing abound. They include the US Supreme Court’s citation of research on tests of restrictions on police shootings (Fyfe 1980, 1982; Sherman 1983), Tennessee v. Garner (471 U.S. 1 [1985]), a decision that fostered a major reduction in police killings of citizens in the United States (Tennenbaum 1994). Overall reductions in homicide have been associated with an increased targeting of US police work toward micro-level crime “hot spots” (Weisburd and Lum 2005; Police Executive Research Forum 2008; Braga and Weisburd 2010), where research showed that crime is heavily concentrated (Sherman, Gartin, and Buerger 1989). By 2012, evidence-based targeting for problem-oriented policing (POP) has, since Goldstein (1979) first proposed it, gradually become respected as real police work, not “social work,” supported in some agencies with new case management systems. Crime analysts are more likely to identify repeat crime places, victims, suspects, situations, and other patterns to support targeting decisions for patrol, POP, and detectives. Even reactive detective work is more proactive, police-initiated, in gathering new evidence at crime scenes, using scientific meth380 Lawrence W. Sherman ods to identify suspects and prove cases (Roman et al. 2009). What patrol officers do when responding to crime scenes has become more likely to be a method that has been tested. Where officers are at any moment is more likely to be tracked by global positioning satellite (GPS) systems generating weekly management analysis reports. The best test of evidence-based policing is whether it has improved public safety and police legitimacy. There is certainly a correlation over time between the rise of evidence and a decline in serious crime, in both the United States and the United Kingdom. Proof that serious crime dropped because of evidence-based policing, however, is more elusive. There is some micro-level analysis of the New York City crime drop that offers a greater basis for causal inference than national trends (Zimring 2012), but it is impossible to rule out other plausible causes that also match the timing of the crime drop. There is also a danger that a sharper focus on crime can undermine police legitimacy, especially since proactive policing is structurally less legitimate than reactive policing with a personal victim’s endorsement (Reiss 1971, p. 11). Yet there is no evidence that the rise of evidence in policing has caused reductions in police legitimacy. Indeed, more data about legitimacy have been incorporated into police management as part of evidencebased tracking than ever before. Overall, the vast scale of the rise of evidence in policing leaves it without a fair comparison group, just as it does in medicine (Sagan 1987), leaving causation unknowable. Still, the macro trends in reduced mortality suggest that we should celebrate the rise of evidence in both medicine and policing. The changes from 1975 to 2012 raise the further question of what Anglo-Celtic policing will look like by 2025. The answers cannot be predicted just by extrapolating the trends of four decades, or even— ideally—by a causal model for prediction (Silver 2012). But by explaining why policing changed as it did from 1975 to 2012, a forecast using the same causal model can identify a range of possible scenarios for policing in 2025. Chief among these will be the possible global influence of the 2012 creation of the first professional body ever charged by a national government with recommending police practices on the basis of continuous review of new research evidence on what works: the United Kingdom’s College of Policing (http://www.college.police .uk). This body has tremendous potential to follow the pathway to innovation Johnson (2010) associates with such major advances as the printing press, which was inspired by the wine press: a lateral-thinking Rise of Evidence-Based Policing 381 style of adaptation of an idea used in one setting (such as evidencebased medicine) to another (such as evidence-based policing). The Rise of Evidence. The causal model this essay suggests for these changes is a dynamic system that links external demands on police (to cut crime, cut costs, or build legitimacy) to research evidence on how to meet those demands, supported by the increasing availability of innovative technologies. While external factors drive the availability of both research evidence and technology, how policing responds depends on a fourth factor: the governmentally generated human capital of education and police professionalism. The role of government policies is crucial in setting salaries and recruitment standards that shape the quality of police skills and creativity in general and of police leaders in particular. Those human resources are so influential that the entire causal model can reasonably be said to call for evidence-based professionalism. Inside this abstract causal model are many human beings struggling hard for change: influential leaders of policing, research, and efforts to combine the two. These include UK Home Secretary R. A. Butler creating a research unit in 1958; future FBI Director Clarence Kelly supporting the Kansas City Patrol Experiment in 1971 (Kelling et al. 1974); Minneapolis Police Chief Anthony V. Bouza supporting the first random assignment of arrest in 1981 (Sherman and Berk 1984); criminologist David Weisburd and his colleagues (Weisburd et al. 2004, 2006; Weisburd, Groff, and Yang 2012) analyzing criminal careers of places in Seattle and Jersey City; New York Police Department (NYPD) Commissioner William Bratton creating COMPSTAT in the 1990s; Assistant US Attorney General Laurie Robinson making evidence from police testing readily accessible at http://www.crime solutions.gov in 2010; and UK Chief Constable Peter Neyroud’s proposal for the evidence-based College of Policing in 2011. What these leaders generally shared—with, most notably, NYPD Commissioner and US Police Foundation President Patrick V. Murphy—was a loosely articulated but deeply held belief that greater use of research could help transform policing into a more legitimate and respected profession. A profession is widely defined as a public-interest occupation that restricts entry to those who have mastered knowledge and skills needed to provide a particular set of complex services. In the past century, the level of knowledge deemed necessary for a profession has become 382 Lawrence W. Sherman linked to requirements for university-based education. It is only a century, for example, since the radical proposal to force doctors to attend university was presented to the United Kingdom’s Haldane Commission (Sherman 2011b). Similarly, proposals in repeated US commission reports in the 1960s and 1970s recommended that a university degree be a requirement for appointment as a police officer (Sherman and National Advisory Commission on Higher Education for Police Officers 1978). That idea remains controversial and widely unacceptable on both sides of the Atlantic. By conventional definition, then, policing has yet to become a profession. Yet by the same definition, policing has made great strides toward professionalism. The past 40 years have seen a huge increase in the educational levels of police leaders. Almost half of English chief constables in 2010 had been educated at the Cambridge University Police Executive Program, and three decades of major city chiefs in the United States attended the nondegree Harvard Police Executive Sessions. The role of universities in disseminating police research has been a remarkably successful top-down change process. Rising levels of police chief education have also fostered more support for research. First led by external scholars, and now increasingly led by police leaders in partnership with universities, the production of new research evidence has helped police respond to external demands for improvement. At minimum, research is perceived to be helpful with such demands. That is especially true when there is internal or external controversy about which police methods are best and whether new technologies are cost-effective. Even a radical change in structures of police accountability can enhance the value of police research evidence. The 2012 devolution of control over chief constables in England and Wales to locally elected “police and crime commissioners,” for example, created a potential for disputes between elected officials and professional police leaders. One possible solution is the professional College of Policing setting independent, evidence-based standards of police practice. Whether this institution, or its cognates in other countries, can create a global demand for more production and use of evidence is a crucial question for the next decade. Targeting, Testing, and Tracking. Both the demand for, and uses of, research evidence have become clustered around three strategic principles: Rise of Evidence-Based Policing 383 1. Police should conduct and apply good research to target scarce resources on predictable concentrations of harm from crime and disorder. 2. Once police choose their high-priority targets, they should review or conduct tests of police methods to help choose what works best to reduce harm. 3. Once police agencies use research to target their tested practices, they should generate and use internal evidence to track the daily delivery and effects of those practices, including public perceptions of police legitimacy. The growing adoption of those three principles has given shape to what is increasingly called evidence-based policing (EBP). Broadly defined as the use of best research evidence on “what works” as a guide to police decisions (Sherman 1998), the EBP framework was meant to be only a method of making decisions rather than a substantive strategy for police operations. In the 15 years from 1997 to 2012, however, police have shaped their own emergent definition of EBP as a substantive strategy for managing large police agencies on the basis of these three principles. The rise of the triple-T strategy did not emerge from any theoretical plan to use evidence in such a coherent strategy. The clustering of evidence around three key strategic tasks was driven as much by innovative police leadership as by police scholarship. Its success resulted from the surprisingly rational convergence of police reform with a flood of new research in criminology. This essay has four sections. Section I discusses the key social trends, institutions, and people who helped shape research aiming to improve policing. The triple-T strategy emerged from the users of that evidence. Section II offers major examples of how much the evidence has grown since 1975 and how widely police have applied that knowledge. These examples address all three triple-T principles, including instances in which evidence and practice have sharply diverged. They are selected to highlight the wide range of what has already been accomplished, even as so much still remains to be done. Section III analyzes conceptual issues and confusion that hold back progress in targeting, testing, and tracking. The issues are arguably little different from those arising in the development of evidence-based practices in professional baseball, election campaigns, or marketing. They reflect the current struggle between “system I” and “system II” 384 Lawrence W. Sherman thinking (Kahneman 2011) in a wide range of endeavors, all of which face the same challenge policing faces—to get much better at what they do in the next decade. In the conclusion I suggest 10 things institutions can do to institutionalize research evidence in ways that foster more fairness and effectiveness in democratic policing. These are the most important points I make in this essay: 1. The evidence base for police decisions has grown enormously since 1975. 2. Use of that evidence lags behind the knowledge, but use has also grown. 3. Most police practices, despite their enormous cost, are still untested. 4. Targeting and testing require highly reliable measures of crime and harm. 5. Crime rates and counts are by themselves misleading; a crime harm index offers far better evidence to guide police decisions. 6. Police in 2012 used evidence on targeting much more widely than evidence from testing. 7. Research on tracking police outputs remains largely descriptive and incomplete, with great room for using new technologies to improve the quality of evidence. 8. More use of evidence can increase police legitimacy, both internal and external. 9. The State Boards of Police Officer Standards and Training in the United States and the College of Policing in the United Kingdom will be key institutions in making policing more effective, along with the practitioner-led Society for EvidenceBased Policing. I. The Rise of Evidence for Police Decision Making The rise of evidence in policing is mostly an Anglo-American story, with events on each side of the Atlantic influencing the other. Both sides began with parallel social trends in the 1960s that challenged police legitimacy. Both the United Kingdom and the United States faced rising crime rates and increasing racial tensions. Authoritative commissions in both set out ambitious agendas for “upgrading” the police, as one think tank put it (Saunders 1970). Both saw policing Rise of Evidence-Based Policing 385 become a matter of national politics in unprecedented ways. The different responses of police in each country helped build a special intellectual relationship that was conceptually far deeper than the military collaboration of World War II. One continuing theme on both sides of the Atlantic has been that “they do policing better” on the other side. A. Setting the Stage By the early 1970s, Anglo-American policing faced a kind of Christmas tree of ideas for police reform, each idea an ornament unconnected to the others. One idea was that police needed more research, including experiments and demonstration projects, with criminologists and other social scientists becoming part of the police agency workforce focusing on crime prevention (President’s Commission 1967, pp. 25– 27). There was no consensus about the questions research should answer. There was no particular vision of a new policing strategy to which research could contribute. Yet belief in innovation and evaluation as a strategy in itself became a major force in police reform (Reiss 1992). That belief led to public and charitable investments in new institutions for police research. The growing role of the British Home Office in police research helped inspire the creation of a similar function in the US Department of Justice. Yet neither government agency’s efforts had much impact until after the Ford Foundation in 1970 funded the Police Foundation in Washington to “foster innovation and improvement in American policing.” The bold experiments the foundation conducted led government funders to test policing more deeply. In less than a decade, the Police Foundation’s and government-supported research discredited the intellectual basis for the three-Rs strategy of random patrol, rapid response, and reactive investigations. What replaced that strategy was not a coherent new theory. Only in retrospect can we use the triple-T framework of targeting, testing, and tracking to make sense of what emerged. The multicentered work of examining current practices, designing innovations, and evaluating new programs produced something that scholars now call “emergence,” the confluence of properties arising from a combination of elements not found in any one of the elements ( Johnson 2001). The essential new property is the capacity to lead police organizations with dynamic evidence rather than static doctrine. This property arose initially from the destruction of the prevailing three-Rs model. That, in turn, opened 386 Lawrence W. Sherman the door to considering the implications of basic research provided by an earlier generation of thinkers. The most important basic research had documented the existence of police discretion to choose different strategies rather than being handcuffed by the three Rs. Most notably, Reiss’s (1971) conception of police work as divided between proactive and reactive discretion stimulated much creative thinking. At the same time, Banton’s (1964) research in Britain and Wilson’s (1968) research in the United States showed how the discretionary choices of police organizations differ on the basis of their political and cultural environments, despite their sharing a common legal system. These insights expanded the concept of police discretion from the level of case-by-case to agency-by-agency decision making and helped to set the stage for the emergence of triple-T. But that new, evidence-based strategy took almost four decades to emerge from the accumulation of new research evidence. What happened was a process of “prosumption” (Tapscott and Williams 2006) in which the producers of elements of the triple-T strategy were simultaneously its consumers. B. Testing The three-Rs strategy was intellectually discredited after three major efforts were made to assess the effects of key elements. These studies did not adequately support the conclusions attributed to them. They did not show that random patrol had no effect on crime, that rapid police response had no deterrent effect, or that detectives made no contribution to detection, prosecution, and general deterrence of crime. Regardless of what the research evidence really did show, those conclusions became widely accepted by police leaders and police scholars, then and now. 1. Random Patrol. The Police Foundation’s Kansas City Preventive Patrol Experiment (Kelling et al. 1974) launched the rise of evidencebased policing. The first attempt to undertake a scientifically controlled test of the effects of patrol staffing levels, it proved that bold experiments were possible in policing. The experimental design of withdrawing patrols from five patrol beats (and doubling it in five others) was stunning and unprecedented. That the “sky did not fall” made the world safe for further bold experiments. The reported conclusion— that frequency of police patrols did not affect crime—opened many minds to think more critically about police strategy. The experiment’s Rise of Evidence-Based Policing 387 leadership by a police chief who was subsequently appointed FBI director also suggested that research, even with negative results, could be good for police career advancement. It is unfortunate that the Kansas City experiment did not actually support its conclusion. The limitations of the experiment’s research design have been extensively reviewed elsewhere (Sherman 1986, 1992b; Sherman and Weisburd 1995). The main problem is that the actual frequency of patrols was unmeasured and may well have been identical in all beats because of reactive responses to calls (Larson 1976). Moreover, there were large differences across treatment groups in certain crime rates but too few beats to call the differences “statistically significant.” But the most important critique has never been published before: that random patrol was not compared to any other pattern of patrol, so there was no evidence showing that random patrol did not “work” relative to other patterns of patrol, such as concentration on crime hot spots. Policing still needs a strong randomized trial comparing random patrol to hot spots patrol. Nonetheless, the conclusion that random patrol did not work was widely accepted. Tens of thousands of police officers lost their jobs in the aftermath of the study, which came coincidentally just before a financial crisis in many US cities. None of that stopped the widespread use of random patrol. But it did help drive a research agenda seeking alternative police strategies. 2. Rapid Response. The theory that marginally faster response times would catch and deter more criminals was effectively falsified by a National Institute of Justice–funded research project led by staff of the Kansas City Police Department (1977). The study reported that it was necessary to divide crimes into victim-offender “involvement” crimes (e.g., robbery, assault, rape) and after-the-crime “discovery” crimes (e.g., burglary, car theft). It then focused response time analysis on involvement crimes, with “response time” including three time periods: crime occurrence to calling the police (“reporting time”), police receipt of call to dispatch (“dispatch time”), and “travel time” of police from receipt of dispatch to arrival at the scene. Using systematic observation methods and interviews of victims, the Kansas City study found that there was no correlation between response-related arrest probability and reporting time once the reporting time exceeded 9 minutes. The average reporting time for involvement crimes was 41 minutes (Kansas City Police Department 1977, vol. 2, pp. 23, 39). Replications of the 388 Lawrence W. Sherman reporting time segment in other US cities found similar results (Spelman and Brown 1981). What the research did not show is that a capacity for rapid response had no effect on crime. The Kansas City study was not an experiment comparing police agencies or areas with very rapid responses to similar agencies or areas with slower responses, or even no response. Much remains to be learned about the effect of any rapid response capacity compared to none, the latter being the state of police services in much of the world, from India to parts of rural America. What mattered historically was the growing recognition that there were better things police could do with their time than just wait for calls. While the research did not prove that point, it helped open the minds of police leaders and scholars about what those other things