Ideas in American Policing
By Lawrence W. Sherman
Ideas in American Policing presents commentary and insight from leading criminologists on issues of interest to scholars, practitioners, and policymakers. The papers published in this series are from the Police Foundation lecture series of the same name. Points of view in this document are those of the author and do not necessarily represent the official position of the Police Foundation.
©1998 Police Foundation and Lawrence W. Sherman. All rights reserved.
Lawrence W. Sherman is professor and chair of the Department of Criminology and Criminal Justice at the University of Maryland at College Park. He was the Police Foundation’s director of research from 1979 to 1985.
The new paradigm of “evidence-based medicine” holds important implications for policing. It suggests that just doing research is not enough and that proactive efforts are required to push accumulated research evidence into practice through national and community guidelines. These guidelines can then focus in- house evaluations of what works best across agencies, units, victims, and officers. Statistical adjustments for the risk factors shaping crime can provide fair comparisons across police units, including national rankings of police agencies by their crime prevention effectiveness. The example of domestic violence, for which accumulated National Institute of Justice research could lead to evidence-based guidelines, illustrates the way in which agency-based outcomes research could further reduce violence against victims. National pressure to adopt this paradigm could come from agency-ranking studies, but police agency capacity to adopt it will require new data systems creating “medical charts” for crime victims, annual audits of crime reporting systems, and in-house “evidence cops” who document the ongoing patterns and effects of police practices in light of published and in-house research. These analyses can then be integrated into the NYPD Compstat feedback model for management accountability and continuous quality improvement.
Most of us have thought of the statistician’s work as that of measuring and predicting . . . but few of us have thought it the statistician’s duty to try to bring about changes in the things that he [or she] measures.
—W. Edwards Deming
—— 2 ——
Of all the ideas in policing, one stands out as the most powerful force for change: police practices should be based on scientific evidence about what works best. Early in this century, Berkeley Police Chief August Vollmer’s partnership with his local university helped generate this idea (Carte and Carte 1975), which was clearly derived from that era’s expansion of the scientific method into medicine, management, agriculture, and many other fields (Cheit 1975). While science had greater initial impact in those other professions during the first half of the century, policing in recent decades has been moving rapidly to catch up. However, any assessment of this idea in modern policing must begin with an accurate benchmark: catching up to what? More complete evidence on the linkage between research and practice suggests a new paradigm for police improvement and for public safety in general: evidence-based crime prevention.
For years, Sherman (1984, 1992) and others have used medicine as the exemplar of a profession based upon strong scientific evidence. Sherman has praised medicine as a field in which practitioners have advanced training in the scientific method and keep up-to-date with the most recent research evidence by reading medical journals. He has cited the large body of randomized controlled experiments in medicine—now estimated to number almost one
million in print (Sackett and Rosenberg 1995)—as the highly rigorous scientific evidence used to guide medical practices. He has suggested that policing should therefore be more like medicine.
Sherman was right about the need for many more randomized experiments in policing, but wrong about how much medicine was really based on scientific research. New evidence shows that doctors resist changing practices based on new research just as much as police do, if not more so. Closer examination reveals medicine to be a battleground between research and practice, with useful lessons for policing on new ways to promote research. Those lessons come from a new strategy called “evidence-based medicine,”1
“widely hailed as the long-sought link between research and practice” (Zuger 1997) to solve problems like the following (Millenson 1997, 4, 122, 131):
• An estimated 85 percent of medical practices remain untested by research evidence.
• Most doctors rarely read the 2,500 medical journals available, and instead base their practice on local custom.
• Most studies that do guide practice use weak, non- randomized research designs.
Medicine, in fact, seems just as resistant to the use of evidence to guide practice as are fields with lower educational requirements, such as policing. The National Institutes of Health (NIH) Consensus Guidelines are a case in point. NIH convenes advisory boards to issue to physicians recommendations that are based on intensive reviews of research evidence on specific medical practices. These recommendations usually receive extensive publicity, and are reinforced by mailings of the guideline summaries to some one hundred thousand doctors. But according to a RAND evaluation, doctors rarely change their practices in response to publication of these guidelines (Kosecoff et al. 1987, as cited in Millenson 1997). Thus three years after research found that heart attack patients treated with calcium antagonists were more likely to die, doctors still prescribed this dangerous drug to one-third of heart attack patients. Eight years after antibiotics were shown to cure ulcers, 90 percent of ulcer patients remained untreated by antibiotics (Millenson 1997, 123–25).
Evidence Cops The struggle to change
medical practice based on research evidence has a long history, with valuable implications for policing. In the 1840s, Ignaz Semmelweiss found evidence that maternal death in childbirth could be reduced if doctors
1 The term “evidence” in this mono- graph refers to scientific, not criminal, evidence.
—— 3 ——
washed their hands before delivering babies. He then tried to apply this research to medical practice in Vienna, which led to his being driven out of town by his boss, the chief obstetrician. Hundreds of thousands of women died because the profession refused to comply with his evidence-based guidelines for some forty years. The story shows the important distinction between merely doing research and attempting to apply research to redirect professional practices.
One way to describe people who try to apply research is the role of “evidence cop.” More like a traffic cop than Victor Hugo’s detective Javert, the evidence cop’s job is to redirect practice through compliance rather than punishment. While this job may be as challenging as herding cats, it still consists of pointing professionals to practice “this way, not that way.” As in all policing, the success rate for this job varies widely. Fortunately, the initial failures of people like Semmelweiss paved the way for greater success in the 1990s.
Consider Scott Wein- garten, M.D., of Cedars-Sinai
Hospital in Los Angeles. As director of the hospital’s Center for Applied Health Services Research, Weingarten is an evidence-cop-in-residence. His job is to monitor what the 2,250 doctors are doing to patients at the hospital and to detect practices that run counter to recommendations based on research evidence. He does this through prodding rather than punishment, convening groups of doctors who treat specific maladies to discuss the research evidence. These groups then produce their own consensus guidelines for practices that become hospital policy. Thirty- five such sets of guidelines were produced in Weingarten’s first four years on the job (Millenson 1997, 120).
What NIH, Weingarten, and the 1995 founders of the new journal called Evidence-Based Medicine are all trying to do is to push research into practice. Just as policing has become more proactive at dealing with crime, researchers are becoming more proactive about dealing with practice. This trend has developed in many fields, not just medicine.
Increased pressure for “reinventing government” to focus on measurable results is reflected in the 1994 U.S. Government Performance Results Act (GPRA), which requires all federal agencies to file annual reports on quantitative indicators of their achievements. Education is under growing pressure to raise test scores as proof that children are learning, which has led to increased discussion of research evidence on what works in education (Raspberry 1998). And the U.S. Congress has required that the effectiveness of federally funded crime prevention programs be evaluated using “rigorous and scientifically recognized standards and methodologies” (House 1995, sec. 116). All this sets the stage for a new paradigm for making research more useful to policing than it has ever been before.
Key Questions In suggesting a new paradigm
called evidence-based policing, there are four key questions to answer: What is it? What is new about it? How does it apply to a specific example of police practice? How can it be institutionalized?
What is it? Evidence-based policing is the
use of the best available research on the outcomes of police work to implement guidelines and evaluate agencies, units, and officers. Put more simply,
One way to describe people who try to apply research is the role of “evidence cop.”
—— 4 ——
evidence-based policing uses research to guide practice and evaluate practitioners. It uses the best evidence to shape the best practice. It is a systematic effort to parse out and codify unsystematic “experience” as the basis for police work, refining it by ongoing systematic testing of hypotheses.
Evaluation of ongoing operations has been the crucial missing link in many recent attempts to improve policing. If it is true that most police work has yet to go “beyond 911” (Sparrow, Moore, and Kennedy 1990), the underlying reason may be a lack of evaluation systems that clearly link research-based guidelines to outcomes. It is only with that addition that policing can become a “reflexive” or “smart” institution, continuously improving with ongoing feedback.
The basic premise of evidence-based practice is that we are all entitled to our own opinions, but not to our own facts. Yet left alone to practice individually, practitioners do come up with their own “facts,” which often turn out to be wrong. A recent survey of 82 Washington State doctors found 137 different strategies for treating urinary tract infections (Berg 1991). No doubt the same result could be found for handling domestic disturbances. A study evaluating the accuracy of strep throat diagnoses based on unstructured examination by experienced pediatricians found it
far inferior to a systematic, evidence-based checklist used by nurses. The mythic power of subjective and unstructured wisdom holds back every field and keeps it from systematically discovering and implementing what works best in repeated tasks.
A prime example of the power of systematic, ongoing evaluations comes again from medicine. In 1990, the New York State Health Department began to publish death rates for coronary bypass surgery grouped by hospital and individual surgeon. This action was prompted by research showing that while the statewide average death rate was 3.7 percent, some doctors ran as high as 82 percent. Moreover, after adjusting for the risk of death by the pre-operation condition of the patient caseload, patients were 4.4 times more likely to die in surgery at the least successful hospitals than at the best hospitals. Despite enormous opposition from hospitals and surgeons, these data were made
public, revealing a strong practice effect: the more operations doctors and hospitals did each year, the lower the risk-adjusted death rate. Using this clear correlation to push low-frequency surgeons and hospitals out of this business altogether, hospitals were able to lower the death rate in these operations by 40 percent in just three years (Millenson 1997, 195).
Evidence-based policing is about two very different kinds of research: basic research on what works best when implemented properly under controlled conditions, and ongoing outcomes research about the results each unit is actually achieving by applying (or ignoring) basic research in practice. This combination creates a feedback loop (fig. 1) that begins with either published or in-house studies suggesting how policing might obtain the best effects. The review of this evidence can lead to guidelines taking law, ethics, and community
Figure 1. Evidence-Based Policing.
—— 5 ——
culture into account. These guidelines would specify measurable “outputs,” or practices that police are asked to follow. Their varying degrees of success at delivering those outputs can then be assessed by tracking risk-adjusted “outcomes,” or results over a reasonably long follow-up period. These outcomes may be defined in several different ways: offenses per 1,000 residents, repeat victimizations per 100 victims, repeat offending per 100 offenders, and so on. The observation that some units are getting better results than others can be used to further identify factors associated with success, which can then be fed back as new in-house research to refine the guidelines and raise the overall success level of the agency. Such research could also be published in national journals or at least kept in an agency database as institutional memory about success and failure rates for different methods.
What is new about it? Skeptics may say that there is
nothing new in evidence-based policing, and that other paradigms already embrace these principles. On closer examination, however, we will see that no other paradigm contains the principles for its own implementation. No other paradigm contains a principle for both changing practices and measuring the success of those changes with risk-adjusted
outcomes research (like bypass surgery death rates). No other paradigm—not even NYPD’s Computerized Crime Comparison Statistics (Compstat) strategy (Bratton with Knobler 1998)— uses scientific evidence to hold professionals accountable for results in peer-reviewed and even public discussions of outcomes evidence.
Evidence-based policing is clearly different from, but very helpful to, all three present paradigms of policing. Incident- specific policing, or 911 responses, currently lack any outcomes measure except time out of service. Police officers who take too much time to handle a call are sometimes accused of shirking and are urged by supervisors to work faster.2 But no one tracks the rate of repeat calls by officer or unit to see how effective the first response was in preventing future problems. Evidence-based policing could
use such outcomes to justify longer time spent on each call on the basis of an officer’s average results, rather than issuing a crude demand that he or she stay within an average time limit. It could also place much more emphasis on learning how to deal with each call most effectively and preventively, a question that currently gets little attention.
Community policing, however defined, is not clearly linked to evidence about effectiveness in preventing crime. It is much more about how to do police work—a set of outputs— than it is about desired results, or outcomes. Working with the community and listening to and respecting community members are all important elements of the paradigm. But that paradigm alone has been easy for many officers to ignore. Adding the accountability systems from the paradigm of evidence-based policing could actually make police far more active in working with the community.
Problem-oriented policing is clearly the major source for
2 This sounds oddly like the pressure for drive-in, drive-out childbirth health insurance now barred by federal law.
Evidence-based policing is clearly different from, but very helpful to, all three present paradigms of policing.
—— 6 ——
evidence-based policing. Herman Goldstein’s writings (1979, 1990), as well as John Eck and William Spelman’s SARA model (1987), clearly emphasize assessment of problem-solving responses as a key part of the process. Yet there is no clear statement about the use of scientific evidence either in selecting strategies for responding to problems or in monitoring the implementation and results of those strategies (Sherman 1991). Reports on problem-oriented policing have so far produced little evidence either from controlled tests or outcomes research. Because the paradigm stresses the unique characteristics of each crime pattern, problem- oriented policing has not been used to respond to highly repetitive situations like domestic assaults or disputes. Few comparisons of different methods for attacking the same problem have been developed. Few officers are even held accountable for not implementing a problem-solving plan they have agreed to undertake. Problem-oriented policing has clearly revolutionized the way many police think about their objectives, moving them away from a narrow focus on each incident to a broader focus on patterns and systems. But in the absence of pressure from an evidence-based approach to evaluating success and management accountability, problem-oriented policing has been kept at the margins of police work.
NYPD’s Compstat strategy (Bratton with Knobler 1998) has pushed the results accountability principle farther than ever before, but it has not used the scientific method to assess cause and effect. Successful managers are rewarded, but successful methods are not pinpointed and codified.
What evidence-based policing adds to these paradigms is a new principle for decision making: scientific evidence. Most police practice, like medical practice, is still shaped by local custom, opinions, theories, and subjective impressions. Evidence-based policing challenges those principles of decision making and creates systematic feedback to provide continuous quality improvement in the achievement of police objectives (see Hoover 1996). Hence the inspiration for this paradigm is not only medicine and its randomized trials, but also the principles of quality control in manufacturing developed by Walter Shewhart (1939) and W. Edwards Deming (1986). These principles were initially rejected by U.S. business leaders, but were finally embraced in the 1980s after Japanese industries used them to far surpass U.S. manufacturers in the quality of their products.
What makes both policing and medicine different from manufacturing, of course, is the far greater variability in the raw material to be processed—human beings. That is what gives the gold standard of evaluation research, the randomized
controlled trial, both its strength and its limitations. The strength of the research design, pioneered in policing by the Police Foundation, is its ability to reduce uncertainty about the average effects of a policy on vast numbers of people. The limitation of the research design is that it cannot escape variability in treatments, responses, and implementation.
The variability of treatments in policing is much like that in surgery, which stands in sharp contrast to pharmaceuticals. While the chemical content of medical drugs is almost always identical, the procedural content of surgery varies widely. Similarly, the style and tone each officer brings to a citizen encounter varies enormously and can make a big difference in the outcome of a specific case. Dosage, timing, and follow-up of both drugs and police work can vary widely in practice.
Even holding treatment constant, there is evidence that both patients and offenders respond to treatments with wide variations. Some of these responses, allergic reactions, can kill some people with treatments that cure most others. Offenders are known to vary in their responses to police actions by individual, neighborhood, and city. And implementation of new practices based on controlled experiments in both medicine and policing varies according to how well research is communicated, how much information is created
—— 7 ——
about whether practices actually change, and how much reinforcement there is for the change, both positive and negative.
Evidence-based policing assumes that experiments alone are not enough. Putting research into practice requires just as much attention to implementation as it does to controlled evaluations. Ongoing systems for researching implementation can close the feedback loop to create the principle of industrial quality improvement.
How does it apply to a specific example of police practice?
The policing of domestic violence offers a clear illustration of what is new about the evidence-based paradigm. Domestic violence has been the subject of more police practices research than any other crime problem. The research has arguably had little effect on police practice, at least by the new standards of evidence-based medicine. Yet the available evidence offers a fair and scientifically valid approach for holding police agencies, units, and officers accountable for the results of police work, as measured by repeated domestic violence against the same victims.
The National Institute of Justice (NIJ) and the Police Foundation have provided policing with extensive information on what works to prevent repeated violence. The
research has also shown that, like surgery, police practices vary greatly in their implementation. These variations in practice cause varying results for repeat offending against victims. Even holding practice constant, responses to arrest vary by offender, neighborhood, and city. Finally, research shows very poor compliance with mandatory arrest guidelines after they are adopted (Ferraro 1989).
There are many varieties of arrest for misdemeanor domestic violence. The offender may or may not be handcuffed, arrested in front of family and neighbors, given a chance to explain his version of events to the police, or
treated with courtesy and politeness. Do these variations on the theme of arrest make a difference? They should, according to the “defiance” theory of criminal sanction effects (Sherman 1993). And they did in Milwaukee, according to Raymond Paternoster and his colleagues (1997). The Milwaukee evidence reveals that controlling for other risk factors among some 800 arrested offenders, those who felt they were not treated in a procedurally fair and polite manner were 60 percent more likely to commit a reported act of domestic violence in the future (fig. 2). This finding suggests three ways
Figure 2. Repeat Domestic Violence and Police Fairness.
Source: Paternoster, et al.
—— 8 ——
to push research into practice: 1) change the guidelines for making domestic violence arrests to include those elements that would enable offenders to perceive more “procedural justice”; 2) hold police accountable for using these guidelines by comparing rates of repeat victimization associated with different police units; and 3) compute these rates using statistical adjustments for the pre- existing level of recidivism risks.
The NIJ research provides other evidence for ways that police can reduce repeat offending in misdemeanor domestic violence. Rather than a one-size-fits-all policy, the evidence suggests specific guide- lines to be used under different conditions. Offenders who are absent when police arrive—as they are in some 40 percent of cases—respond more effectively to arrest warrants than offenders who are arrested on the scene (Dunford 1990). Offenders who are employed are deterred by arrest, while offenders who are unemployed generally increase their offending more if they are arrested than if they are handled in some other fashion (Pate and Hamilton 1992; Berk et al. 1992; Sherman and Smith 1992). Offenders who live in urban areas of concentrated poverty commit more repeat offenses if they are arrested than if not, while offenders who live in more affluent areas commit fewer repeat offenses if they are arrested (Marciniak 1994). All of these
findings could be changed by further research, but for the moment they are the best evidence available.
This research evidence could support guidelines for policing domestic violence that differed by neighborhood and absence or presence of the offender. It could also support guidelines about listening to suspects’ side of the story before making arrest decisions and generally treating suspects with courtesy. Other evidence, such as the extremely high-risk period for repeat victimization in the first days and weeks after the last police encounter (Strang and Sherman
1996), could be used to fashion new problem-oriented strategies. Most important, the existing research can be used to create a fair system for evaluating police performance on the basis of risk- adjusted outcomes. That evidence (fig. 3) shows that the likelihood of a repeat offense is strongly linked to the number of previous offenses each offender has.
Once the risk of repeat offending can be predicted with reasonable accuracy, it becomes possible to use those predictions as a benchmark for police performance. Just as in the bypass surgery death rates in New York, the outcomes of policing can be
Figure 3. Risk of Repeat Domestic Assault by Priors.
Milwaukee Domestic Violence Experiment
0 1 2 3 7
—— 9 ——
controlled for the risk level inherent in the caseload they face. Using a citywide database of all domestic assaults, now running over ten thousand cases per year in cities like Milwaukee, a model can be constructed to assess the risk of repeat offending in each case. The overall mix of cases in each police precinct or for each officer can generate an average risk level for that caseload. Each police patrol district can then be evaluated according to the actual versus predicted rate of repeat offending each year (fig. 4). All patrol districts in the city can then be compared on the basis of their relative percentage difference between expected and actual rates of repeat domestic assault (fig. 5).
By constructing information systems for this kind of outcome research, police departments can focus on an objective that has only previously been measured in major experiments. Making the goal of policing each domestic assault the outcome of a reduced repeat offending rate rather than the output of whether an arrest is made would have several effects. One is that crime prevention would get greater attention than retribution for its own sake. While not everyone would welcome that, it is consistent with at least some police leaders’ view of the purpose of the police as a crime prevention agency (Bratton with Knobler 1998). Another effect would be to seek out and
Figure 4. Observed vs. Expected Risk of Repeat Domestic Violence.
PCT 1 PCT 2 PCT 3 PCT 4 PCT 5
Figure 5. Observed vs. Expected Ranking by Precinct.
—— 10 ——
even initiate more research on what works best to prevent domestic violence. In the world as we now know it, no one in policing—from the police chief to the rookie officer—has any direct incentive to reduce repeat offending against known victims. No one in policing is held accountable for accomplishing, or even measuring, that objective. As a result, no one knows whether repeat victimization rates get better or worse from year to year. Using outcomes evidence to evaluate performance would make police practices far more victim- centered, the top priority being that of preventing any further assaults.
How can it be institutionalized?
The strongest claim about evidence-based policing is that it contains the principles of its own implementation. The principles of using evidence both to change and evaluate practice can be applied to a broad institutional analysis of implementation. Thus while the changes described
above would have to occur one police agency at a time, there are certain national forces that can help start the ball rolling. This can be seen, for example, in national rankings of big-city police agencies, as well as national mandates for improving police data systems to provide better evidence. Yet even such external pressures will not succeed without internal evidence cops to import, apply, and create research evidence.
No institution is likely to increase voluntarily its accountability except under strong external pressure. It is unlikely that evidence-based policing could be adopted by …