Journal of Experimental Criminology

, Volume 7, Issue 2, pp 149–181

A randomized controlled trial of different policing strategies at hot spots of violent crime

Authors

    • NORC at the University of Chicago
  • Christopher S. Koper
    • Police Executive Research Forum
  • Daniel J. Woods
    • Police Executive Research Forum
Article

DOI: 10.1007/s11292-010-9120-6

Cite this article as:
Taylor, B., Koper, C.S. & Woods, D.J. J Exp Criminol (2011) 7: 149. doi:10.1007/s11292-010-9120-6

Abstract

Focusing police efforts on “hot spots” has gained acceptance among researchers and practitioners. However, little rigorous evidence exists on the comparative effectiveness of different hot spots strategies. To address this gap, we randomly assigned 83 hot spots of violence in Jacksonville, Florida, to receive either a problem-oriented policing (POP) strategy, directed-saturation patrol, or a control condition for 90 days. We then examined crime in these areas during the intervention period and a 90-day post-intervention period. In sum, the use of POP was associated with a 33% reduction in “street violence” during the 90 days following the intervention. While not statistically significant, we also observed that POP was associated with other non-trivial reductions in violence and property crime during the post-intervention period. In contrast, we did not detect statistically significant crime reductions for the directed-saturation patrol group, though there were non-significant declines in crime in these areas during the intervention period. Tests for displacement or a diffusion of benefits provided indications that violence was displaced to areas near the POP locations, though some patterns in the data suggest this may have been due to the effects of POP on crime reporting by citizens in nearby areas. We conclude by discussing the study’s limitations and the implications of the findings for efforts to refine hot spots policing.

Keywords

Problem-oriented policingViolent crimeRandomized experimentHot spots

Introduction

Police interventions focused on “hot spots”—small geographic places or areas where crime is concentrated—have gained widespread acceptance among practitioners and researchers as an effective approach to reducing crime. The use of crime mapping to identify hot spots is now common among police agencies (Weisburd and Lum 2005; Hickman and Reaves 2006: 30), and police cite hot spots enforcement as a leading approach to the reduction of violence and other crime problems (Police Executive Research Forum 2008). There is also substantial consensus among researchers that hot spots policing can be effective in reducing crime and disorder (National Research Council 2004).

However, the body of research on hot spots policing is still relatively small, and it is not yet sufficient to demonstrate what types of strategies work best for hot spots generally or for particular types of hot spots as defined by types of crime problems or other features. To explore these issues and add to the growing body of work on hot spots policing, this paper presents the results of a randomized experiment conducted in Jacksonville, Florida, to test and contrast the effects of two commonly used hot spots strategies, directed patrol and problem-oriented policing (POP), at hot spots of violent crime.

Hot spots and hot spots policing

Although there is no universal definition of the term “hot spot” (Eck 2005), it is typically used by researchers, as we do here, to refer to specific addresses, intersections, street blocks, or clusters of blocks where crime is concentrated (e.g., Sherman, Buerger et al. 1989; Mastrofski et al. 2009; Weisburd 2008).1 Studies in several cities, for example, have shown that approximately half of crime occurs at 5% or less of a city’s addresses and intersections (e.g., Pierce et al. 1988; Sherman, Gartin and Buerger 1989; Weisburd et al. 2004). Further, the concentration of crime at these places tends to be stable over time (Weisburd et al. 2004; also see Braga et al. 2010; Weisburd et al. 2009).

Hot spots are often nodes for business, leisure, and/or travel activities, and they have features or facilities that create criminal opportunities and facilitate offending (Eck and Weisburd 1995). In the language of routine activities theory (Cohen and Felson 1979), they are places that bring together motivated offenders, suitable targets, and an absence of capable guardians. Examples include locations with bars, convenience stores, parks, bus depots, apartment buildings, parking lots, shopping centers, motels or hotels, adult businesses, and the like (e.g., Sherman, Buerger et al. 1989: 45; Braga et al. 1999: 551–552; also see Eck and Weisburd 1995).

Focusing on hot spots has the potential to make police more effective and efficient in a number of ways. For one, it concentrates their attention on the places where crime is most likely to occur. Police can also arguably establish a more visible presence and generate greater perceptual effects in the small space of a hot spot than over larger areas like a patrol beat or an entire jurisdiction (Sherman and Weisburd 1995). Further, officers can try to address the underlying conditions that contribute to crime and disorder at these places through problem-solving efforts that may include enforcement and investigative operations as well as prevention measures implemented in cooperation with place managers (Eck 1994) and other stakeholders (such as business owners and managers, residents, and other government agencies) with interests and/or responsibility for the area.

Several empirical studies have also demonstrated the effectiveness of hot spots policing. Braga’s (2007) review of nine experimental and quasi-experimental studies of enforcement-oriented hot spot interventions (including a variety of patrol, order maintenance, crackdown, and/or problem-solving activities) found that these efforts reduced at least some form(s) of crime or disorder in seven of the nine cases. Further, in five studies that addressed the issue, there were no obvious or consistent signs that crime was displaced to nearby areas; on the contrary, in three of these cases, there was evidence that crime reduction benefits extended to areas outside the hot spot—i.e., a diffusion of crime control benefits (Clarke and Weisburd 1994). The National Research Council (NRC) of the National Academy of Sciences has also provided a positive assessment of hot spots policing, stating that the research on hot spots constitutes the “…strongest collective evidence of police effectiveness that is now available” (National Research Council 2004: 250; also see Weisburd and Eck 2004).2

One caveat to this conclusion is that the body of evidence on hot spots policing is still modest. For example, the number of randomized trials testing hot spots policing (eight, according to a recent compilation by Lum et al. 2010) is small compared to that in other areas such as corrections research. Another caveat is that hot spots strategies may not be optimal for all types of crime problems. Studies of hot spots policing have examined effects on a variety of crimes, including violent, property, drug, and disorder offenses, with varying results. However, Braga’s (2007) review, which included a meta-analysis of results from five randomized experiments, suggests that the effects of hot spots policing are most pronounced on disorderly behaviors; although violent and property crimes declined on average across the studies, these effects were not statistically significant overall (36).

Impacts on violent crime, which is the chief concern of our study, tended to be the smallest in Braga’s meta-analysis. This could reflect the impulsive, expressive nature of many violent crimes (which may make them harder to prevent) and the rarity of violent crime in very small locations. On the other hand, the particularly high concentration of violence in a relatively small number of places would seem to weigh in favor of using hot spots strategies to curb violence. Sherman et al.’s study in Minneapolis, for example, revealed that all robberies occurred at 2.2% of the city’s places over a 1-year period and that all assaults occurred at 7% (1989: 43). Similarly, all gun assaults in Boston over a 29-year period occurred in just 11.5% of the city’s street segments, and about three-quarters occurred in 4.8% (Braga et al. 2010: 41). Further, the relative concentration of violence at small places—i.e., the relative difference between the number of addresses and places where violence could occur and the number where it does occur—is greater for violent crime (particularly robbery, assault, and domestic disturbances) than for property crime (Sherman et al., 1989a).3 Displacement from hot spot approaches may also be lessened when addressing violent crime, given that violent offenders tend to commit their offenses closer to home (e.g., see reviews in Eck and Weisburd 1995 and Gabor and Gottheil 1984), thus making spatial displacement less likely, and that substitution of targets (i.e., victims) seems less likely for assaultive violence. Moreover, research discussed below suggests that hot spots interventions are more successful in reducing serious offending, including violence, when focused on places chosen specifically for their levels of serious crime and when officers are focused on the alleviation of those particular problems (Braga et al. 1999).

Assessing the comparative effectiveness of hot spots strategies

Another key issue for advancing hot spots policing is determining what hot spots approaches are most effective. Prior studies of hot spots policing have examined a variety of strategies including directed patrol or fixed presence (e.g., Sherman and Weisburd 1995; Lawton et al. 2005); order maintenance and drug enforcement crackdowns (Braga and Bond 2008; Braga et al. 1999; Weisburd and Green 1995); raids on crack houses (Sherman and Rogan 1995); and other forms of problem-solving that have included situational crime prevention, nuisance abatement, clean-up activities, improvement of social services, and other measures (e.g., Braga and Bond 2008; Braga et al. 1999; Eck and Wartell 1998; Mazerolle et al. 2000; Sherman, Gartin and Buerger 1989). There has been little comparative assessment of these approaches. Moreover, these strategies have often been combined in hot spots evaluations, further complicating this task. Consequently, as others have noted (Braga 2007; National Research Council 2004), there is not yet sufficient evidence to determine what types of police approaches are most optimal at hot spots.

To date, only one study has attempted to compare the effects of different types of police activities at hot spots. In that study, Braga and Bond (2008) evaluated a Lowell, Massachusetts, project in which police implemented a problem-oriented policing (POP) strategy at 17 randomly selected hot spots over a 1-year period.4 The authors broadly categorized the police actions as a “policing disorder” strategy that involved a mix of situational crime prevention measures (e.g., improving lighting, razing abandoned buildings, etc.), social service activities (e.g., connecting people with mental health services and improving opportunities for youth recreation), and order maintenance enforcement (585). Overall, the program reduced calls-for-service at the target hot spots by 20% relative to changes in a control group of hot spots. Reductions in serious crimes were even greater—around one-third for burglary and non-domestic assault and over 40% for robbery (592). Moreover, a post-hoc statistical mediation analysis suggested that these effects stemmed primarily from the situational crime prevention measures (594-595); the misdemeanor arrests generated effects of marginal statistical significance, and the social service strategies produced non-significant results.5

Generalizing from these results, they would seem to suggest that problem-solving, preventive strategies (particularly those linked to situational crime prevention) are more effective at hot spots than are enforcement-oriented strategies.6 In this study, we attempt to test this issue more directly. Our study extends and complements the Lowell study and earlier hot spots research through a randomized experimental evaluation that compares the effectiveness of two general police strategies—directed patrol and problem-oriented policing (POP)—relative to both one another and a control condition at hot spots of violent crime. To our knowledge, this is the first study to compare the outcomes of different hot spots strategies through experimental methods. In the next subsections, we briefly review prior research on directed patrol and POP, particularly as it pertains to hot spots policing and violence.

Directed patrol at hot spots

Directed patrol, which involves assigning officers to intensively patrol particular areas at particular times (while often freeing them from answering calls-for-service), is perhaps one of the most basic and common strategies that agencies use at hot spots (Koper 2008; Police Executive Research Forum 2008).7 In the first experimental study of hot spots policing, Sherman and Weisburd (1995) evaluated the impact of directed patrol at 55 randomly selected hot spots in Minneapolis. During the course of the 1-year study, officers concentrated patrol dosage on the experimental hot spots during high-risk times of the day and night, spending as much time as possible at these locations when not answering calls-for-service. Hence, the treatment consisted of “intensified but intermittent” police presence (1995: 634). The project’s emphasis was on increasing police presence at hot spots; what the officers did at these locations, if anything, was left to them and appeared to vary widely according to anecdotal accounts. Despite a breakdown in the study protocol during 2 months, officers generally spent 2–3 h per day at the experimental locations. For most of the year, this raised police presence in the experimental spots to levels that were two to three times as high as those in a group of 55 control hot spots that received standard patrol (1995: 639).

An analysis of the experiment’s effects indicated that the intensified patrols reduced total calls to the police by 6 to 13% relative to changes in the control locations (results were sensitive to the inclusion of periods during which there were problems with treatment fidelity or measurement of outcomes) (1995: 641–643). These results were driven largely by the intervention’s impact on “soft” crimes (e.g., disturbances, drunken behavior, break-in alarms, and vandalism). Effects on “hard” crimes (e.g., shootings, stabbings, rapes, other assaults, robberies, and thefts of and from automobiles) were generally non-significant and smaller, though in the expected direction.

Another illustration of a patrol-only intervention at hot spots was Philadelphia’s Safe Streets program, in which pairs of officers were assigned to 24-h, 7-days-a-week fixed presence at 214 of the city’s most problematic drug addresses and intersections (Lawton et al. 2005). As in the Minneapolis study, this program simply emphasized increased police presence at the selected locations. In an evaluation of the program’s first 18 weeks, Lawton et al. (2005) compared before and after changes in crime at the intervention locations to those in 73 matched non-intervention locations. They found that both violent and drug crimes declined in the intervention sites without changing in the comparison sites. Focusing on the former results, the program appeared to prevent one violent crime per week in the intervention sites, while also reducing violence by lesser amounts in adjoining areas of 0.1 to 0.2 miles surrounding the target locations (440).8

Problem-oriented policing at hot spots

The POP model of policing, first articulated by Herman Goldstein (1979; 1990), calls for police to transcend reactive, incident-driven policing by studying and addressing underlying problems that contribute to crime and disorder in the community. Goldstein’s notion was for police to take proactive, preventive action against the causes of continuing crime and disorder issues. Further, Goldstein argued that police responses to these problems should not be limited to traditional law enforcement actions but, rather, should also include the use of civil law and reliance on other municipal and community resources. Eck and Spelman (1987) later developed the well-known SARA model for implementing POP, which consists of four steps: scanning for problems, analysis of problems, development and implementation of responses, and follow-up assessment of results. POP thus represents a process of identifying problems and developing responses rather than any specific type(s) of response.

POP has become a popular strategy that police use to tackle a wide variety of crime types. Observers have noted that in practice POP efforts often fall short of the POP ideal in that they involve limited analysis, limited community partnership efforts, and heavy reliance on enforcement tactics and other relatively easy situational crime prevention responses—what some have called “shallow” problem-solving (e.g., Braga and Weisburd 2006; Braga and Bond 2008; Cordner and Bielbel 2005; Eck 2006). Implementation problems have also complicated the assessment of some POP projects (e.g., Sherman, Gartin and Buerger 1989; Weisburd and Green 1995). Yet despite these limitations, POP has generally received favorable evaluations (National Research Council 2004). A recent meta-analysis of ten rigorous experimental and quasi-experimental studies found that the strategy is generally successful in reducing crime and disorder, though the effects tend to be modest overall (Weisburd et al. 2010).9

A number of studies have examined the application of POP to hot spots (Baker and Wolfer 2003; Braga et al. 1999; Braga and Bond 2008; Mazerolle et al. 2000; Sherman, Gartin and Buerger 1989; Weisburd and Green 1995), and most of these have shown that the efforts reduced some forms of crime and disorder (also see Braga 2007 and Weisburd et al. 2010).10 Results have been more mixed with respect to violent crime—interventions studied by Mazerolle et al. (2000), Sherman, Gartin and Buerger 1989, and Weisburd and Green (1995), for example, did not reduce violence—but only one rigorous study of POP and hot spots has focused specifically on hot spots of violent crime.

In a randomized experiment conducted in Jersey City, New Jersey, Braga et al. (1999) evaluated a project in which officers of a violent crimes unit conducted POP efforts at 12 violent crime hot spot intersection areas (i.e., problem intersections and their adjoining streets) over a 16-month period and compared them to 12 comparable control areas.11 Using the SARA framework, officers studied problems at the sites through analysis of data and discussions with community members and then implemented various strategies tailored to the unique problems of each place. Common strategies used included situational crime prevention measures, aggressive order maintenance enforcement, intensified drug enforcement, and clean-up activities (554). In general, officers placed substantial emphasis on the reduction of disorder at the locations. Analysis of incident reports and calls-for-service to police revealed consistent, though not always statistically significant, reductions in robbery and non-domestic assaults as well as in property crimes, drug offenses, and more minor criminal and disorderly behavior at the experimental hot spots relative to a control group (561–566). Most analyses also suggested that there was no significant crime displacement or diffusion of benefits to the two-block areas surrounding the treatment hot spots.

The project we describe below also tests the use of POP at violent crime hot spots. In this particular effort, however, officers were encouraged to deemphasize the use of enforcement responses and to rely more on situational crime prevention measures, civil law, and the leveraging of other government and community resources.

Methods

Research site

We conducted this study in the city of Jacksonville, Florida, with the Jacksonville Sheriff’s Office (JSO). JSO has about 1,400 sworn officers. Jacksonville is the largest city in Florida. Since 1968, as a result of the consolidation of the city and county government, Jacksonville is the largest city in land area in the contiguous United States (758 square miles). Consequently, the majority of Jacksonville's metropolitan population resides within the city limits, making it the most populous city proper in Florida and the twelfth most populous in the United States.12 Jacksonville is the principal city in the Greater Jacksonville Metropolitan Area, a region with a population of more than 1,313,228.13

Like many large cities, Jacksonville has a violent crime problem. The number of violent crimes in Jacksonville has gone up from 2003 to 2008. Homicides went from 95 in 2003 to 116 in 2008, with a high point of 125 in 2007. Forcible rapes went from 222 in 2003 to 276 in 2008. Robberies went from 2,412 in 2003 to 3,061 in 2008, with a high point of 3,245 in 2007. Aggravated assaults went from 4,271 in 2003 to 4,935 in 2008. With such a large number of violent crimes in Jacksonville, this made it an excellent site from a research perspective for a study on violence. At the time of the development of this project which started in late 2007, Jacksonville had been coined the “Murder Capitol of Florida.” In 2007, media attention around the increase in violence began to intensify, especially since the annual increase of homicides from 2005 (96) to 2006 (115) to 2007 (125) were the kinds of increases not seen since the crack epidemic years more than a decade ago. The increase in violent crime in the mid 2000s also signaled a major shift after years of declining violence in the Jacksonville area.

In recent years, JSO’s primary responses to violence had been to target high crime areas for overtime-funded directed patrol and to focus investigative and prosecutorial efforts on Jacksonville’s most problematic offenders. For this project, however, JSO experimented with a more geographically focused approach to violence reduction that involved concentrating patrol and problem-solving efforts on well-defined “micro” hot spots of violence.

Description of interventions

As discussed below, we took 83 violent hot spots and randomly assigned them to one of three conditions: 40 control hot spots, 21 saturation/directed patrol hot spots (we use this hybrid term to capture the fact that officers were directed to specific hot spots and that their extended presence at these small locations, which typically lasted for several hours at a time, amounted to a saturation of the areas), or 22 problem-oriented policing (POP) hot spots. Each of these three conditions was maintained for a 90-day period. The intervention period was short compared to that of many other hot spots experiments which have tested interventions lasting for several months or more than a year (e.g., see review in Braga 2007). With short-term interventions, there may be a greater likelihood that effects will decay quickly after the intervention ends (e.g., see Sherman and Rogan 1995). The short intervention period also raises concern about the ability of POP officers to diagnose problems and implement effective solutions within the given timeframe. Yet while the intervention period was short, the intensity of the intervention was high, particularly in the POP areas. As described below, POP officers conducted problem-solving activities full-time, 7 days a week and were able to complete many POP responses at each location.14 Further, our analysis examines changes in crime during the 90 days following the intervention to allow for the possibilities that the effects of POP would take more than 90 days to materialize and/or that the effects of either or both interventions would decay quickly.

Another caveat is that, given our focus on violent crime, we were only able to identify 83 hot spots for the experiment. As we discuss later in more detail in the discussion section, a sample size of 83 cases creates some limitations in our ability to detect small statistically significant differences across the study conditions. The practical implications of this is that we are left with some uncertainty regarding whether differences that do not reach statistical significance (i.e., p ≤.05) are actually different but are not showing up as statistically significant due to a modest sample size.

The control hot spots were provided with standard policing efforts involving traditional patrol operations, without the introduction of any additional resources. The saturation/directed patrol hot spots were provided with additional focused/precision policing to create a heightened police presence in specific locations with high concentrations of violent crime. JSO determined the patrol dosage and schedule (i.e., days of week and times of day) for each hot spot based on analysis of the location’s crime levels and high-risk times. During the selected days and times, pairs of officers in separate cars worked one to three hot spots at a time (officers assigned to multiple hot spots covered locations in close proximity).15 A sergeant was assigned to oversee the saturation effort at all times. Total officer-hours from the experimental patrols at the saturation/directed patrol locations averaged about 53 per week.

The POP hot spots had officers using a policing strategy that involved the identification and analysis of specific crime and disorder problems in these spots, in order to develop effective response strategies in conjunction with ongoing assessment. Officers working in small teams, along with a crime analyst, were encouraged to explore the “root cause” of the violence problem in the hot spot and come up with ways of solving it. In some cases the teams focused on the offenders, the community, or the need for some kind of environmental crime prevention element. In some cases, the officers worked with residents and other city agencies to develop custom responses to particular problems, and in other cases they used well-tested approaches presented in the Office of Community Oriented Policing Services (COPS Office) POP Guides. For all of the POP hot spots, a high degree of importance was placed on creativity and discretion for the officers.

In total, 60 officers and four crime analysts were assigned to the POP intervention. These officers were divided into two groups, or shifts, that worked different days of the week, thus providing coverage seven days a week. On any given day, each location was worked by one or more officers (usually one to two) and a crime analyst. The officers spent their full shifts working their assigned spots. In total, JSO dedicated 2,100 officer-hours per week to the POP efforts for an average of roughly 95 officers-hours per week per hot spot.16 (However, depending on the nature of their POP activities, officers may not have been physically present in the hot spots for all of this time.) The POP projects were overseen by a total of six sergeants, two lieutenants, and two managerial/supervisory staff of the crime analysis unit.

The police personnel in these spots addressed the underlying factors driving crime in these spots, worked closely with community partners to address the problems, and used the SARA (Scanning, Analysis, Response and Assessment) model. POP officers received 3 days of training in POP and intelligence-led policing, taught by outside experts and JSO staff, including intense training on implementing the SARA model. Officers working the POP spots were provided flexibility in their work schedules to operate at varying times as needed (e.g., working at times when the majority of crimes were committed or working during standard business hours to facilitate meetings with business, government, and community partners).

Officers implemented a wide array of measures at the POP locations. The most common were situational crime prevention measures, such as repairing fences, installing or improving lighting, and erecting road barriers. Officers commonly worked with business owners and rental property managers regarding security measures, business practices, and other forms of prevention and collaboration. Other activities fell into the realms of community organizing (e.g., conducting community surveys and other forms of citizen outreach), social services (e.g., improving recreational opportunities for youth), code enforcement, aesthetic community improvements (e.g., removing graffiti or cleaning up a park), and nuisance abatement. Enforcement and investigation were also used in some locations, though JSO project managers generally encouraged officers to rely more on prevention and deeper measures whenever possible. Overall, the POP officers implemented 283 discrete POP measures across the 22 locations.

Experimental design

Among the flaws found in many policing intervention studies are designs with non-comparable comparison groups (see Mazerolle et al. 2006). While there are a number of exceptions, many policing intervention studies make little attempt to draw comparison groups in ways that maximize the likelihood that they will be similar to the intervention/treatment group. The problem with these types of studies is that although measured differences can be statistically controlled, the many unmeasured variables related to the outcome variable (e.g., susceptibility to change) cannot be controlled. Randomized controlled trials (RCTs) are typically thought of as the best method for eliminating threats to internal validity in evaluating social policies and programs (Berk et al. 1985; Boruch, et al. 1978; Campbell 1969; Campbell and Stanley 1963; Dennis and Boruch 1989; Riecken, et al. 1974). RCTs provide the best counterfactual describing what would have happened to the treatment group if it had not been exposed to the treatment (Rubin 1974; Holland 1986).

The first step in our project was to conduct geographic analyses of the most violent hot spots within Jacksonville. For selection purposes, our team focused on “street violence” (i.e., non-domestic violence) from 2006 through May of 2008. We utilized spatial density analyses to identify hot spots and experimented with different settings so as to balance considerations of area size (our intent was to focus on micro places), sample size, and base rates. All selected hot spots also had to show signs of recent activity—specifically, one or more incidents of violence from January through May of 2008. This process resulted in the selection of 83 hot spots of street violence that averaged 0.02 square miles in size and included a variety of locations such as problem intersections and blocks, apartments, stores, hotels, and bars/entertainment locations. We ensured that each hot spot was a distinct land parcel and that the locations were generally a city block or more apart from one another. Final adjustments to a few locations were also made by JSO staff based on their knowledge of the areas.

Next, these 83 hot spots were randomly assigned to one of three conditions using computer-generated random numbers (see Shadish et al. 2002). All of the assignments were followed flawlessly by the JSO.17 We used a stratified random allocation procedure (see Boruch 1997) and randomized hot spots within statistical “blocks” (based on the number of violent non-domestic incidents in each hot spot) to allow for the likely substantial variation across places (Weisburd and Green 1995).18 The strata for random assignment included the following: 9–17 incidents (35 locations); 18–39 incidents (30 locations); 42–87 incidents (15 locations); and 109–163 (three locations). Within these blocks, the hot spots were randomly assigned to one of three conditions: 40 control hot spots (which included standard patrol services at the current level of police activity at the time), 21 saturation/directed patrol hot spots or 22 POP hot spots. 19 Each of these three conditions was maintained for a 90-day period.

Procedures were also established to monitor the integrity of the assignment process and monitor for expectancy, novelty, disruption, and local history. Supervisors tracked officer activities to check that the officers were in their assigned hot spot applying their assigned condition. We also conducted detailed video-taped interviews and “ride-alongs” with the study officers and other patrol officers to assess their implementation of their study condition and conduct treatment integrity checks (e.g., query them on their adherence to the study protocols). No problems were revealed through these treatment integrity checks. We did not have any issues with officers straying out of their assigned areas (which none did, except for a couple of emergency cases where the officers were needed to provide backup).

Measures

We collected a variety of traditional police outcome measures of enforcement activity for the hot spots and surrounding areas (to assess for displacement/diffusion effects), including: incident/Uniform Crime Report (UCR) data, arrest data, calls-for-service (911) data, and data on field interviews and other self-initiated police activity (e.g., traffic stops and investigation of miscellaneous matters).20 For our UCR, arrest and calls-for-service data we created three sub-measures for each data source, but the self-initiated activity and the field stops were just total counts for each of those measures (further differentiation was not available).

We used UCR data to measure the occurrence of crime and created three sub-measures based on UCR Part I data, including: Violent crime (aggravated assault, forcible rape, murder, and robbery), non-domestic violent crime (excludes any of the aforementioned violent crimes that involved a domestic/intimate partner relationship between the victim and perpetrator), and property crime (arson, burglary, larceny-theft, and motor vehicle theft). We used a similar coding scheme for our three calls-for-service sub-measures and our three arrest sub-measures. Note that although many prior studies of hot spots policing have examined program impacts on minor forms of crime and disorder, our evaluation focuses on violence and serious property crime because JSO conducted this initiative specifically as a means of reducing serious crime, especially violence. Our focus on these outcomes also addresses a particularly important question for policy and research, given that prior studies have not provided consistent evidence that hot spots policing reduces violence and other serious crime.

For all of these above measures, we created a 100-foot buffer around each hot spot. That is, a “hit” would “count” for a hot spot for the purposes of our research if it occurred either on the specific hot spot or within 100 feet of the hot spot (e.g., this allows us to include parking lots along the hot spots and other similar areas in the immediate proximity of the hot spot). We believe it makes sense to include these areas within 100 feet of the hot spot, for they were covered by the officers on these hot spots.

Our crime outcome measures were collected for the 90-day periods before and during the intervention for each hot spot and for the 90-day period immediately after the intervention. We hypothesized that the effects of POP would be more likely to persist (or simply become more evident) during the post-intervention period as the POP efforts were completed and their effects became fully realized. Further, as others have noted (e.g., Braga et al. 1999; Weisburd and Green 1995), examining post-intervention impacts lessens the possibility that the results will be confounded by reporting effects (i.e., a change in the propensity of people in a program area to report crimes to police during the program period).

Results

The first sets of analyses (see Table 1) describe the key analytic variables and summarize the nature of the distribution of our data. Table 1 includes means and standard deviations for all of study variables. Tables 2 through 5 present multivariate models to be described below.
Table 1

Means and standard deviations for three study conditions and entire sample

 

Assigned condition

 
 

CONTROL (n = 40)

SATURATION (n = 21)

SOLVING (n = 22)

Total (n = 83 all vars)

 

Mean

Std. deviation

Mean

Std. deviation

Mean

Std. deviation

Mean

Std. deviation

N

Self-initiated police activity

1 year prior to treatment

105.5

101.2

109.5

85.6

155.5

191.1

119.7

128.4

83

90-day treatment period

105.9

106.5

255.3

163.8

200.8

183.9

168.9

157.2

83

90-day post-treatment

118.2

108.2

114.2

90.8

175.0

199.7

132.2

135.9

83

Police field stops

1 year prior to treatment

17.0

21.6

18.8

14.4

29.5

49.6

20.8

30.5

83

90-day treatment period

14.7

16.1

24.3

17.9

32.1

39.7

21.8

25.7

83

90-day post-treatment

17.2

15.9

17.1

19.6

23.8

29.4

18.9

21.1

83

Calls-for-service (911 calls)

1 year prior to treatment - Any violence

4.4

4.5

4.9

4.5

5.0

5.9

4.7

4.9

83

90-day treatment period - Any violence

3.3

3.5

2.7

2.5

3.7

4.5

3.2

3.5

83

90-day post-treatment - Any violence

2.9

4.2

3.9

4.2

4.2

6.1

3.5

4.7

83

1 year prior to treatment - Non-domestic violence

4.3

4.2

4.8

4.5

4.9

6.0

4.6

4.7

83

90-day treatment period - Non-domestic violence

3.2

3.4

2.6

2.5

3.6

4.5

3.2

3.5

83

90-day post-treatment - Non-domestic violence

2.9

4.2

3.8

4.0

4.1

6.1

3.4

4.7

83

1 year prior to treatment - Property crime

18.7

19.4

23.1

18.2

22.7

19.0

20.9

18.9

83

90-day treatment period - Property crime

15.3

15.3

17.9

20.1

17.0

13.7

16.4

16.1

83

90-day post-treatment - Property crime

18.2

21.4

20.1

21.6

17.2

13.3

18.4

19.4

83

Arrests

1 year prior to treatment - Any violence

1.2

2.0

0.8

1.1

1.7

2.1

1.2

1.9

83

90-day treatment period - Any violence

0.7

1.6

0.6

0.7

1.0

1.0

0.7

1.3

83

90-day post-treatment - Any violence

1.0

1.4

0.4

0.8

0.7

0.9

0.8

1.2

83

1 year prior to treatment - Non-domestic violence

1.0

1.9

0.5

0.7

1.5

1.9

1.0

1.7

83

90-day treatment period - Non-domestic violence

0.5

1.2

0.5

0.7

0.8

0.8

0.6

1.0

83

90-day post-treatment - Non-domestic violence

0.8

1.1

0.3

0.7

0.5

0.8

0.6

0.9

83

1 year prior to treatment - Property crime

4.2

14.6

3.9

9.2

2.2

3.1

3.6

11.2

83

90-day treatment period - Property crime

3.0

10.4

5.0

15.3

2.5

3.3

3.4

10.6

83

90-day post-treatment - Property crime

4.2

16.8

5.9

19.1

1.6

2.6

3.9

15.1

83

Uniform crime reports (UCR)/incidents of crime

1 year prior to treatment - Any violence

3.6

3.9

3.0

3.8

4.5

4.4

3.7

4.0

83

90-day treatment period - Any violence

2.5

3.0

1.9

1.6

3.4

3.9

2.5

3.0

83

90-day post-treatment - Any violence

2.9

3.7

2.8

3.1

3.4

4.6

3.0

3.8

83

1 year prior to treatment - Non-domestic violence

3.3

3.7

2.8

3.5

4.3

4.2

3.4

3.8

83

90-day treatment period - Non-domestic violence

2.1

2.3

1.7

1.6

2.9

3.4

2.2

2.5

83

90-day post-treatment - Non-domestic violence

2.5

2.9

2.5

2.7

2.9

4.4

2.6

3.3

83

1 year prior to treatment - Property crime

13.1

15.0

15.0

16.0

11.0

10.6

13.0

14.1

83

90-day treatment period - Property crime

10.0

13.2

13.7

20.1

11.0

12.1

11.2

14.9

83

90-day post-treatment - Property crime

11.2

18.9

15.3

22.4

10.5

9.4

12.0

17.8

83

Table 2

Implementation models for the 90-day intervention period (n = 83)

 

Arrests

 

Self-initiated

Field stops

Any violence

Non-domestic violence

Property crime

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

Treatment condition (reference: control)

 Saturation

2.91 (.439)***

1.85 (.390)**

0.97 (.398)

0.94 (.390)

1.01 (.285)

 Problem-solving

1.50 (.220)**

1.33 (.281)

1.35 (.490)

1.36 (.469)

1.45 (.380)

Prior year activity pre-treatment

1.00 (.001)

1.00 (.004)

1.08 (.128)

1.10 (.131)

1.22 (.045)***

Prior 90-day activity pre-treatment

1.00 (.001)***

1.03 (.009)***

1.25 (.100)**

1.13 (.097)

0.94 (.018)**

Type of hotspot (reference: mixed use)

 Residential

1.03 (.165)

0.81 (.178)

0.70 (.234)

0.68 (.235)

0.59 (.168)

 Business

1.08 (.175)

0.82 (.184)

0.35 (.163)*

0.37 (.169)*

0.86 (.235)

 

Lnalpha

–1.24 (.160)

–0.64 (.180)

–1.17 (.808)

–1.13 (.560)

 
 

alpha

0.29 (.046)

0.52 (.094)

0.31 (.250)

 

0.32 (.182)

 

Log likelihood

–461.23

–310.18

–85.26

–79.56

–135.46

 

Likelihood ratio X2

93.82

60.28

27.20

17.56

44.29

 

Prob > X2

0.00

0.00

0.00

0.01

0.00

 

Pseudo R2

0.09

0.09

0.14

0.10

0.14

* p <.05, ** p<.01, ***p<.001

Analysis for pre-treatment differences across the three study conditions

As seen in Table 1, no pre-treatment differences emerged in our three study conditions based on any of the study measures including: pre-treatment levels of self-initiated police activity, field stops by officers, calls-for-service for non-domestic (“street”) violence, calls-for-service for any violence (domestic and non-domestic), calls-for-service for property crimes, arrests for non-domestic violence, arrests for any violence, arrests for property crimes, UCR reported non-domestic violence, UCR reported any violence and UCR reported property crimes. We also examined two other variables, location type (pure residential, pure business or mixed residential and business21) and size of hot spot (measured in square miles22). There was no evidence of any statistical differences for our three study conditions based on either location type (X2 = 4.45 [df = 4], p = 0.35) or size of hot spot (F = 0.41 [df = 2/80], p = 0.67). Combined with evidence from Table 1, our data suggests that our random assignment process worked as planned and created comparable intervention/control conditions.

Changes during the experimental period

Table 1 also presents implementation and outcome data for the 3-month period when the interventions were implemented. A review of Table 1 reveals only two statistically significant differences (statistical tests not shown on Table 1 but are reported here in the text) across the three study conditions for the outcomes of self-initiated police activity and field stops by officers. Hot spots receiving the POP intervention (mean = 200.8) and the saturation/directed patrol intervention (mean = 255.3) had significantly more officer self-initiated activity (F = 8.0 [df = 82], p < .001) than the control group hot spots (mean = 163.8) (Bonferroni test for POP to control comparison = p = .048 and saturation/directed patrol to control = p = .001). Our finding of higher levels of self-initiated activity is consistent with our design to have the interventions provide more policing services than standard patrol. Also, hot spots receiving the POP intervention (mean = 32.1) (but not the saturation/directed patrol intervention with a mean of 24.3) had significantly more officer field stops (F = 3.6 [df = 82], p = .031) than the control group hot spots (mean = 14.7) (Bonferroni test for POP to control comparison  p = .030 and saturation/directed patrol to control = p = .468). However, no differences emerged for the POP or the saturation/directed patrol hot spots compared to the control group hot spots for our violence or property crime arrest or outcome measures.

Changes during the 3-month follow-up period

Finally, Table 1 presents outcome data for the 3-month period after the interventions were implemented. A review of Table 1 reveals no statistically significant univariate differences across the three study conditions for any of our outcome measures. Our 3-month follow-up results are further tested in a series of multivariate models (see below) where other factors can be assessed and held constant.

Use of special overtime resources after the intervention period

We also examined whether there were differences in post-treatment police activity in our hot spots. The JSO periodically uses special overtime resources to provide extra coverage to high crime areas. During our experimental period, JSO agreed not to use these special resources and kept them out of our study areas (where we had different resources to deploy). However, after the conclusion of our intervention period the JSO used these overtime resources throughout Jacksonville, including some of our study areas. While from a research standpoint it might have been better to not have the study areas treated in this way, to avoid any complications with disentangling intervention effects, it was not possible for the research team to avoid this post-treatment activity given the real-world problems that JSO was attempting to address. Nevertheless, we were concerned that if these special overtime resources were deployed disproportionately across some of our study conditions that our outcome measures for one of our intervention groups might be artificially affected. Our data suggest that there was a balance across the three study groups in exposure to these post-intervention activities.

From May 1 to June 20, 2009, 26 officers working on overtime conducted evening and nighttime directed patrols (generally from 6:00 p.m. to 3:00 a.m.) within designated high crime areas averaging between 4 and 5 square miles in size. Some of our study hot spots were included in these areas, so we tested for differences across study groups in the share of hot spots that may have been affected by these patrols. These percentages were 30% for the control group, 38% for the saturation/directed patrol group, and 32% for the POP group. No statistically significant differences emerged from an analysis of these percentages (F = .20 [2, 80], p = 0.82).

In addition, during the last 3 weeks of the follow-up period, JSO began new or “maintenance” POP initiatives in several of the study hot spots, including locations from all three study groups. A pair of officers was assigned to each of these locations. The control group included 21% of the locations targeted by this initiative, the saturation/directed patrol group included 28%, the POP group included 30%, and locations not included in the study accounted for the remaining 21%. No statistically significant differences emerged from an analysis of these percentages (F = .59 [2, 80], p = 0.56). In the multivariate models described below, we control for whether a hot spot was affected by either the new patrol or POP initiatives.

Multivariate models

Although not strictly necessary because we are working with experimental data, we will also introduce a set of covariates to our outcome models. Introducing covariates is increasingly common in analyzing data from randomized experiments (Patel 1996). The introduction of covariates allows us to improve the precision of the treatment comparisons by reducing error variance and correcting for any major imbalances in the distribution of these covariates across the treatment and control groups that may have occurred due to chance (Armitage 1996). Adding covariates also can help adjust for the natural variation between cases within the comparison groups (Gelber and Zelen 1986). However, the introduction of covariates could introduce bias into our study if they are unevenly distributed across the treatment and control groups, especially given our relatively small sample size. Our tests for differences across the treatment and control groups on the covariate measures showed little statistical differences.

To follow are three sets of statistical models related to measuring the implementation of the interventions and outcomes associated with the interventions. For the first set of models, we will assess the implementation of the two interventions compared to the control groups in changing the level of self-initiated police activity, field stops, and arrests (for non-domestic violence, any violence, and property crimes) by officers during the 90-day treatment period. For the second set of statistical models, we examine the effectiveness of the two interventions in changing crime at the hot spots during the 90-day intervention period. The final set of models assesses the residual effects of the interventions during the 90-day post-intervention period. In some cases we use Poisson regression and in other cases we use negative binomial regression,23 based on the distribution of the data.24 For each of the three implementation regression models we introduced a set of five independent variables to the models. First, we have two variables indicating the type of intervention applied to the hot spot (the reference category is the control group/standard patrol). Next, we have covariates for the level of the outcome variable during both the prior 90 days and the same 90-day period of the prior year (to assess seasonal effects).25 Finally, we have an independent variable for the type of hot spot (the reference category for business/residential is “mixed use”). Our models for crime outcomes during the post-intervention period also include two post-period indicators to control for the additional patrol and POP activities described previously. Also note that the prior 90-day measure in these models corresponds to the 90-day experimental period.26

Models measuring indicators of the implementation of the interventions

We assessed the implementation of the two interventions compared to the control groups in changing the level of self-initiated police activity, field stops and arrests (for non-domestic violence, any violence, and property crimes) by officers during the 90-day treatment period.

In Table 2, we present our regression results for the implementation models. First, for the 90-day treatment period, our negative binomial model of self-initiated police activity (log likelihood = –461.22, X2 = 93.8, p < .001, Pseudo R2 = .09) shows that the assignment to the saturation/directed patrol condition (β = 2.9, z = 7.06, p < .001) and assignment to the problem-solving condition (β = 1.5, z = 2.77, p < .01) were both significantly related to the level of self-initiated police activity during the intervention period. The saturation/directed patrol group (199% higher count of events) and the problem-solving group (50% higher count of events) were associated with higher levels of self-initiated police activity during the intervention compared to the control group.

Next, for the negative binomial model of police field stops during the 90-day treatment period (log likelihood = –310.18, X2 = 60.3, p < .001, Pseudo R2 = .09), we observed that assignment to the saturation/directed patrol condition (β = 1.85, z = 2.92, p < .01) was significantly related to the level of police field stops during the intervention period. The saturation/directed patrol group (85% higher count of events) was associated with higher levels of police field stops during the intervention compared to the control group. While not statistically significant, the POP group was associated with higher levels of police field stops during the intervention (33% higher count of events) compared to the control group (β = 1.33, z = 1.33, p = .18).

Finally, our negative binomial regression model for violent crime arrests during the treatment period (log likelihood = –85.26, X2 = 27.2, p < .001, Pseudo R2 = .14) revealed that none of the intervention assignment variables were statistically significant. Neither of the intervention variables (saturation/directed patrol and problem-solving) was statistically significant in this model for arrests for violence during the 90-day treatment period. Also, neither of the intervention variables was statistically significant in our other arrest models for non-domestic/street violence and property crime arrests during the 90-day treatment period (see Table 2).

Crime impact models

In the crime outcome models (see Tables 3 and 4), we examined changes in crime during the 90-day periods of and following the intervention. In both sets of models, we examined the effectiveness of the two interventions compared to the control group in changing levels of crime (based on calls-for-service and UCR measures) controlling for pre-intervention crime levels and other covariates (discussed earlier). We hypothesized that the effects of the saturation/directed patrols would be strongest during the intervention period but that these effects would be likely to decay during the subsequent months. In contrast, we hypothesized that the effects from the POP intervention might be more evident during the post-intervention period because officers would need time to analyze problems and implement solutions. We also expected that the effects from POP would be less likely to decay over time.
Table 3

Crime impact models for the 90-day intervention period (n = 83)

 

Calls-For-Service/911

Uniform crime reports/Incidents of crime

Any violence

Non-domestic violence

Property crime

Any violence

Non-domestic violence

Property crime

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

Treatment condition (reference: control)

 Saturation

0.80 (.172)

0.82 (.177)

0.87 (.122)

0.88 (.205)

0.96 (.235)

0.97 (.155)

 Problem-solving

0.89 (.185)

0.91 (.191)

0.93 (.126)

1.09 (.227)

1.16 (.253)

0.97 (.151)

Prior year activity pre-treatment

1.06 (.029)*

1.07 (.208)*

1.02 (.006)***

1.04 (.040)

1.06 (.042)

1.06 (.013)***

Prior 90-day activity pre-treatment

1.12 (.025)***

1.12 (.025)***

1.02 (.005)***

1.14 (.033)***

1.15 (.039)***

1.00 (.009)

Type of hotspot (reference: mixed use)

 Residential

0.91 (.205)

0.89 (.206)

0.82 (.117)

1.23 (.271)

1.11 (.262)

0.79 (.127)

 Business

1.04 (.248)

1.09 (.264)

0.83 (.115)

0.88 (.232)

0.97 (.264)

0.84 (.136)

 

lnalpha

–1.47 (.433)

–1.47 (.439)

–1.77 (.255)

–1.77 (.610)

–1.73 (.665)

–1.57 (.270)

 

alpha

0.23 (.099)

0.23 (.101)

0.17 (.044)

0.17 (.104)

0.18 (.118)

0.21 (.056)

 

Log likelihood

–168.58

–166.38

–270.7

–151.23

–143.94

–242.44

 

Likelihood ratio X2

46.56

48.00

90.96

47.18

40.25

89.28

 

Prob > X2

0.00

0.00

0.00

0.00

0.00

0.00

 

Pseudo R2

0.12

0.13

0.14

0.13

0.12

0.16

* p <.05, ** p<.01, ***p<.001

Table 4

Crime impact models for the 90-day post-intervention period (n = 83)

 

Calls-For-Service/911

Uniform crime reports/Incidents of crime

Any violence

Non-domestic violence

Property crime

Any violence

Non-domestic violence

Property crime

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

Treatment condition (reference: control)

 Saturation

1.32 (.324)

1.34 (.332)

0.88 (.108)

1.27 (.262)

1.27 (.228)

1.00 (.149)

 Problem-solving

0.89 (.223)

0.87 (.222)

0.86 (.102)

0.80 (.166)

0.67 (.126)*

0.95 (.137)

Prior year activity pre-treatment

1.09 (.034)**

1.09 (.035)**

1.02 (.006)**

0.99 (.024)

1.00 (.016)

1.02 (.011)

Prior 90-day activity pre-treatment

1.15 (.034) ***

1.15 (.034)***

1.03 (.006)***

1.23 (.038)***

1.27 (.030)***

1.03 (.010)**

Type of hotspot (reference: mixed use)

 Residential

0.82 (.204)

0.85 (.216)

1.11 (.142)

0.88 (.175)

1.00 (.182)

0.97 (.145)

 Business

0.64 (.175)

0.66 (.185)

0.80 (.104)

0.50 (.122)**

0.53 (.121)**

0.67 (.109)*

Post-treatment police activity 1 (May 1 to June 20)

1.46 (.349)

1.46 (.353)

1.13 (.143)

1.22 (.256)

1.06 (.206)

1.18 (.184)

Post-treatment police activity 2 (After June 20)

0.82 (.209)

0.83 (.211)

0.98 (.128)

0.78 (.180)

0.85 (.178)

0.87 (.140)

 

lnalpha

–1.01 (.312)

–0.98 (.313)

–2.17 (.279)

–2.10 (.658)

 

–1.83 (.284)1

 

alpha

0.36 (.113)

0.38 (.118)

0.11 (.032)

0.12 (.081)

 

0.16 (.046)

 

Log likelihood

–168.62

–167.84

–268.72

–153.40

–144.35

–240.30

 

Likelihood ratio X2

56.97

56.23

111.96

66.48

159.36

104.56

 

Prob > X2

0.00

0.00

0.00

0.00

0.00

0.00

 

Pseudo R2

0.14

0.14

0.17

0.18

0.36

0.18

* p <.05, ** p<.01, ***p<.001

Table 3 presents outcome model results for the implementation period. These models did not reveal statistically significant effects from either of the interventions. However, the saturation/directed patrol locations showed decreases in all crime measures (albeit not statistically significant). These reductions ranged from 4–20% for the violence measures (most estimates ranged from 12– 20%) and from 3–13% for the property crime measures. Results were more mixed for the POP locations, ranging from an 11% reduction in calls about any violence to a 16% increase in incident reports for non-domestic violence.

Models for the post-intervention period appear in Table 4. Our main finding on the effects of the interventions on crime during this period is for our “street violence” (all non-domestic UCR Part I violent incidents) 90-day post-intervention model. For our Poisson regression model for “street violence” (log likelihood = –144.3, X2 = 159.4, p <.001, Pseudo R2= 0.36), we observed that assignment to the problem-solving condition (β = 0.67, z = –2.13, p =.03) was significantly related to our “street violence” 90-day post-intervention outcome. More specifically, the use of the problem-solving intervention was associated with a 33% reduction in the count of “street violence” during the 90-day post-period.

While neither of our other two UCR models yielded statistically significant results for either of our intervention variables, the results were in the predicted direction for our problem-solving measure. That is, for our negative binomial regression model for all forms of Part I UCR violence (log likelihood = –153.39, X2 = 66.5, p < .001, Pseudo R2 = .18) we observed that our problem-solving measure was in the predicted direction (β = 0.80, z = –1.07, p = 0.28), but was not below the statistically significant level of 0.05. Also, our negative binomial regression model for UCR property crime (log likelihood = –240.30, X2 = 104.6, p <.001, Pseudo R2= 0.18) revealed that our problem-solving measure was in the predicted direction (β = 0.95, z = –0.34, p = 0.77), but was not below the statistically significant level of 0.05.

Although not statistically significant, both total and non-domestic violent incidents increased 27% in the saturation/directed patrol locations, reflecting a rebound after the end of the special patrols. Property offenses remained unchanged.

Next we examined the effects of our interventions on calls-for-service for “street violence” (non-domestic violence) 90-day post-intervention. For our negative binomial regression model for calls involving “street violence” (log likelihood = –167.8, X2 = 56.2, p <.001, Pseudo R2=.14), we observed that neither the problem-solving nor the saturation/directed patrol yielded statistically significant results. However, the problem-solving measure was in the predicted direction (β = 0.87, z = –0.54, p = 0.59), but was not below the statistically significant level of 0.05. Similar findings emerged for our negative binomial regression model for calls involving all forms of violence (log likelihood = –168.62, X2 = 56.9, p < .001, Pseudo R2 = .14), where we observed that our problem-solving measure was in the predicted direction (β = 0.889, z = –0.47, p = 0.64), but was not below the statistically significant level of 0.05. Additionally, our negative binomial regression model for calls involving property crime (log likelihood = –268.72, X2 = 111.9, p < .001, Pseudo R2 = .14), where we observed that our problem-solving measure was in the predicted direction (β = 0.86, z = –1.28, p = 0.20), but also was not below the statistically significant level of 0.05.

In the saturation areas, in contrast, there were increases of 32–34% in calls for total and non-domestic violence and a 12% decline in calls for property offenses. However, none of these changes were statistically significant.

Visual assessment of potential displacement

Our final set of analyses assess if crime displacement or diffusion of benefits occurred from our hot spots to areas adjacent or near these spots. We examine displacement/diffusion in the areas adjacent to our study hot spots that are beyond the 100 foot buffer of the hot spot but also within 500 feet of the respective hot spot.27 If displacement or diffusion was occurring, we would expect there to be statistical changes in these areas immediately adjacent to the hot spots. In our final model (see Table 5) we examined crime outcomes 90 days after the interventions in the areas adjacent to our hot spots. Specifically, we examine the effectiveness of the two interventions compared to the control group in changing 90-day post-period levels of crime (based on calls-for-service and UCR measures) in the areas adjacent to our hot spots, controlling for pre-intervention crime levels and other covariates (discussed earlier).
Table 5

Crime displacement impact models for the 90-day post-intervention period (n = 83)

 

Calls-For-Service/911

Uniform crime reports/Incidents of crime

Any violence

Non-domestic violence

Property crime

Any violence

Non-domestic violence

Property crime

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

IRR (SE)

Treatment condition (reference: control)

 Saturation

0.97 (.237)

0.98 (.243)

0.97 (.145)

1.55 (.491)

1.20 (.390)

1.05 (.209)

 Problem-solving

1.71 (.360)*

1.69 (.364)*

0.94 (.139)

1.25 (.355)

1.16 (.346)

1.15 (.216)

Prior year activity pre-treatment

1.15 (.037)***

1.15 (.037)***

1.03 (.010)**

1.19 (.052)***

1.21 (.080)**

1.10 (.023)***

Prior 90-day activity pre-treatment

1.14 (.051)**

1.14 (.053)**

1.04 (.011)***

1.32 (.085)***

1.32 (.132)**

1.05 (.028)

Type of hotspot (reference: mixed use)

 Residential

0.79 (.178)

0.80 (.180)

0.94 (.146)

1.40 (.399)

1.06 (.298)

0.74 (.144)

 Business

0.94 (.217)

0.90 (.212)

1.01 (.161)

1.34 (.454)

1.06 (.357)

0.92 (.191)

Post-treatment police activity 1 (May 1 to June 20)

1.42 (.314)

1.44 (.321)

1.07 (.156)

0.84 (.241)

1.10 (.311)

1.68 (.301)**

Post-treatment police activity 2 (After June 20)

0.85 (.205)

0.84 (.206)

1.07 (.174)

1.12 (.317)

1.09 (.339)

0.77 (.152)

 

lnalpha

  

–1.75 (.292)

  

–1.81 (.489)

 

alpha

  

0.17 (.051)

  

0.16 (.080)

 

Log likelihood

–122.61

–121.74

–244.50

–92.44

–90.48

–178.11

 

Likelihood ratio X2

76.17

74.36

68.78

68.58

35.09

67.31

 

Prob > X2

0.00

0.00

0.00

0.00

0.00

0.00

 

Pseudo R2

0.24

0.23

0.12

0.27

0.16

0.16

p <.05, ** p<.01, ***p<.001

Neither of our interventions produced statistically significant changes in any of our crime outcome models based on UCR data for the areas adjacent to our hot spots. Our main finding on the effects of the interventions on crime in adjacent hot spots is for our any violence (all domestic and non-domestic violent incidents based on calls-for-service) and our street violence (all non-domestic violent incidents based on calls-for-service) 90-day post-intervention models. First, for our Poisson model for calls-for-service for any violence (log likelihood = –122.6, X2 = 76.2, p < .001, Pseudo R2 = .24), we observed that assignment to the problem-solving condition (β = 1.71, z = 2.53, p = .01) was significantly related to our calls-for-service for any violence in the areas adjacent to our hot spots. The use of the POP intervention was associated with a 29% increase in the count of calls-for-service for any violence.

Second, for our Poisson model for calls-for-service for street violence (log likelihood = –121.7, X2 = 74.4, p < .001, Pseudo R2 = .23), we also observed that assignment to the problem-solving condition (β = 1.69, z = 2.45, p = .01) was significantly related to calls-for-service for street violence in the areas adjacent to our hot spots. The use of the problem-solving intervention was associated with a 31% increase in the count of calls-for-service for street violence.

Discussion

Focusing police efforts on hot spots has gained widespread acceptance among researchers and practitioners. However, little evidence has been amassed as to what types of policing strategies work best for hot spots, including interventions targeted at different types of hot spots, such as hot spots of violent crime. In particular, we are unaware of any experimental tests in the published literature on the effectiveness of problem-oriented policing compared to saturation/directed patrol at hot spots of violence. To address this gap, we identified 83 hot spots of violence in Jacksonville, Florida and randomly assigned them to problem-solving, saturation/directed patrol, or control conditions for a 90-day experimental period. Problem-solving activities were conducted by a special team of supervisors, officers, and crime analysts. The saturation/directed patrol efforts were conducted by a mix of on-duty and overtime-funded officers and guided by allocations determined through analysis of crime trends and patterns. The control group involved standard or routine patrol services.

For the first set of multivariate models, we assessed the implementation of the two interventions compared to the control groups in changing the level of self-initiated police activity, field stops and arrests by officers during the 90-day treatment period. First, for the 90-day treatment period of self-initiated police activity model we observed that the saturation/directed patrol group and the problem-solving group were associated with higher levels of self-initiated police activity during the intervention compared to the control group. Next, for the 90-day treatment period of field stops model we found that the saturation/directed patrol group was associated with a statistically higher level of police field stops during the intervention compared to the control group. While not statistically significant as in our earlier examination of means of field stops, the POP group was associated with higher levels of police field stops during the intervention compared to the control group. We note this non-statistically significant result for a reason. As discussed earlier, the interpretation of our substantive results is complicated by our modest sample size (n = 83) and associated limited statistical power. While we find it useful to present our statistical tests of significance to provide a general guide for interpreting our data, we think it is also important to consider the direction, consistency and magnitude of the effects. For example, we believe it is worth noting that a series of non-significant findings in the same predicted direction emerged for our 90-day treatment period of arrest models. Our models for the 90-day treatment period of arrests for non-domestic/street violence, any violence, and property crime all had POP intervention variables that were associated with more arrests compared to the control group (but the results were all non-significant).

Our finding of higher levels of self-initiated activity, field stops, and possibly arrests is consistent with our design to have the interventions provide more policing services than standard patrol. The JSO provided close to a million dollars in overtime to implement the interventions, including funding 60 officers and eight supervisors to do POP and an average of over 50 officer-hours per week at the saturation/directed patrol hot spots. The POP and saturation/directed patrol teams maintained detailed implementation logs to assure that the work was being done. The POP teams also were required to make presentations to command staff on their efforts and results, further enhancing their accountability to the sheriff.

Moving onto the second set of statistical models, we examined the effectiveness of the two interventions compared to the control groups in changing levels of crime, as measured by calls-for-service and UCR measures, during both the 90-day intervention period and the following 90 days. Our examination of the latter period allowed us to explore the residual effects over time of the POP compared to the saturation/directed patrol intervention, while minimizing potential reporting effects that could obscure our findings. In sum, we found no statistically significant effects from treatment during the intervention period, though the crime measures declined by up to 20% in the saturation/directed patrol locations. Our main finding on the effects of the interventions on crime is for our “street violence” (all non-domestic UCR Part I violent incidents) 90-day post-intervention model. The use of the problem-solving intervention was associated with a 33% reduction in the count of “street violence” during the 90-day post-period. While neither of our other two UCR measures for any violence or property crime yielded statistically significant results for either of our intervention variables during this period, the results were in the predicted direction for our problem-solving measure. We observed that our problem-solving group was associated with a 20% reduction in the count of any Part I UCR violence and a 5% reduction in the count of property crime as measured by the UCR, but not below the statistically significant level of 0.05. A similar pattern emerged for our calls-for-service measures. We observed that our problem-solving group was associated with a 13% reduction in the count of calls for non-domestic/street violence, 11% reduction in the count of calls-for-service for any violence, and a 14% reduction in calls involving property crime. However, none of these calls-for-service results reached the traditional 0.05 level of statistical significance.

One possible reason for the absence of larger effects of the POP intervention may be related to the relatively short intervention period of 90 days for this experiment. Given that police officers often struggle implementing POP (see Scott 2000) and that POP was new to the JSO, 90 days may not be enough time to develop a good understanding of the hot spots and implement an appropriate intervention. After 90 days, thin POP responses may need to be re-adjusted and other strategies might need to be implemented.28 Adjustment of response is an important part of the POP process that may not occur in only a 90-day time frame and this may have constrained the effectiveness of the POP process in Jacksonville.

Yet despite this shortcoming, the intervention was intensive, and results imply that the POP efforts began to produce the desired effects once they were implemented and had time to take hold. The 33% reduction found for incidents of non-domestic violence during the follow-up period implies that the intervention prevented roughly 31 serious non-domestic incidents across the 22 POP locations during just this 90-day period.29 In principle, this effect may have lasted well beyond that time.

In the area of crime displacement, however, we observed two statistically significant results that require close examination. The use of the problem-solving intervention was associated with a 29% increase in the count of “any” violence based on calls-for-service to the police and a 31% increase in the count of “street” violence based on calls-for-service to the police during the 90-day post-period for areas adjacent to our hot spots. However, neither of our interventions produced statistically significant changes in any of our crime outcomes based on UCR data for the areas adjacent to our hot spots. One interpretation of these results is that the POP intervention that reduced street violence in our hot spots also increased the sensitivity of residents and workers in areas adjacent to the hot spots about a more effective and engaging police presence. This higher sensitivity may have led to greater reporting of crime in these areas. People in these areas may have become aware of the POP efforts, for example, through direct observation or discussion with other neighbors. In some instances, POP efforts may have also required officers to contact people in nearby areas or address issues in those areas that contributed to problems in the target locations. For all these reasons, people living or working near the target locations may have thus attempted to do their part by increasing their calls to police about violent crime. Support for this interpretation can be found in the fact that no differences were observed in our UCR measures of crime for either of our interventions compared to the control group for the adjacent areas.

A related point is that the calls for service measures showed both smaller crime reductions in the POP locations (for the post-intervention period) and larger increases in areas adjacent to the POP locations. Together, these patterns may suggest a general upward reporting bias in the calls for service. Similar to other official data sources, calls for service data rely on recognition of criminal behavior coupled with bringing it to the attention of the police. However, calls for service data typically contain higher counts of events then other official records mostly because these data represent an unfiltered account that might include behavior that is not criminal but is initially reported as criminal or incidents that cannot be verified. In Klinger and Bridges research (1997) they found a number of false positive calls for service where a citizen reported non-criminal behavior as criminal.

Of course, an alternative interpretation is that the greater number of calls-for-service for violence in areas near our hot spots is evidence of crime displacement. If true, this result is contrary to that of many prior hot spots studies. This may be due in some measure to the nature of violent crime and violent crime hot spots. As discussed earlier, hot spots policing has generally been less effective in reducing violence, due perhaps to the more impulsive, expressive nature of many violent crimes. We also noted that violent offenders commit offenses closer to home and that violent crime is more geographically concentrated than property crime. While these factors might make distant displacement of violence less likely, they may also increase the likelihood of displacement to very nearby areas, at least in the short run. Hence, in response to the altered opportunity structures at the POP locations in Jacksonville, some offenders who frequented those locations may have slightly moved their hangouts and activities to places that were very close, familiar, and similar.

Another possible reason for actual crime displacement may have been the abbreviated nature of JSO’s intervention. Our POP intervention of 90 days may have been long enough to reduce violent crime in the targeted hot spots but not long enough to stop these violent crimes altogether. Perhaps criminals identified the targeted hot spots as places where they would likely get caught, but noticed considerably less police activity in the nearby areas. This observation by criminals might have been reinforced when the intense policing activity stopped after 90 days. That is, the criminals were still weary of re-entering the former targeted locations but felt greater comfort to commit violent crimes in the nearby areas that never received the intense POP intervention. If the intervention proceeded for a longer period of time (e.g., 6 or 12 months) these same violent criminals might have begun to see the inevitability of getting caught by the police and desisted from violent crime activity in these nearby areas. The abbreviated nature of our intervention might have also interacted with the absence of any POP interventions ever being used in Jacksonville in a systematic manner, as implemented in our study. That is, the criminals near the study violent hot spots may have been unaccustomed to seeing POP and when it stopped after 90 days assumed it was like other policing efforts that start and disappear soon after (leading them to become a bit less concerned about being caught by the police at least in the areas near the targeted spots).

Under either interpretation, POP leading to reporting effects or crime displacement, our results suggest that law enforcement agencies need to be cognizant of possible effects of implementing POP in areas near the intervention site. That is, agencies may need to not only implement POP in specific hot spots of violent crime, but also provide some type of intervention in the surrounding area to either address actual displacement of violent crime or greater reporting of it when it happens.

Limitations

Prior to drawing any conclusions from our research, it is important to consider limitations associated with our study. Like other randomized control trials, our study has a number of strengths related to the strong counterfactual we created. We have good evidence that our random assignment process worked as planned (we detected no pre-treatment differences and experienced no misassignments connected with the random assignment process) and created comparable intervention/control conditions. We have a high degree of confidence in our ability to describe what would have happened to the treatment group if it had not been exposed to the treatment. One downside of our experimental assignment process was the creation of somewhat artificial conditions under which we asked the JSO to operate. While the officers in our study carefully followed the assignment pattern dictated by the experiment, the officers did not seem to like being confined to small hot spots. While our approach is not very different from other hot spot policing strategies used by the JSO and other agencies, the officers would have preferred to move more naturally through the city’s high crime areas (e.g., work many more hot areas in a given shift and move away from hot areas that happen to be very slow on a given night).

Another potential limitation of our study is the number of hot spots included in our research. Even in a city such as Jacksonville, which had a fairly high violent crime rate (at the top of the list within Florida around the time of the study), we struggled to identify 83 hot spots for violent crime. While a sample size of 83 is not a small study, especially in this context, it does provide some limitations in statistical power. For example, while this study had a sample size comparable to or larger than that of many prior hot spot experiments (Braga and Bond 2008; Braga et al. 1999; Weisburd and Green 1995), it is not very large compared to other experiments in criminal justice, which can have hundreds of cases (e.g., see Davis and Taylor [1999] for a review of batterer treatment experiments). With only 83 hotpots and relatively low base rates, our statistical power was limited to finding medium to large effect sizes.30 In future research of this type, researchers may need to consider using multiple cities. What this means for our project is that when differences emerge in our project that do not reach traditional levels of statistical significance (i.e., p< .05), we are left with uncertainly regarding whether our difference is due to random fluctuation or is a real difference that is not emerging as significant due to a modest sample size.

Our project is also limited to official police measures (UCR and calls-for-service) and all of the problems associated with those measures (Sherman, Buerger et al. 1989; Klinger and Bridges 1997). Given that both of these data sources measure the occurrence of crime with error they can lead to some inconsistent results. For example, in our multivariate models involving the POP intervention we found statistically significant reductions in the post-period in the POP sites in the UCR data, but not in the calls-for-service. However, we believe this inconsistency emerged due to our modest sample size and related statistical power problems, for while the calls-for-service data for POP were not statistically significant they were in the predicted direction of reductions in calls-for-service. Also, in our displacement analyses we found no evidence of displacement in the UCR data, but some possible evidence of displacement through increases in calls-for-service in surrounding areas. We attempted to explain this finding by noting that perhaps people in surrounding areas are aware of the intervention and start calling the police more. While we believe that both of our explanations are credible, like many other hot spot experiments that use police data we might also be facing a measurement error problem that is leading to the lack of consistency in some of our results. One possible way for researchers in future work to address this concern would be to develop community survey-based measures of changes in crime levels associated with hot spot interventions. However, high-quality community surveys are very expensive to conduct and can also include measurement error (e.g., recall or telescoping problems).

Conclusions

Overall, we believe that we have good evidence that our interventions were implemented faithfully and with a high level of intensity. Our major finding is that the use of POP was associated with a 33% reduction in the count of street violence during the 90-day post-period. Also, while not statistically significant, we observed that our POP group was associated with other non-trivial reductions in violence and property crime. On balance, we believe that the implementation of POP can lead to lasting violent crime reductions within violent hot spots. In contrast, we did not find any statistically significant crime reductions for the saturation/directed patrol group, though crime levels were lower for this group during the intervention period itself. As pointed out earlier, our small sample size and associated low statistical power may be masking important differences as non-significant.

The effects of POP were not apparent during the intervention period, most likely because officers needed much of that time to analyze problems and implement solutions. It is also possible that reporting effects obscured program impacts during the intervention period. After the POP projects were completed, however, their effects became more apparent. If the problem diagnoses and solutions were sound, the benefits of the POP projects at the hot spots may have lasted well beyond the study period, and perhaps indefinitely, though we cannot address that issue with certainty (our study was limited to only a 90-day follow-up period).

The estimated effects of POP on violence were generally within the range of those found in Braga et al.’s (1999) study of POP at violent places in Jersey City. In that study, which involved a much longer intervention, effects on robbery and assault ranged from 9 to 47% (not all of which were statistically significant). Most notably, they found a 31% reduction in reports for non-domestic assault (p =.09), which is virtually identical to that found for our measure of officially-recorded non-domestic violence.31

On the other hand, our results also suggest that some violence may have been displaced to areas just outside the hot spots. While we believe that this finding of crime displacement is most likely a reporting effect (see earlier discussion), we cannot say with certainty that the POP intervention was entirely successful in reducing violence if this displacement actually occurred. This finding is contrary to that of several other hot spots studies, but it reinforces the importance of efforts to monitor and address displacement when implementing hot spots strategies.

In explaining the null findings for directed/saturation patrol, a number of factors may have relevance including reporting effects and undetected departures from the study protocol (i.e., while we have no evidence of this, officers could have strayed from their assigned hot spots). Further, the condensed nature of our intervention in three months may have limited the effects of the intervention compared to if the intervention was operated for a longer period of six or twelve months (the more typical intervention length used in other hot spot studies).

In addition, the directed patrol schedules, while variable across hot spots, were fixed and may have become discernible to people in the area, thus facilitating temporal displacement. Dosage also varied across the locations and may have been insufficient in some cases (e.g., some locations received the patrols nearly every day, while others were patrolled only a few days per week). It is also possible that some of the violent crime hot spots In Jacksonville are unique, with more entrenched violent conditions that make them more resistant to directed patrol strategies. The lack of residual (i.e., post-intervention) effects from directed patrol is perhaps to be expected, however, particularly in comparison to POP (but see Sherman 1990).

On the other hand, studies of enforcement-oriented interventions at micro hot spots have not, on the whole, shown large or statistically significant effects on violent crime (Braga 2007: 36). Studies of directed patrol, in particular, have produced inconsistent results with respect to violence reduction at micro hot spots: in Minneapolis, 2–3 additional hours of intermittent patrol per day did not significantly reduce “hard crime” (Sherman and Weisburd 1995; also see Braga 2007: 36), while in Philadelphia, it took around-the-clock police presence to reduce violence (Lawton et al. 2005). In terms of daily dosage, the Jacksonville effort fell between these points. It was also done for fewer months (only 3 months).

Taken at face value and in context with other findings, nonetheless, our results suggest that assigning officers to specific hot spots of violence for hours at a time is not an optimal approach for reducing serious crime. Other research suggests, for example, that a better patrol strategy may be to assign officers to several hot spots and have them make 10–15 min stops at these locations throughout their shift, preferably on a random basis (Koper 1995).32

On the positive side, our study provides evidence that POP can be an effective strategy in reducing violence at hot spots—and one that can produce more lasting effects—though police need to be aware of the potential for displacement or reporting effects in nearby areas. Additionally, there is still much evidence supporting the efficacy of hot spots policing. However, our findings suggest that further testing and refinement of hot spots strategies should continue.

Footnotes
1

Practitioners use the term hot spot somewhat more loosely, often referring to both small places and larger areas like neighborhoods and patrol beats (Koper 2008; also see Police Executive Research Forum 2008).

 
2

Though see Rosenbaum (2006) and Taylor (2009) for critiques of hot spots policing.

 
3

However, this does not tell us whether violent places are more likely to be clustered near one another, which is another important consideration (Groff et al. 2010).

 
4

As we discuss below, POP is a strategy that calls for police to analyze and address underlying issues that contribute to chronic crime and disorder problems.

 
5

To varying degrees, police conducted these three types of activities at all of the target hot spots. Hence, the comparisons between these categories were not based on experimental assignment.

 
6

More informal evidence on the most effective hot spots strategies comes from a recent survey with a national convenience sample of police agencies affiliated with the Police Executive Research Forum (PERF), a membership organization for police chiefs and sheriffs in medium to large jurisdictions. Respondents indicated that the strategies they use most commonly and find most effective at violent crime hot spots include directed patrol, targeting known offenders, problem solving, and community partnerships (Koper 2008; also see Police Executive Research Forum 2008). As noted earlier, however, the respondents’ used the term hot spot in a broader sense than it is used here.

 
7

Note that our discussion focuses on directed patrol studies involving micro hot spots of the sort discussed previously. See Sherman et al. (2002), for example, for a review of the broader body of research on directed patrol.

 
8

The comparison areas also experienced a drop in violence that was two-thirds as large as that in the target locations, but this change was not statistically significant. The authors also examined changes in crime at the city level but found that the program effects were not large enough to produce a statistically significant reduction in violence for the city as a whole (438).

 
9

Results were overwhelmingly positive and showed much larger effects in 45 less rigorous POP studies examined by Weisburd et al. (2010).

 
10

Some might argue that hot spots policing, in any form, is a variety of POP in that it involves targeted responses based on crime analysis.

 
11

The length of the intervention period was reported in Braga (2007: 29).

 
12

U.S. Census (2006). July 1, 2006 estimates. Retrieved March 10, 2010, from www.census.gov/popest/cities/tables/SUB-EST2006-01.xls.

 
13

U.S. Census (2008). Annual estimates of the population of metropolitan and micropolitan statistical areas:

April 1, 2000 to July 1, 2008. US Census Bureau. Retrieved March 10, 2010, from www.census.gov/popest/metro/tables/2008/CBSA-EST2008-01.xls.

 
14

Prior studies of hot spots policing have generally lacked specific information about officer-hours committed to the intervention, thus precluding contrasts between the intensity of those efforts and that in Jacksonville.

 
15

JSO used a mix of on-duty officers and officers on overtime.

 
16

This is based on 30 officers working the POP locations per day for a 10-h shift, 7 days a week.

 
17

We discussed the option of an “override process” as a safety valve for the JSO. That is, if a location is deemed by the Sheriff to require an intervention, then that place will receive it. Despite this option, no “overrides” were deemed necessary by the JSO.

 
18

This type of randomized block design, of allocating cases randomly within groups, minimizes the effects of variability on a study by ensuring that like cases will be compared with one another (see Fleis 1986; Lipsey 1990; Weisburd 1993).

 
19

It is worth noting that all three conditions (saturation, problem-solving, and the control group) received standard patrol services, except the control group received no other interventions beyond standard patrol services.

 
20

Based on collection of data at the area level, we did not have any missing data for these measures.

 
21

The control group hot spots were 40% residential, 37.5% business, and 22.5% mixed use. The saturation group hot spots were 24% residential, 29% business, and 47% mixed use. The POP hot spots were 36% residential, 27% business, and 37% mixed use.

 
22

The average size of the control, saturation, and POP spots were, respectively, .22 square miles, .23 square miles, and .28 square miles.

 
23

We did not use ordinary least squares (OLS) regression because of the limited distribution of our data (most hot spots had a few crimes occur and very few had more than ten occur), and the potential for violating the normality assumption of OLS regression. Also, in these types of cases OLS can yield negative predicted values and inefficient, inconsistent, and biased estimates (Long 1997).

 
24

All count models were estimated using STATA 10.1 xt commands for cross-sectional time series data.

 
25

For example, when our outcome variable is police self-initiated activity, we have independent variables for police self-initiated activity the year before the intervention and police self-initiated activity during the 90-day period immediately before the experimental period.

 
26

For instance, when our outcome variable is calls-for-service 90 days after the treatment period we have an independent variable for calls-for-service during the 90-day experimental period.

 
27

Because some of our hot spots were in close proximity to one another, CFS/incidents could occur in the buffer zones of multiple hot spots. When this occurred, each individual CFS/incident was counted against each nearby hot spot. Approximately, one-third of the hot spots had overlap in their buffer zones, but the distribution of this overlap was approximately equally across the assigned treatments. We conducted some sensitivity analyses to determine if this double-counting unduly affected our results. First, we excluded two hot spots (both assigned to POP) that had near 50% overlap in their displacement counts and found our results did not change. Second, we constructed both a percent overlap indicator and a dichotomous overlap indicator to be included in separate analyses. Neither of these variables was significant nor had an appreciable effect on the assigned treatment estimates. Finally, we tested interaction terms between the above overlap variables and assigned treatment and found that these analyses also did not change our substantive conclusions about the impact of assigned treatment on our crime impact outcomes.

 
28

We thank an anonymous peer reviewer of an earlier version of this paper for this comment.

 
29

The average number of UCR non-domestic violent incidents in the POP locations was 2.9 during the post-intervention period (see Table 1). Our model implies that this represents a 33% reduction from what would have been expected without the intervention. From this, we estimate that the POP intervention prevented an average of roughly 1.4 crimes per hot spot, or about 31 crimes across the 22 POP locations.

 
30

Using GPOWER software (Erdfelder et al. 1996), we conducted a power test examining differences of means with an ANOVA test. Detecting a drop of approximately one-third in our violence measures (a “small” standardized effect size of 0.17 based on our sample sizes and standard deviations) with an alpha level of 0.05 and 80% power would require a total sample size of 339 cases. With our actual sample size (83 cases), we have an 80% chance of detecting a medium to large effect size of .35 with an alpha level of 0.05.

 
31

We calculated these effect sizes using data reported by Braga et al. (1999: 563). They also found significant reductions of 18 to 40% in property crime.

 
32

This notion is also consistent with the more general body of studies on directed patrol, many of which suggest that directed patrol is an effective strategy for reducing crime (e.g., see review in Sherman et al. 2002). Prior directed patrol studies have generally focused on well-defined but larger areas than those examined in this study. Hence, the strategy may be more optimal when officers can cover larger numbers of hot spots over somewhat larger areas.

 

Acknowledgements

The authors thank the Jacksonville Sheriff’s Office for its strong commitment to the research project throughout the organization including the crime analysis unit, managers Matt White and Jamie Rousch, the Operation Safe Street officers, and other JSO commanders. Also, we are very appreciative of former PERF research fellows Rachel Bambery of the New Zealand Police Service and Sergeant Jeff Egge of the Minneapolis Police Department for their assistance with different stages of the project.

Copyright information

© Springer Science+Business Media B.V. 2010