Abstract
2. BACKGROUND FOR THE REVIEW
Patrol is typically the largest function in police agencies around the world, and the majority of officers tend to be assigned to general service duties (Bayley, 1992). Patrol officers generally spend their time responding to emergency calls for service from the public, deterring crime through their presence, and carrying out special assignments from supervisors. In recent years, it has become increasingly recognized that police agencies can have a beneficial impact on crime and disorder (Lum, Koper, & Telep, 2011; National Research Council, 2004; Telep & Weisburd, 2012; Weisburd & Eck, 2004). Police patrol officers have likely played a major role in police efforts to effectively address crime as these officers make up a substantial portion of police resources and are on the front lines responding to crime and citizen concerns on a daily basis. An important question is the extent to which increased police presence through increased police patrols impact crime and disorder. If police can deter crime through their presence, does increasing the quantity of this presence help reduce crime and disorder? Being present, of course, is not the only activity patrol officers engage in, but it a major component of patrol and one that is important to examine systematically because agencies around the world devote such extensive resources to police patrol.
Our review follows the definition of “police patrol” from Kelling et al. (1974: 2) in the Kansas City preventive patrol study (see below): “to most, police patrol is equated with the visible presence of a police officer in the community, whether on foot or in a marked automobile.” Thus, for our purposes, presence is the key component of patrol. We can also think of patrol work as generally falling into one of two broad categories: “answering assigned calls and conducting general surveillance” (Whitaker, 1982: 13). We are especially interested in the second category because the police have greater control over the time devoted to non-call presence and the ability to increase (or decrease) it. As we discuss below, this “general surveillance” can take on different forms with varying levels of focus.
We recognize this is a simplification of the complex multitude of activities that fall under patrol work. But this categorization helps make clear what is and is not within the scope of this review, as we emphasize below in our summary of the theory and prior research guiding our inclusion criteria. Not all forms of police being physically present in an area qualify as patrol presence. The concept of “general surveillance” emphasizes a crime prevention focus through police visibility. This is different than officers being present to deliver a program in schools or attend a community meeting. Our emphasis is on the role of patrol presence for immediate crime prevention purposes.
Based on the deterrence literature, crime should be reduced if police patrol can effectively increase the certainty that criminal activity will be observed and punished. One mainstay of policing since the 1930s has been random preventive patrol by automobile. Random patrol generally involves officers randomly driving around their beat or assigned area in downtime between calls for service (see Kelling & Moore, 1988). While preventive patrol has traditionally been allocated in beats or precincts, its proponents have emphasized the importance of providing the widest possible geographic coverage (Reppetto, 1976). The idea is to create a sense of omnipresence and to maximize deterrence by keeping offenders on their toes about when an officer will drive by next (see Kelling et al., 1974).
This effort to create ambiguity in the minds of offenders about when police presence will next occur is supported by recent work in the deterrence literature. Loughran and colleagues (2011: 1056) emphasize the potential benefits of changing levels of patrol, noting “With the same amount of resources, an alteration of police policy by manipulating the ambiguity of the certainty of arrest would enable them to enhance the deterrent effect they have. Our finding with respect to ambiguity suggests it would be worthwhile for law enforcement to exploit any vagueness in perceptions of the risk of punishment.” Loughran et al.'s (2011) recommendations are particularly relevant to hot spots policing (see below), but the ideas would apply to larger areas as well if the police are present with enough frequency to create ambiguity in the minds of offenders as to when the next period of presence will occur. Additionally, crime is expected to be deterred at the time officers are driving through (or sitting in) a particular area, because the certainty of punishment is highest when officers can physically view any crime or disorder activity.
Durlauf and Nagin (2011) also argue that the certainty of punishment is key to understanding the deterrent effects of police patrol. Nagin (2013) expands on this argument, noting that the police role as “sentinel” or guardian is central to a deterrence regime. If the number of police or their targeting of offenders can be heightened to a significant degree than significant crime prevention benefits are likely to be gained: “increasing the visibility of the police by hiring more officers and by allocating existing officers in ways that heighten the perceived risk of apprehension consistently seem to have substantial marginal deterrent effects” (Durlauf & Nagin, 2011: 14).
The importance of police as guardians relates to the arguments in routine activities theory (Cohen & Felson, 1979). Cohen and Felson (1979) argue that crime occurs when a motivated offender and suitable target come together in space and time in the absence of capable guardianship. Police are the best example of formal guardianship that can ideally disrupt the opportunities that could lead to criminal events. Thus, even if the number of motivated offenders and suitable targets remain constant, if police can increase their levels of capable guardianship at places, then crime could potentially be reduced.
Increasing police presence can occur in a number of ways. As noted above, a common method dating back to the 1930s is changing levels of random patrols in beats or neighborhoods. We note here that our focus in this review is not on the effectiveness of random patrol per se, but rather the effectiveness of increasing police presence. Thus, we begin our review of the literature with a discussion of random preventive patrol both because it has traditionally been a common police presence allocation method and because increasing (or decreasing) levels of random patrol has been the subject of some prior research (see below).
In the major study in this area, the Kansas City Preventive Patrol Experiment (Kelling et al., 1974)5 increasing (or decreasing) levels of preventive patrol did not have a significant impact on crime or victimization. In the Kansas City study, 15 beats in the South District were assigned to one of three conditions. In the five control beats, preventive patrol was left at the normal level of one car per beat. In the five reactive beats, preventive patrol was totally eliminated. Officers still responded to 911 calls in these beats but were instructed to not randomly patrol these areas between calls. In the five proactive beats, random patrol was increased with two or three cars assigned to each beat, thus doubling or tripling the amount of random preventive patrol. The results called into doubt the effectiveness of random patrol as a crime prevention tool. There were no significant differences across the groups in reported crime, arrests, and levels of reported victimization.
The conventional wisdom after the Kansas City study was that random patrol did not work and that increasing or decreasing police presence would not impact crime rates (see Bayley, 1994; Weisburd & Eck, 2004). Gottfredson and Hirschi (1990: 270), for example, argued that, “no evidence exists that augmentation of police forces or equipment, differential patrol strategies, or differential intensities of surveillance have an effect on crime rates.” Subsequent research examining the concentration of crime at place also called into question the logic behind random preventive patrol. Because crime is not randomly distributed across beats, but is instead highly concentrated (e.g., see Sherman, Gartin, & Buerger, 1989; Weisburd, Groff, & Yang, 2012), randomly allocating patrol resources does not seem to be the most efficient or effective way to control and prevent crime (see Telep & Weisburd, 2012). These findings on the high levels of concentration of crime at place helped influence the hot spots policing model of increased presence discussed more below.
Despite these concerns with the logic underlying the approach, random preventive patrol is routinely dismissed as an ineffective strategy that police should not be using based largely on the results of just the Kansas City study. But as Sherman and Weisburd (1995) note, the small sample of beats in the study and the low base rates for many crimes created low statistical power, which made it difficult for the evaluation to discern a significant difference between the study groups even if one had existed (see also Sherman, 1992). Larson (1975) also notes that the patrol levels across the groups may not have been as different as intended, in part because officers were entering the reactive beats frequently to respond to calls for service. Other studies examining beat-based patrol increases found more positive results in terms of crime reduction (e.g., see Press, 1971; Schnelle et al., 1977), although these studies also suffered from methodological flaws. A more systematic examination of the impact of increasing patrol in beats or large geographic areas may thus provide a stronger answer to the question of “does random preventive patrol work?” than simply citing the Kansas City study as the final answer (see Sherman, 2013).
In a recent review of systematic reviews in policing, Telep and Weisburd (2016) argued that while most police innovations in policing (see Weisburd & Braga, 2006) have been covered by an existing review, more traditional tactics in policing have received less attention. While these “standard model” tactics (Weisburd & Eck, 2004) such as random preventive patrol are generally seen as outdated, they continue to occupy a substantial portion of police time and resources and so more systematic inquiry into their effects would be worthwhile. This is not to necessarily suggest that additional studies collected in this review will demonstrate that random preventive patrol is a highly effective approach. But it also seems imprudent to argue that a tactic is ineffective largely based on the results of a single study that suffers from issues of methodology and statistical power. A systematic review offers an opportunity to carefully assess the existing evidence base for random preventive patrol and other methods of increasing police presence.
Increasing police presence is not limited to random patrols at the beat level. Increasing preventive police patrols has also been an important component in a number of interventions at smaller units of geography than the police beat. Many of these interventions have focused on crime hot spots. Crime hot spots are small units of geography with high rates of criminal activity. The specific geographic area that makes up a hot spot varies across studies, ranging from individual addresses or buildings (e.g., Sherman et al., 1989) to single street segments (i.e. both sides of a street from intersection to intersection; e.g., Sherman & Weisburd, 1995) to small groups of street segments with similar crime problems such as a drug market (e.g., Weisburd et al., 2006). As noted above, because crime is so highly concentrated, the hottest spots in a city, regardless of the specific unit of analysis chosen, are responsible for a large chunk of a jurisdiction's crime problem. The concentration of crime at place has become well established through research over the last two decades. From Sherman et al.'s (1989) groundbreaking finding that 3.5 percent of addresses in Minneapolis included 50 percent of crime calls to the police (see also Pierce, Spaar, & Briggs, 1988), to more recent work by Weisburd and colleagues in multiple cities that shows that just five percent of street segments produce 50 percent of the crime each year (Weisburd et al., 2004; Weisburd, Telep & Lawton, 2014; Weisburd & Amram, 2014; Weisburd et al., 2012), there is growing consensus that crime is highly concentrated at small micro geographic units in cities. Indeed, Weisburd et al. (2012) argue that we have to recognize a “law of crime concentrations at places.”
Hot spots policing, also sometimes referred to as place-based policing (see Weisburd, 2008), covers a range of police responses that all share in common a focus of resources on the locations where crime is highly concentrated. The original hot spots policing experiment in Minneapolis (Sherman & Weisburd, 1995) focused on trying to increase patrol levels on high crime street blocks up to three hours per day. Sherman and Weisburd (1995) used computerized mapping of crime calls to identify 110 hot spots of roughly street-block length. Police patrol was doubled on average for the experimental sites over a 10-month period. Officers in Minneapolis were not given specific instructions on what activities to engage in while present in hot spots. They simply were told to increase patrol time in the treatment hot spots. The study found that the experimental as compared with the control hot spots experienced statistically significant reductions in crime calls and observed disorder.
More recently, additional rigorous studies have focused on increasing police patrol presence in high crime or at risk places. The Sacramento Police Department, for example, used 15 minute stops by officers in a random order to increase police presence on high crime street segments (Telep, Mitchell, & Weisburd, 2014). The study was guided by recommendations by Sherman (1990) and Koper (1995) to deliver police presence in medium-length doses in an unpredictable order to maximize the deterrent effect of the police. In Sacramento, the approach was successful as treatment group hot spots had significantly fewer calls for service and Part I crime incidents than control group hot spots when comparing the three month period of the experiment in 2011 to the same period in 2010.
Ratcliffe and colleagues (2011) evaluated the impact of using foot patrol to increase patrol levels in high crime areas in Philadelphia. Results suggested significant declines in violence in the treatment hot spots compared to the control sites. The intervention was particularly effective for hot spots that reached a threshold of violence (i.e. the hottest hot spots). Other presence-based hot spots studies include Di Telia and Schargrodsky's (2004) evaluation of the impact of adding police officers to guard Jewish and Muslim buildings following a terrorist attack in Buenos Aires, and Taylor, Koper, and Woods' (2011) analysis of the impact of increased presence in lengthy crackdowns in high crime locations in Jacksonville, FL. These hot spots or micro place interventions generally show stronger evidence of effectiveness, in part because police are maximizing their deterrent ability by focusing in on the highest crime places.
Hot spots are much smaller than the police beats and neighborhoods used in many interventions designed to increase random patrols, but the logic here is similar. The goals of hot spots interventions focused on increasing police presence are to both deter criminal activity and increase levels of guardianship. As Koper (1995) argues, the deterrent ability of the police is likely enhanced in these smaller units, because potential offenders are more likely to see police and be aware of their increased presence.
In reviewing the evidence on police presence, Nagin (2013) finds overall support for using increased presence in small geographic units to reduce crime. He argues “The evidence is clear that large changes in police presence do affect crime rates. The change in presence may be the result of an unplanned event, such a terror alert that triggers a large increase in police officers in public spaces, or it may be a strategic response to a known crime problem, such as in hot spots policing deployments. In either case, crime rates are reduced in places where police presence has been materially increased” (Nagin, 2013: 42).
Not all hot spots policing studies, however, focus exclusively on increasing police presence. Indeed, a number of different approaches can be taken to dealing with crime hot spots, including nuisance abatement programs (e.g., Mazerolle, Price, & Roehl, 2000), problem-oriented policing (e.g., Braga et al., 1999; Weisburd & Green, 1995), and drug house raids (e.g., Sherman & Rogan, 1995). We are not including studies examining these other approaches to dealing with high crime places. We recognize that these other types of interventions may also involve increases in police presence. Nonetheless, in an effort to avoid problems of disentangling the effects of multiple treatments simultaneously (e.g., the SARA model of problem-oriented policing vs. increased presence), we will focus only on those hot spots studies focused entirely (or almost entirely) on increasing police presence. We also recognize that this will still lead to some overlap with the studies included in the hot spots policing systematic review by Braga, Papachristos, and Hureau (2012). Any hot spots interventions focused on the strategies listed above or other strategies not related to simply increasing police presence will not be covered by our review and as a result, we do not expect the overlap between the two reviews to be substantial. Additionally, as we noted above and discuss below, we also expect our group of eligible studies to include non-hot spots studies focused on increasing police presence in larger geographic areas, such as beats. In their moderator analysis, Braga and colleagues (2012) found that hot spots studies focused on increased police presence had a significant impact on reducing crime and disorder, although the magnitude of this impact was less than that of problem-oriented hot spots studies. As we describe below, we will code any activities that study authors note that police engaged in while increasing patrols, but the primary focus of the program must be to simply increase police presence and not to engage in any particular problem solving or enforcement activity.
Moderator analyses in this review will hopefully allow for comparisons of increased police presence in beat-based random patrol vs. hot spots policing, an important comparison for examining the relative benefit of adopting hot spots policing over a more standard patrol model. This will ideally provide important new insights to the evidence base on patrol strategies because, to date, there has been no direct comparison of random patrol to other patrol approaches like hot spots policing (Sherman, 2013). 1
We suspect that many police interventions examining increased patrol and/or presence will focus on either the police beat or a micro place (e.g., hot spot) as the unit of analysis, although police could increase patrol levels at any unit of geography from very small (e.g., a particular school or interaction) to very large (e.g., an entire jurisdiction). Our main requirements (discussed below) are that the increase in police presence be the focus of the intervention and that the evaluation examine crime and disorder outcomes using a rigorous research design.
Just as the unit of analysis for interventions that increase police presence can vary, so can the dosage or intensity of increased police presence. Dosage and intensity can be assessed by examining both the number of extra officers added to each geographic unit and the amount of time these extra officers are spending in the unit relative to comparison sites. These levels can vary considerably from study to study and could include, for example, having an extra patrol car conducting random patrols in a beat 24 hours a day (e.g., the proactive beats in Kelling et al., 1974), having a single officer present in a hot spot 24 hours a day (e.g., Lawton, Taylor, & Luongo, 2005) or having a crackdown team of multiple officers present in a hot spot for a few hours at a time (e.g., Taylor et al., 2011). To the extent possible based on information provided by authors, we will code the dosage (number of officers) and intensity level (amount of time spent) for each eligible study. We will not limit our review to any particular dosage or intensity level as long as treatment sites are receiving increased patrol compared to control sites. We plan to use a similar coding scheme regardless of the geographic unit in which the police presence occurs. To the extent possible based on information provided in eligible studies, we will note the number of extra officers and/or patrol cars added to the intervention site (whether that be a hot spot, police beat, or some other unit) and the amount of time these officers were spending each day in the designated intervention site.
Finally, in any sort of place-based police intervention, there is a risk that police actions will shift crime or disorder to other places where programs are not in place. This phenomenon is usually termed displacement, and it has been a major reason for traditional skepticism about the overall crime prevention benefits of place-based prevention efforts (see Reppetto, 1976). The assumption that displacement is an inevitable outcome of focused crime prevention efforts though has been replaced by a new assumption that displacement is seldom total and often inconsequential (see Weisburd et al., 2006). Displacement in place-based police interventions at both the micro-level (Bowers et al., 2011) and the meso-level (Telep et al., 2016) have been the focus of prior systematic reviews and both of these reviews find more evidence for a “diffusion of crime control benefits” (Clarke & Weisburd, 1994) to areas nearby than for displacement. We will collect and code data on displacement in our eligible studies (see below), but suspect our findings will be in line with these two prior Campbell reviews focused specifically on the topic.
3. OBJECTIVES OF THE REVIEW
The objective of this systematic review is to synthesize the extant empirical evidence (published and unpublished) on the effects of increased police patrol presence on crime and disorder. Specifically, this review will seek to answer the following questions: How does increased police patrol presence affect crime and disorder? Do crime and disorder effects vary based on the intensity of the police intervention? Do crime and disorder effects vary based on the geographic unit of analysis for the police intervention (e.g., hot spots vs. police beats)? Do the effects of police patrol presence vary by the types of crimes examined? Do different types of police patrol have differential influences on crime and disorder (e.g., foot patrol vs. motorized patrol)?
4. METHODOLOGY
4.1 Criteria for inclusion and exclusion of studies in the review
The eligibility criteria are as follows:
Population. Studies must involve geographic areas of a city or other jurisdiction as the unit of analysis. There must be at least one geographic area of the jurisdiction designated as the treatment/intervention area. There is no limit to the number of treatment/intervention sites in a particular jurisdiction.
Intervention. The intervention must be an instance of a police or law enforcement agency increasing police patrol presence in a specified geographic area in an attempt to address crime and/or disorder. Thus, we are not interested in the effectiveness of a particular strategy or tactic per se. The intervention must involve increasing patrol resources in a specified geographic area. The geographic area is not restricted to a particular size and could include (but is not limited to) street segments, police beats, or neighborhoods. The police intervention must be limited to (or focused almost entirely on) increasing the amount of time officers are spending in the geographic area for the purpose of immediate crime prevention. For example, a hot spots intervention focused entirely on increasing officer time in the hot spots would be included, while a problem-oriented hot spots approach would not. We will code any activities police are engaged in during times of increased patrols. Data on these activities may include, for example, information on arrests made, citations written, and pedestrian or vehicle stops made. To be eligible, the intervention must be focused primarily on increasing police presence and not on any particular enforcement activity or other strategy (e.g., problem solving, zero tolerance arrests).
Comparator conditions: A study must include a control/comparison group that did not receive the intervention. The group must be of the same geographic size as the target area (e.g., comparing a treatment beat to a comparison beat). In other words, we will not include designs in which a target beat, neighborhood, etc. is compared to the rest of the jurisdiction.
Outcomes: The study must examine at least one crime or disorder related outcome. This could include measures related to total crime or disorder or total amount of a particular crime or disorder type. Most studies will likely use official police department data to measure crime outcomes, typically data from citizen calls for service or incident reports. Some studies may use victimization data collected from interviews or surveys to measure crime outcomes. Disorder outcomes will usually be measured from some combination of calls for service, incident reports, and researcher observations of physical and social disorder.
Designs: A study much be a true experiment (randomized controlled trial) or a quasi-experiment with a comparison group
Exclusion criteria
We will exclude any studies that do not include a comparison group and only assess the impact of police presence on crime by examining outcomes before and after an intervention (pre-post studies) because of our concerns about the internal validity of these studies. Because we want to ensure we are capturing all relevant studies of the impact of increased police presence, we will collect pre-post studies, but will not include these studies in our main analysis. We may decide to conduct a separate examination of these pre-post studies depending on the number of studies we identify. We will not exclude studies on the basis of language or geographical location. Limited resources do not allow us to search in languages other than English, but we will obtain and translate non-English language studies with English abstracts or keywords that are identified in our search.
4.2 Search strategy for identification of relevant studies
Several strategies will be used to perform an exhaustive search for literature fitting the eligibility criteria. First, a keyword search will be performed on an array of online abstract databases (see lists of keywords and databases below). Second, we will review the bibliographies of past reviews of police presence. 2 Third, we will perform forward searches for works that have cited seminal studies on increasing police presence. 3 Fourth, we will review the tables of contents of leading journals in the field to ensure our searches have not missed relevant articles. 4 Fifth, we will search the publications of several research and professional agencies (see list below). Sixth, after finishing the above searches and reviewing the studies as described later, we will e-mail the list to leading scholars knowledgeable in the area of police presence. These scholars will be defined as those who authored a study that appears on our inclusion list. This is likely to identify additional studies, as these experts may be able to refer us to studies we may have missed, particularly unpublished pieces such as dissertations. Finally, we will consult with an information specialist at the outset of our review and at points along the way in order to ensure that we have used appropriate search strategies.
The following databases will be searched: Academic Search Premier (EBSCOhost) Australian Criminology Database (CINCH) Criminal Justice Abstracts Criminal Justice Periodical Index Dissertation Abstracts EconLit GEOBASE GeoBib GeoRef Global Policing Database (pending availability) Google/Google Books/Google Scholar
5
Hein Online Index to Current Urban Documents JSTOR LexisNexis Academic MedLine National Criminal Justice Reference Services (NCJRS) Abstracts PAIS International PolicyFile ProQuest PsycINFO Research Now Rutgers School of Law-Newark Gray Literature Database Social Science Abstracts (SocialSciAbs) Social Science Citation Index (ISI Web of Knowledge) Social Science Research Network (SSRN) Sociological Abstracts SocINDEX WorldCat Web of Science Worldwide Political Science Abstracts
The publications of the following groups will be searched: Center for Problem-Oriented Policing (Goldstein Award submissions, Tilley Award Submissions, Situational Crime Prevention Evaluation Database) Center for Evidence-Based Crime Policy (including the Evidence-Based Policing Matrix) Criminal Justice Press (Crime Prevention Studies, volumes 1-27) Institute for Law and Justice Justice Research and Statistics Association- State Statistical Analysis Centers (SACs) Publication Library Office of Community Oriented Policing Services (COPS Office) Police Executive Research Forum Police Foundation Rand Corporation Urban Institute Vera Institute for Justice
The following agencies' publications will be searched and the agencies will be contacted if necessary (we will use English language versions of websites when available): Australian Institute of Criminology French National Center for Scientific Research (CNRS) Danish National Police (Politi) German Federal Criminal Police Office (Bundeskriminalamt) Finnish Police (Polsi) Home Office (United Kingdom) Ministry of Justice (United Kingdom) Netherlands Institute for the Study of Crime and Law Enforcement (NSCR) Netherlands Police (Politie) New Zealand Police New Zealand Ministry of Justice Norwegian Ministry of Justice and the Police Royal Canadian Mounted Police Swedish National Council on Crime Prevention (Brå) Swedish Police Service
The websites and policing publications of the following publishers will be searched: Emerald Insight Oxford Journals Sage Publications Science Direct Taylor & Francis Wiley
The following keywords will be used to search the databases listed above (in all cases where police is listed we would also use policing and “law enforcement”):
6
Police AND “increased presence” Police AND presence AND crime Police AND “increased patrol” Police AND patrol AND crime Police AND “hot spots” Police AND “preventative patrol” Police AND “preventive patrol” Police AND “random patrol” Police AND crackdown*
Several strategies will be used to obtain full-text versions of the studies found through searches of the various abstract databases listed above. First, we will attempt to obtain full-text versions from the electronic journals available through the George Mason University, Arizona State University, and University of Cambridge library research databases. When electronic versions are not available, we will use print versions of journals available at the libraries. If the journals or books are not available at George Mason University, Arizona State University, or Cambridge, we will make use of the Washington Research Library Consortium (WRLC) and the Interlibrary Loan Office (ILL) to try to obtain the journal from the libraries of other area schools. If these methods do not work, we will contact the author(s) of the article and/or the agency that funded the research to try to get a copy of the full-text version of the study.
4.3 Description of methods used in the component studies
The studies included in this review will use methodologies that are variations of a treatment versus comparison group research design with pre- and post-test measures. Some studies may have additional follow-up comparisons. The intervention will involve an increase in police patrol or presence in a specified geographic area. The exact length of the increase, the exact dosage of the increase, and the exact geographic area will vary by study, but in each study, the target geographic areas will receive an increase in police patrol/services compared to what these areas were receiving prior to the start of the intervention. In general the outcomes will likely be drawn primarily from official data on crime and disorder. The studies will also vary in their method of assignment to treatment and comparison areas. Some will use randomized methods to assign treatment to geographic areas. Quasi-experiments in which specific areas are assigned to treatment or control by a police department and/or research team will likely be common as well.
4.4 Criteria for determination of independent findings
It is possible that some studies will report multiple findings on different outcomes and/or different samples. In the case of independent samples, the results will be treated as separate findings and all such results will be included in the analysis. An example of this would be a study reporting on the effects of increased police presence in multiple target sites. .
However, some studies could have multiple primary outcomes. In cases with multiple effect sizes representing multiple outcomes (e.g., property crime and violent crime) we will use three methods of arriving at representative scores for the studies. We expect to consider the mean, the largest, and the smallest effect size to gain both a general sense of treatment impacts as well as instances in which treatment seemed to be most or least effective (see Bowers et al., 2011; Weisburd et al., 2008). In some studies many effect sizes are reported, but the authors are primarily concerned with specific outcomes. We will code all primary outcomes identified by the authors. We plan to analyze outcomes by crime type (if possible) to avoid making comparisons across dissimilar outcomes. In the event we have multiple effect sizes from a single study for a particular crime outcome (e.g., both incident and victimization data on robbery), we will create a mean effect so that each study is only represented once per outcome in any meta-analyses. Finally, some studies may involve multiple sites within a single city or locality. Such cases will be treated as one study with sub units, and independent effect sizes for primary outcomes will again be created in the same manner as above.
4.5 Details of study coding categories
All eligible studies will be coded on a variety of criteria (including details related to them) including: Reference information (title, authors, publication etc.) Nature and description of selection of site Nature and description of selection of comparison group/site or period The unit of analysis (hot spot, police beat, police district, neighborhood etc.) The sample size Methodological type (randomized experiment or quasi-experiment) A description of the police patrol presence intervention A description of any activities police engaged in during periods of increased police patrol presence. A description of the strategy for measuring displacement and diffusion including description of catchment area (when applicable) Dosage intensity and type Implementation difficulties The statistical test(s) used Pre and post outcome measure statistics in intervention area(s) and comparison area(s) (and catchment area(s) if applicable) Reports of statistical significance (if any) for both the effects of the intervention (and the displacement/diffusion effects if applicable) Effect size The conclusions drawn by the authors
A graduate research assistant at George Mason University (Sean Wire) will code each eligible study under the guidance of Dr. David Weisburd. A graduate research assistant at Arizona State University will also independently code each study under the guidance of Dr. Cody Telep. All titles and abstracts identified through our search procedure will first be screened by these graduate assistants. The full text of potentially eligible studies will be reviewed in collaboration with Dr. Weisburd and Dr. Telep. All studies meeting our eligibility criteria will be fully coded. The two coding teams will be in close communication throughout the coding process and Dr. Telep and Dr. Weisburd will monitor and review all full study coding and have biweekly calls to discuss coding progress. Where there are questions or discrepancies arise in the coding of studies across universities, all of the review authors will participate in a call to discuss and review the study and determine the final coding decision. Zotero will be used to assist in the search process and assembling of titles and abstracts. A database created in FileMaker Pro will be used to track the eligibility and coding process. A preliminary coding sheet is included in the Appendix.
4.6 Statistical procedures and conventions
Meta-analytic procedures will be used to combine data from studies. For eligible studies, with enough data present, effect sizes will be calculated in Biostat's Comprehensive Meta-Analysis Version 2.2. We will attempt to use standardized measures of effect sizes as suggested in the meta-analytic literature (e.g., Lipsey & Wilson, 2001) and plan to convert all effect sizes to Cohen's d (Rosenthal, 1994), while recognizing that our eligible studies may create some complications in effect size calculation. For small samples, we will consider using Hedges's (1981) g instead of Cohen's d to reduce bias in our effect size estimates. We will follow the approaches used by previous meta-analyses of the policing literature, especially when there is overlap between our review and the hot spots policing review (Braga et al., 2012). In instances where count-based or other regression models are used, we will use regression coefficients and standard errors to generate effect sizes. We suspect that many interventions will report primarily on before and after crime counts in treatment and comparison areas. We will use t-tests from these comparisons to calculate effect sizes when possible. When such data are not available, we will use the method suggested by Farrington and colleagues (2007) to create an odds ratio. We recognize that this is not a true odds ratio and the standard error estimates for this effect size may be biased (see Bowers et al., 2011). For interventions that report on victimization data, we anticipate these studies will utilize proportions or percentages to describe the prevalence of victimization. In this case, we could use these data to calculate a true odds ratio that could then be converted to a Cohen's d.
We will be as transparent as possible in describing how we have calculated effects for each study and will be cautious in combining effect sizes from these odds ratio calculations with effect sizes calculated from other data sources. Mean effect sizes will be computed across studies and weighted (using the inverse variance weighting procedure) to account for the greater precision of effect size estimates from larger samples. We assume a random effects model using the method of moments approach (Kelley & Kelley, 2012) for meta-analysis of effect sizes, which accounts for the diversity of interventions involving increased police presence. We will present all meta-analysis results using forest plots that include 95% confidence intervals for all estimated effect sizes.
We also hope to examine contextual or moderating features of crime and disorder effects. Though it is difficult to know at the outset, we think it important to assess differences in effects based on variables such as initiative type/type of strategy, crime type targeted, and unit of geography used for intervention. Planned moderators include whether the study was a randomized experiment or a quasi-experiment, whether the study focused on crime hot spots/micro units or larger units of geography, and whether the intervention focused on all crime or a particular crime type. Our moderator analyses will utilize the analog to the ANOVA method (see Lipsey & Wilson, 2001) for categorical moderator variables and meta-analytic regression analysis for continuous moderator variables or analyses involving multiple moderators. At the same time, we recognize that such analyses will be dependent on the number of studies that are available for inclusion in the meta-analysis.
Finally, publication bias is a concern in every meta-analysis. As such, we will use traditional methods to test for the sensitivity of the findings to publication bias in the experimental and quasi-experimental studies. These methods will include a comparison of the mean effect size for published and unpublished studies and examining a funnel plot of studies to assess whether studies are distributed symmetrically around the combined mean effect size. If we have a sufficient number of eligible studies, we will conduct additional analyses available in Comprehensive Meta-Analysis including a trim-and-fill analysis (Duval & Tweedie, 2000) and Begg and Mazumdar's (1994) rank correlation test.
4.7 Treatment of qualitative research
Qualitative studies will not be included in the current review.
5. TIMEFRAME
6. PLANS FOR UPDATING THE REVIEW
The authors plan to update the review every five years.
7. ACKNOWLEDGEMENTS
We thank David Wilson, Charlotte Gill, members of the Campbell Steering Committee, and the peer reviewers for their helpful comments on earlier versions of this protocol.
8. STATEMENT CONCERNING CONFLICT OF INTEREST
Cody Telep has previously been involved in a hot spots policing study in Sacramento (Telep et al., 2014) examining the impact of increased police presence on calls for service and serious crime.
David Weisburd has previously been involved in multiple hot spots policing studies examining the effects of increased police presence on crime and disorder, including experimental interventions in Minneapolis (Sherman & Weisburd, 1995) and Sacramento (Telep et al., 2014). Sean Wire has not previously conducted any work on the effects of increased police presence on crime and disorder.
David Farrington has not previously conducted any work on the effects of increased police presence on crime and disorder.
Footnotes
10. APPENDIX: CODING SHEET
1
2
These will include Sherman, 1997; Sherman and Eck, 2002; Telep and Weisburd, 2012; Weisburd and Eck, 2004
3
The seminal pieces that will be used here include: Di Telia and Schargrodsky, 2004; Kelling et al., 1974; Sherman and Weisburd, 1995
4
These journals will include: American Journal of Police, British Journal of Criminology, Crime and Delinquency, Criminology, Criminology and Public Policy, Journal of Criminal Justice, Journal of Quantitative Criminology, Journal of Research in Crime and Delinquency, Justice Quarterly, Police Quarterly, Policing: An International Journal of Police Strategies & Management, Policing: A Journal of Policy and Practice, Police Practice and Research: An International Journal, Policing and Society.
5
We recognize that Google and Google Scholar may yield an unmanageable number of studies (“hits”) and we thus will adjust/restrict our keyword search (e.g., add words such as evaluation or research) and our document search (e.g., only search for PDFs) as necessary when querying these databases.
6
We ideally can use a more simplified police AND patrol keyword that would cover keywords 3-4 and 6-8, but we will are unsure the number of hits this more generic keyword will uncover. If the number of hits is manageable, we will use this simpler search strategy.