Abstract
The spatial analysis of crime and community problems can inform police operations by revealing where resources can be most effectively deployed. Advances in understanding the spatial concentrations of crime show that some locations are repeatedly victimized and that some nearby locations are at an elevated risk for a subsequent crime during a relatively short period of time. These are known as repeat and near-repeat phenomena. Police may be able to have a strong preventive impact on crime if these risk patterns can be identified and disrupted. This analysis reports on whether a specialized, proactive patrol unit deployed to high-crime areas was effective in disrupting repeat and near-repeat patterns of shootings. Results suggest the proactive unit did not disrupt concentrations of shootings in a meaningful way. To improve effectiveness, police practitioners and researchers should seek to understand the factors driving these patterns and then design specific interventions to address them.
In response to problems of gun-related violence, police have partnered with a variety of agencies to launch innovative programs (see, for instance, Kennedy, Piehl, & Braga, 1996; McGarrell, Chermak, Weiss, & Wilson, 2001; Skogan et al., 2008) and have experimented with proactive efforts to seize more guns from the streets (Koper & Mayo-Wilson, 2006). Combining knowledge about the spatial concentrations of crime in hot spots with research evidence that shows police patrols can have significant effects on gun violence suggests that police are in a strong position to effectively respond to and prevent gun violence. Deploying police resources to locations where crime is unusually concentrated is referred to as “hot spots” or place-based policing (Sherman & Weisburd, 1995; Weisburd, Morris, & Ready, 2008), and evidence shows this strategy can be effective (Braga, 2001; National Research Council, 2004; but also see Maguire, 2010 and Rosenbaum, 2006).
Hot spots crime analysis can benefit from research that has integrated temporal components. This body of work has identified patterns that show locations nearby to burglaries and shootings remain at an elevated risk of a subsequent crime for a relatively short period of time (Bernasco, 2008; Bowers & Johnson, 2004; Johnson & Bowers, 2004; Ratcliffe & Rengert, 2008; Sagovsky & Johnson, 2007; Townsley, Homel, & Chaseling, 2003; Wells, Wu, & Ye, 2011). This concentration of crime in space and time is known as a near-repeat phenomenon (Morgan, 2001). Studies of near-repeat patterns identify not only locations where there is an elevated risk of crime but also the time bands when this risk is unusually high. If place-based police strategies are implemented in locations experiencing repeat and near-repeat patterns of gun violence, then police may be able to disrupt elements that create this phenomenon. This study documents patterns of near-repeat shootings in Houston, Texas before and during the deployment of a specialized crime reduction unit to high-crime police districts.
Literature Review
A well-established research finding is the distinct concentration of crime risk among small groups of people and places. One of the best known studies to show concentrations of crime at places was carried out in Minneapolis and revealed that only 3% of street addresses and intersections generated 50% of police calls for service (Sherman, Gartin, & Buerger, 1989). Similar results have been reported in a number of other settings, including Vancouver, Boston, Cleveland (Ohio), and Jersey City (New Jersey; Brantingham & Brantingham, 1999; Pierce, Spaar, & Briggs, 1988; Roncek, 2000; Sherman, 1992; Weisburd & Green, 1994, 1995; Weisburd, Maher, & Sherman, 1992).
Analyses of crime patterns have also revealed that relatively small numbers of repeat crime victims and repeat offense locations are at least partially responsible for generating some crime hot spots (Townsley, Homel, & Chaseling, 2000; Trickett et al., 1992). Skogan has reported that repeat multiple victimization is “probably the most important criminological insight of the decade” and the “pilling up of repeat multiple victimization is mostly what makes a high-crime neighborhood a high-crime neighborhood” (Brady, 1996, p. 3). Studies have also demonstrated that victimization risks can be transmitted to surrounding locations. Crimes that occur within a relatively short time period and in a relatively close distance to an initiating event are known as near-repeats (Morgan, 2001).
Analyses of burglaries have revealed near-repeat patterns (Bernasco, 2008; Bowers & Johnson, 2005; Johnson & Bowers, 2004; Johnson et al., 2007; Townsley et al., 2003). Johnson and Bowers (2004), for instance, found that the sites of burglaries, in addition to a catchment area of 300 to 400 m, faced an increased chance of a subsequent burglary for a 1-to-2-month period of time. In another example, Bernasco (2008) examined the extent to which repeat and near-repeat burglaries are carried out by the same offender or offender groups. Using data for 3,624 attempted and completed residential burglaries in The Hague and surrounding areas, he found that burglaries occurring close in space and time are much more likely to be committed by the same offenders than burglaries that are not close in space and time. High percentages of burglary pairs within 1 month and within 200 m were more likely to be committed by the same offenders.
Researchers have not subjected personal crime patterns to the same level of scrutiny as property crimes to determine the existence of repeat and near-repeat patterns. Ratcliffe and Rengert (2008) analyzed patterns of aggravated gun assaults and gun homicides in Philadelphia and identified a near-repeat phenomenon. They explain that “the risk of a repeat shooting within two weeks and one block of a previous event was elevated 33 percent over the normal background risk of a shooting” (Ratcliffe & Rengert, 2008, p. 71). In addition to identifying a significant pattern of near-repeat shootings, the study also revealed that concentrations of near-repeat gun assaults across police sectors differ from the clustering of all shootings. In practical terms, it may be possible to more precisely focus interventions beyond an approach that simply targets areas with the highest volumes of shootings. An important caveat to near-repeat patterns of shootings is that the volume of shooting incidents that could be prevented with a near-repeat focused intervention does not appear to be relatively large (Ratcliffe & Rengert, 2008; Wells, et al., 2011).
Policing Hot Spots
Police agencies have taken advantage of knowledge about significant concentrations of crime and victimization risks within the jurisdictions they serve. The practical implications are straightforward: limited police resources can be mobilized and concentrated in relatively small places that suffer from distinctly high volumes of crime (Sherman & Weisburd, 1995; Weisburd & Green, 1995). This deployment is known as hot spots policing and has become a popular innovation (Police Executive Research Forum [PERF], 2008; Weisburd & Eck, 2004; Weisburd & Lum, 2005).
Systemic reviews of hot spots policing research show empirical support for its effectiveness (Braga, 2001, 2007; Weisburd & Eck, 2004). Braga’s (2007) meta-analysis of research that estimated the effects of hot spots policing on crime and disorder revealed “a medium statistically significant mean effect size favoring the effects of hot spots policing in reducing citizen calls for service in treatment places relative to control places” (p. 19). In most instances, crime displacement has not been detected in the immediate areas that surround the places where police interventions have been used (see Hope, 1994 for an exception).
The lessons from hot spots policing have been effectively used to target gun violence across different urban landscapes (Cohen & Ludwig, 2003; McGarrell et al., 2001; Sherman & Rogan, 1995; Villaveces et al., 2000; see Koper & Mayo-Wilson, 2006 for a systematic review). These studies estimated the effects of police tactics that explicitly intended to enhance patrol presence and traffic and pedestrian stops and increase seizures of illegally possessed guns in public places. Results show that these proactive police efforts, when used to target high-crime locations and suspicious individuals, are effective at reducing gun violence. The results are relatively robust given important differences in terms of the specific police tactics that have been used, the varying sizes of target locations, different intervention dosages, and different evaluation methods.
Scholars have also pointed out the deficiencies that exist in our knowledge of hot spots policing as well as some limitations we should expect to observe when police agencies implement the strategy (Maguire, 2010; Rosenbaum, 2006). Very little systematic evidence exists about how hot spots policing is practiced when the strategy is not being formally evaluated with researcher involvement (for an exception, see PERF, 2008). It may not be reasonable to expect that police departments will implement hot spots policing in a manner that mirrors the conditions in which research has shown it to be somewhat effective and in ways that are most likely to have substantial effects on crime (Maguire, 2010; Rosenbaum, 2006). More specifically, Rosenbaum (2006) is skeptical that police will work to identify the proximate causes of problems that are concentrated in space and time and then construct unique responses that are grounded in those detailed understandings of the problem; generic, directed patrols may not lead to positive, lasting results. In addition, Taylor, Koper, and Woods (2010) cite evidence that suggests police chiefs think of hot spots policing in terms that differ from the way researchers have conceived of the intervention.
Targeted police strategies have been used successfully in the United Kingdom to specifically prevent repeat burglaries (Anderson, Chenery, & Pease, 1995; Chenery, Holt, & Pease, 1997; Forrester, Chatterton, & Pease, 1988; Forrester et al., 1990). These strategies all focus on the small groups of individuals and places that have been victimized. Under the motif of “prevention of revictimization by all locally appropriate means,” the Kirkholt Project was a package of measures which enhanced interagency cooperation and collaboration between neighbors (Farrell & Pease, 1993, p. 21; Forrester et al., 1988, 1990). The results show that the reduction in repeat victimizations exceeded a general decline in victimization (Forrester et al., 1988, p. 24). In contrast, the Huddersfield project incorporated proactive and directed police patrols for the purpose of reducing repeat burglaries. Three different levels of police responses (bronze, silver, and gold) were used to prevent domestic burglaries based on the frequency of prior victimizations (Anderson et al., 1995; Chenery et al., 1997). Officers patrolled around the same time as previous burglaries at least two times per week for the silver response and everyday for the gold response. Overall, the intervention appears to have been effective in reducing repeat burglaries. More specifically, the number of silver and gold responses deployed for burglary victims declined over time, officially recorded repeats declined more in Huddersfield than elsewhere, and a lower number of previous repeat victims reported multiple burglaries in Huddersfield than were reported by repeat victims elsewhere.
Most relevant to the current analysis, Ratcliffe and Rengert (2008) examined the effects of a police intervention aimed at preventing retaliations in Philadelphia. The Philadelphia Police Department implemented a mobile version of its Operation Safe Streets program, called the Priority Corners program (Lawton, Taylor, & Luongo, 2005; Ratcliffe & Rengert, 2008). Operation Safe Streets represented a crackdown on illegal behaviors in high drug-activity locations by stationing police officers in them. In their analysis of Priority Corners, Ratcliffe and Rengert (2008) did not detect a meaningful program impact on near-repeat gun assaults; the likelihood of a near-repeat shooting was unchanged during the time when the program was implemented.
In sum, scholarly findings about the concentrations of crime in space have helped inform police interventions. Hot spots policing can be an effective strategy, and its use has grown. However, in routine practice, there is good reason to suspect hot spots policing may not be implemented in a manner consistent with the way researchers have tested it (Rosenbaum, 2006) and thus may not generate results of the magnitude that research has implied. Emerging research on burglary and shooting patterns has established repeat and near-repeat phenomenon. Police and other intervention specialists may be in a position to capitalize on this emerging finding and further refine their practices, especially the tactics used within a hot spots strategy. The limited evidence suggests police have been effective in combating concentrations of burglaries (Anderson et al., 1995; Chenery et al., 1997) but less effective in disrupting near-repeat shooting patterns (Ratcliffe & Rengert, 2008).
Current Study
This study examines whether a proactive police patrol unit deployed to high-crime locations in Houston was able to disrupt patterns of near-repeat gun assaults and gun homicides. Although evidence from several cities indicates that levels of gun violence are influenced by directed police patrol work (Cohen & Ludwig, 2003; McGarrell et al., 2001; Sherman & Rogan, 1995; Villaveces et al., 2000), studies have rarely examined police effects on near-repeat patterns. The setting for the current study is Houston, Texas, the fourth largest city in the United States with a population of approximately 2.24 million in 2008. According to the 2006-2008 American Community Survey (ACS), the city’s population is about 53.8% White, 24.1% Black, 5.3% Asian, and 41.9% Hispanic or Latino of any race (U.S. Census Bureau, 2009). Violent crimes increased from 23,978 in 2005 to 24,779 in 2008 (U.S. Federal Bureau of Investigation, 2009).
The Houston Police Department (HPD) began deploying a specialized patrol unit known as the Crime Reduction Unit (CRU) to high-crime areas in the city in November 2007. The CRU consisted of 65 officers, including 6 sergeants and 1 lieutenant. Geographic deployment decisions were made through discussions between an executive assistant chief and patrol captains. Areas were selected for CRU attention when localized problems became too large for existing police resources to address. The CRU was deployed 7 days a week to targeted locations and then rotated to new locations approximately every 2 to 4 weeks.
The CRU was expected to reduce crime by proactively patrolling in marked vehicles and making arrests. Discussions with and observations of CRU officers revealed that one of the unit’s objectives was to generate large numbers of felony arrests. Felony arrests and gun seizures represented important measures of productivity for the unit, as evidenced by a posting of felony arrest and gun seizure counts on a whiteboard in the unit’s roll call room. Officers were not required to respond to calls for service and relied on several tactics, including traffic and pedestrian stops, patrols within and nearby problematic apartment complexes, and occasional raids on problematic clubs.
Theoretically, there is reason to be skeptical that CRU activities would generate meaningful reductions in violent crime. Rosenbaum (2006, p. 249) observes that, when implementing hot spots policing, departments will likely employ predictable and generalized tactics: patrols, surveillance, sweeps, and arrests. The CRU tactics were consistent with what Rosenbaum has described. Deployments and tactics did not appear to be focused on unique aspects of crime problems, like outbreaks of gun violence. Until the existing body of knowledge about place-based policing strategies expands, we cannot determine whether these tactics represent those that are commonly employed.
Despite these concerns, it is possible that the proactive, directed patrols would add to the baseline level of police presence in a meaningful way and disrupt crime patterns. The additional visible police presence and increased numbers of stops and felony arrests in high-crime areas may generate processes, including deterrence and routine activity disruption, that would reduce violent street crime. As evidence of this possibility, a survey of police agencies across 56 jurisdictions by PERF (2008, p. 4) revealed that respondents ranked directed patrols fourth out of 18 strategies in terms of perceived effectiveness in dealing with homicide/shooting hot spots. In addition, respondents most frequently cited this strategy as being the most effective in dealing with robbery hot spots and aggravated assault hot spots. This sample of police professionals believes directed patrols will effectively disrupt concentrations of violent crime.
The CRU represents a deviation from the traditional model of policing (Weisburd & Eck, 2004) but was not formally characterized as an attempt to implement “hot spots” policing. For instance, systematic, formal problem analysis was not used to identify small, problematic places and understand the unique sources of crime problems within those places. At the same time, the CRU cannot probably be characterized as a crackdown (Weisburd & Eck, 2004) because it attempted to fill a maintenance function and ensure that positive results lasted. This was accomplished by occasionally returning to locations previously worked and coordinating with other units, so an added degree of enforcement continued for a period of time after the unit moved to work in another location. Existing research has not documented the various ways that police organizations have made sense of “hot spots” policing and actually implement the strategy (Maguire, 2010; Weick, 1995), which makes it difficult to characterize the CRU. The assumption here is that the CRU represents one form of hot spots policing (PERF, 2008; Rosenbaum, 2006) because deployments were to areas experiencing elevated crime levels. This version differs from those that entail elements of problem solving (Braga & Bond, 2008; Taylor et al., 2010) and versions that have been implemented with careful monitoring for the purposes of experimental testing (e.g., Sherman & Weisburd, 1995; Weisburd & Green, 1995). The idea explored in the current analysis is that proactive patrols deployed to high-crime areas will disrupt mechanisms that generate repeat and near-repeat shooting patterns within police districts.
Data and Analysis Plan
The analysis uses an official crime data set provided by the HPD that contains information about the date, time, and location of incidents known to police. The 5,717 events examined in this study include aggravated assaults with a firearm, justifiable homicides with a firearm, and murders with a firearm that occurred between January 2007 and August 2008. We refer to these crimes as gun assaults and shootings. Two periods are examined: the 10-month period before CRU deployment (January 2007-October 2007) and the 10-month CRU deployment period (November 2007-August 2008). The study compares patterns of shootings before CRU deployment to patterns during the CRU deployment across the entire city. The purpose of the citywide analysis is to obtain a descriptive picture of patterns before and during the CRU deployment period; no distinct comparison location is examined. It is not possible to draw conclusions about CRU intervention from the citywide analysis because no comparison group is examined.
The second stage of the analysis examines pre- and during-intervention patterns of shootings within and across 23 HPD districts. This analysis is conducted because CRU activity appears to be concentrated in some police districts. The locations of arrests made by the CRU are used as proxy indicator of areas where the CRU concentrated its work. It may be the case that the unit was able to disrupt shooting patterns within smaller portions of the city where it worked more extensively. In this analysis, we compare pre-CRU repeat and near-repeats to repeat and near-repeat patterns during CRU deployment in districts that received greater levels of CRU arrest activity and in districts that received less CRU coverage. The pre-CRU time periods and low-dose districts serve as baseline points of comparison. This does not represent the strongest type of quasi-experimental design, but the use of baseline comparisons within and across police districts allows us to generate conclusions about possible policing effects.
Comparison districts were not selected randomly, which undermines the ability to most effectively determine whether the CRU is responsible for any observed changes in shooting patterns. This is a disadvantage of the design used here. One advantage of the design, however, is that it permits a greater understanding of a strategy as it was implemented by a large police agency with little researcher control over deployment patterns or tactics. In other words, the analysis examines the intervention as it was actually implemented in a large city; it does not test an idealized version of an intervention that may or may not be applicable to actual police deployment. Evidence that shows significant reductions in near-repeat patterns in the districts that received the highest doses of CRU arrest activity would be evidence that suggests the intervention is effective. This conclusion would be strengthened if evidence showed no reductions or small reductions in repeat and near-repeat shootings in districts that received comparatively less CRU arrest activity.
The near-repeat calculator developed by Ratcliffe (2008) is used to determine whether significant repeat and near-repeat patterns exist before and during the intervention at the city and district levels. This software combines a revised Knox test (Knox, 1963, 1964) and a Monte Carlo simulation process to detect near-repeat crime phenomenon. This Knox test divides space (0-d) and time (0-t) into a number of bands, such as from 0 to d1, from d1 to d2, from d2 to d3, and more than d3 and from 0 to t1, from t1 to t2, from t2 to t3, and more than t3. Except for the last band, all of the bands have the same bandwidth. All pairs of incidents are placed into a category that combines spatial and temporal bands (e.g., between d2 to d3 and between t1 to t2). In other words, the number of incident pairs in each space-time band can be identified.
A Monte Carlo simulation is used to test statistical significance. The simulation randomly imputes the times of incidents, while holding their locations constant, because incidents are not assumed to be independent. This assumption is based on previous research that reports significant spatial clustering of crime. Each simulation generates a new value of the number of incident-pairs (simulated number) in each space-time band. With many simulations for each space-time band, all derived values form a distribution that reflects the expected distribution under a null hypothesis of no spatial-temporal relationship. This simulation makes it possible to calculate whether the number of observed events (i.e., shooting pairs) in each bandwidth differs from the simulated number of events. For example, if 999 simulations are run and the observed number in a specific space-time band exceeds the simulated numbers 989 times, the significant level is 1 – 989 / (999 + 1) = .011. That is, the pseudo p value is very low (.011) so the null hypothesis is less likely to be true. Put differently, there is significant space-time clustering in this specific space-time band at the .05 level.
Results
Citywide Analysis
Figure 1 shows a slightly increasing trend in gun assaults during these 20 months, but the number of gun assaults was declining in the 4 months prior to CRU deployment in November 2007. There were 2,883 gun assaults in the pre-CRU period and 2,834 in the CRU deployment period. However, 17 incidents before the intervention and 12 incidents during the intervention are missing x-y coordinates. 1 Thus, in the citywide analysis, 2,866 and 2,822 incidents are used to examine the repeat and near-repeat shooting patterns. Table 1 illustrates that the characteristics of pre- and during-CRU shootings are similar: more than 90% of the cases are aggravated assaults; more than half of these shootings occurred at a house or apartment and about one third occurred in open area locations; and most shootings are not believed to be domestic violence or gang related. 2
Monthly trends in firearm assaults in the pre-CRU intervention period (January 2007 through October 2007) and during the CRU intervention period (November 2007 through August 2008)
Characteristics of Shooting Incidents
The near-repeat calculator is used to, first, describe the overall repeat and near-repeat pattern and, second, to identify which specific incident is the initiator and which is the follow-up in repeat and near-repeat pairs. Researchers must set four parameters for these calculations: spatial/temporal bandwidth, the number of spatial/temporal bands, the type of spatial distance, and the number of simulations. Previous near-repeat analyses have determined that 14 days is an important temporal bandwidth, so 14 days is adopted as the temporal bandwidth in the present study (Johnson et al., 2007; Ratcliffe & Rengert, 2008). Houston is a large city with city blocks of varying sizes and no theoretically or practically defensible spatial bandwidth exists. Previous research on shootings in Philadelphia has established that 400 ft is a useful spatial parameter (Ratcliffe & Rengert, 2008).
Different methods of calculating spatial distances have been used in previous near-repeat research, including Euclidean distances (Bernasco, 2008; Johnson & Bowers, 2004; Townsley et al., 2003) and Manhattan distances (Ratcliffe & Rengert, 2008). A Euclidean distance is the distance between two points that can be measured with a ruler or a straight line. A Manhattan distance is one that exists between two points following a rectangular, grid-like path. In the present study, a Manhattan distance is used because it is closer to the actual distance needed to move between two points in a city (Chainey & Ratcliffe, 2005; Ratcliffe & Rengert, 2008).
Tables 2 and 3 present the results of the near-repeat analysis during the 10-month period of time before the CRU was deployed (Table 2) and during the 10-month period when the CRU was deployed (Table 3). Each cell presents the observed over mean expected frequency for different combinations of spatial and temporal bands. The value in each cell is the ratio of the observed number of space-time pairs to the average number of simulated space-time pairs. Larger ratios indicate greater levels of risk relative to what would be expected if incidents were distributed randomly. For example, the 1.48 value in the cell representing the same location for a 0-14 day period in Table 2 means that once a location experiences a gun assault, the chance of a second one taking place at the same location, within the next 14 days is 48% greater than if there was no discernible pattern.
Observed Over Mean Expected Frequencies and Significance Levels Before the CRU Intervention
p < .05. **p < .01. ***p < .001.
Observed Over Mean Expected Frequencies and Significance Levels During the CRU Intervention
p < .05. **p < .01. ***p < .001.
The pre- and during-CRU period experienced similar patterns of repeat shootings. These results from the citywide analysis suggest the patterns did not change in meaningful ways. For example, the first cell (same location and 0-14 days) in Tables 2 and 3 are significant, showing that the risk of a repeat shooting within 14 days of the initiating event is significantly greater than a random distribution of shootings. In addition, statistically significant near-repeat patterns during relatively brief temporal periods only emerged during the CRU deployment period. During the CRU deployment period, there was a significantly elevated risk of a near-repeat shooting within 1 to 400 ft and 0 to 14 days and 15 to 28 days of an initial shooting (Table 3). These patterns were not statistically significant in the 10-month period before the CRU was deployed (Table 2).
It can also be useful to examine the volume of incidents that comprise these event pairs. Changes may have occurred in the number of follow-up shootings or the number of initiating shootings that generate more than one near-repeat. This pattern may not necessarily be identified with the near-repeat calculator. To examine this possibility the analysis counted the number of initiating and follow-up shootings that were part of repeat and near-repeat sets. Counting these incidents is not necessarily straightforward. By definition, one incident in a near-repeat set must have occurred before any subsequent follow-up. It is possible for an initiating event to produce more than one near-repeat follow-up. For instance, one shooting can be followed by multiple near-repeats. In addition, one shooting can act as the follow-up event to more than one initiator. For instance, one shooting may occur within 200 ft and 7 days of two previous shootings. A single incident may act as a follow-up and an initiator shooting if there are three or more incidents in a set of near-repeats. A shooting is part of the repeat or near-repeat sets no matter if it acts as an initiator or follow-up.
There was a slight decrease in the number and percentage of shootings that were part of repeat or near-repeat sets during the intervention. More specifically, 226 (7.9%) shootings were part of repeat sets before the intervention and 191 (6.8%) shootings were part of repeat sets during the intervention. Although this reduction appears meaningful, the difference is not statistically significant (χ2 = 2.772, p = .10). In addition, 142 (5.0%) shootings were part of near-repeat sets before the intervention and 125 (4.4%) were part of near-repeat sets during the intervention. This difference is not statistically significant (χ2 = 0.877, p = .35).
District-Level Analysis
The spatial distribution of CRU arrests suggests the unit’s work was concentrated in some police districts more than others. Thus, it is possible that the CRU may have disrupted the near-repeat phenomenon within specific parts of the city. Such an effect may not be observed with a city-level analysis. The purpose of the district-level analysis is to determine whether repeat or near-repeat patterns changed across specific locations within the city and compare changes in districts with higher levels of CRU activity to changes in districts with lower levels of CRU activity. Table 4 presents the distribution of CRU arrests across 20 HPD police districts as a percentage of all arrests. Arrests by CRU officers represented 5% or more of all arrests in Districts 2, 3, 7, 10, 15, and 16. These are defined as high-CRU-arrest-activity districts. Six districts received moderate levels of CRU arrest activity, and eight received lower doses of arrest activity.
CRU Arrest Levels and the Risk of Repeat and Near-Repeat Before and During the CRU Intervention
There are no simulated incident-pairs in this cell, so the denominator is zero.
p < .05. **p < .01. ***p < .001.
The district-level analysis uses the same approach as the citywide analysis. First, the near-repeat calculator is used for each district in the pre- and during-CRU periods. This allows for an understanding of whether the patterns changed over time within each district and whether the patterns differed across districts. Second, the analysis counts the number of shootings that comprise repeat and near-repeat sets to understand whether the CRU may have influenced the volume of shootings.
Houston has 23 police districts, but not all are included in the current analysis. Three districts were excluded from the analysis because a small number of shootings occurred in these locations. 3 The district location is missing for 11 incidents that occurred before the CRU deployment period and 14 incidents during the CRU deployment. These 25 shooting incidents are excluded from the analysis, resulting in 2,855 incidents that occurred within these 23 districts before the intervention and 2,808 incidents during the intervention.
Each district is treated as a distinct area within which repeat and near-repeat patterns can exist. Only the shooting incidents that occurred within a specific district are analyzed. However, an incident might have occurred near the border between two districts and it is possible that, within a near-repeat pair, one incident may have occurred just over the border between districts. Without including these boundary incidents, the near-repeat calculator may not correctly identify near-repeat pairs. To overcome this possible limitation, a buffer was created around each district so the area was expanded outward by 400 ft. Within the 400-ft buffer zones, 233 incidents that occurred before the intervention and 193 during the intervention are captured. Giving a district credit for a shooting that occurred slightly outside of its boundaries means the test is conservative.
In each buffered district, near-repeat shootings were identified using the same parameters as the citywide analysis (400-ft spatial bandwidth, 14-day temporal bandwidth, and Manhattan distances). The analysis produced 40 total tables, one for each of the 20 districts for the pre-CRU and during-CRU periods. Each table has 6 spatial bands and 5 temporal bands for a total of 30 cells per table. 4 The current focus is on four spatial-temporal bands: Band 1 (repeat location within 0-14 days), Band 2 (repeat location within 15-28 days), Band 3 (1-400 ft within 0-14 days), and Band 4 (1-400 ft within 15-28 days). Bands 1 and 2 represent the repeat location pattern within the first 2 weeks of the initiator shooting (Band 1) and within the 15-to-28-day period (Band 2). Bands 3 and 4 represent the near-repeat pattern within the first 2 weeks of the initiator (Band 3) and within the 15-to-28-day period (Band 4). Table 4 presents the arrest dosage and the repeat and near-repeat patterns in the pre- and during-CRU periods for each district.
If the CRU influenced repeat and near-repeat shooting patterns, we should expect to see reductions in districts with greater levels of CRU arrest activity and no changes or smaller reductions in districts with lower levels of arrest activity. The overall pattern of results does not support the conclusion that the CRU activity is associated with changes in shooting patterns.
Significant repeat shooting patterns were observed in two high-dosage districts prior to CRU deployment (Districts 10 and 2); none of the six high-dosage districts exhibited significant repeat patterns during the CRU deployment. The same pattern of results was found for near-repeats in these six districts. The next six districts received moderate doses of CRU arrest activity, ranging from 3.22% to 4.97% of arrests. There was a significant repeat or near-repeat pattern prior to CRU deployment in only one of these districts, but four districts exhibited statistically significant patterns during CRU deployment. In District 18, a significant repeat pattern disappeared during CRU deployment. The final eight districts received comparatively less CRU arrest activity (less than 2.5% of all arrests). Contrary to expectations, two districts experienced statistically significant repeat or near-repeat shootings before CRU deployment and both became nonsignificant during CRU deployment.
Table 5 shows the number of shootings in repeat or near-repeat sets in each buffered district before and during the intervention. On average, the number of shootings in near-repeat sets within districts declined by 1.85 shootings during the intervention and the number of shootings in near-repeat sets declined by 1.35 shootings. However, these changes are not statistically significant. The number of shootings in repeat sets declined in 14 districts, increased in 5, and remained unchanged in 1 district. The number of shootings in near-repeat sets declined in 12 districts, increased in 6, and remained unchanged in 2 districts. Eight districts experienced reductions in both the number of shootings in repeats sets and the number of shootings in near-repeat sets.
The Number of Shootings Included in Repeat and Near-Repeat Sets in Each District Before and During CRU Intervention
If the CRU activity was largely responsible for changes in these types of shootings, then we should observe bigger reductions in high-activity districts than in low-activity ones. Table 5 shows that this pattern is not observed. Reductions in the volume of repeat and near-repeat shootings are not concentrated in districts with greater CRU arrest activity. To examine this relationship a bit differently, Spearman’s rank correlation coefficient is calculated. The correlation coefficient between the number of CRU arrests in a district and changes in the number of shootings in repeat sets is –.135 and the correlation for changes in the number of shootings in near-repeat sets is –.141. Greater numbers of CRU arrests are linked to greater decreases in the volume of shootings, but these relationships are not statistically significant.
Conclusions
An appeal of research on repeat and near-repeat phenomena is that it seemingly holds direct policy implications for violence prevention. Understanding when the risks of gun violence change across space and time arms police and prevention specialists with knowledge they can use to make responses more efficient and potentially more effective. An important question that must be answered is the extent to which police agencies can make use of this knowledge and craft interventions with a good chance of disrupting concentrations of shootings. Ratcliffe and Rengert (2008) found that the Priority Corners program in Philadelphia did not affect near-repeat shooting patterns. The research reported here explored the relationship between the deployment of a proactive police patrol and patterns of repeat and near-repeat shootings.
The analysis first examined citywide changes in repeat and near-repeat shootings before and during the proactive patrol unit was deployed to locations across the city. This analysis did not provide any evidence to suggest the unit disrupted repeat and near-repeat shooting patterns (see Tables 2 and 3). The next stage of the analysis examined these patterns within and across police districts to account for the fact that the unit concentrated its work in some locations more than others. This analysis uncovered no evidence to suggest there was a relationship between the CRU deployment and changes in repeat and near-repeat shootings (Tables 4 and 5). Repeat and near-repeat shooting patterns were reduced in two of six districts with the greatest relative concentration of arrest activity, however, reductions were also observed in two districts that received relatively little CRU arrest activity. This suggests there were likely mechanisms other than the CRU that can explain changes over time. The pattern in districts with moderate arrest activity was mixed, with some increases and some decreases in shooting risks. The examination of changes in overall numbers of repeat and near-repeat shootings also showed mixed evidence. The correlations between arrest activity and changes in the number of repeat and near-repeat shootings was in the expected direction but were not statistically significant.
One explanation for not finding a clear and strong effect is that the patrol unit was not specifically tasked with identifying and responding to repeat and near-repeat shootings. Rather, the unit was deployed in a less focused manner, working in locations across the city for 2-to-4-week periods of time and not necessarily attending to outbreaks of gun violence. The unit was given the difficult tasked of responding to a variety of significant crime problems. With a more precise focus on stopping such outbreaks, the unit may have been more successful (see Lum, Koper, & Taylor, 2011). Research from the United Kingdom shows that interventions designed to specifically reduce repeat and near-repeat burglaries can be effective (Anderson et al., 1995; Chenery et al., 1997). The evidence suggests police practitioners wrestling with repeat and near-repeat crime problems should tailor interventions that correspond closely with the nature of the problem.
To make the most effective use of studies that report spatial and temporal concentrations of shooting risks, research must meet several challenges. One challenge is to better identify and specify risk patterns. In both Philadelphia and Houston, repeat and near-repeat shootings represent only a small portion of all shootings. Thus, shootings have differential risks for generating follow-ups that are close in space and time. Research must identify the types of shootings that are most likely to be linked with these follow-ups. These shootings may take place in certain locations, such as near to drug markets, on the borders of gang turfs, or in criminogenic housing locations; or they may involve certain motives or elements, such as ongoing gang disputes and retaliatory violence (see Papachristos, 2009). The ability to better predict when and where follow-ups are most likely to occur will make it possible to design efficient interventions.
In addition, researchers and practitioners must also collaborate to determine the most useful parameters for defining repeat and near-repeat crimes. In this analysis, 14-day temporal and 400-ft spatial parameters were used to define near-repeats. This decision was based on limited existing research and will generate a smaller set of incidents than parameters that are larger, like 400 m (1,300 ft) that has been used in near-repeat research (Bowers & Johnson, 2005; Johnson & Bowers, 2004). Using larger spatial parameters can increase the number of repeat and near-repeat shootings that can trigger responses and then be prevented. However, this must be balanced against the need to identify practical spatial bands where additional police resources can be. For example, findings about repeat burglary patterns in Huddersfield were used to direct police resources. The United Kingdom, were used to direct police resources to specific addresses, not large numbers of city blocks (Anderson et al., 1995; Chenery et al., 1997). Research must also determine whether different spatial bands are needed to better understand interpersonal as opposed to property crimes with stationary targets.
The line of research on repeat and near-repeat crimes is primarily aimed at making policing more efficient and effective. It is clear that police are attempting to maximize efficiency by directing their crime-prevention work and responses to the people and places where problems are concentrated. A next step is to better understand the strategies, like problem solving, and tactics, like seizing more guns and targeting gangs, that are most effective for reducing problems (see Taylor et al., 2010). Generalized police approaches that cast a broad net aimed at deterring potential offenders within hot spots, like the CRU, are less likely to be effective and to generate lasting, positive results than more specifically focused responses (Braga & Bond, 2008; McGarrell et al., 2001; National Research Council, 2004; Rosenbaum, 2006; Weisburd & Eck, 2004; Weisburd et al., 2008). Scholars contend that problem-solving tactics, for example, within hot spots are poised to bring about meaningful change that will produce long-lasting effects (Braga & Bond, 2008; Rosenbaum, 2006; Weisburd et al., 2008; Weisburd & Green, 1995). Police agencies that are implementing some version of hot spots policing or that are considering this strategy would be well served to very carefully consider what officers will actually do when deployed to specific locations.
Crime analysts should explore data to identify emerging repeat and near-repeat phenomena. Finding that some street blocks face elevated risks of gun violence for relatively short periods of time is just the kind of pattern that can spawn new ideas for preventing outbreaks of violence. This analysis should not stop at the point of merely identifying times and places that face elevated risks. First, it is important to recognize that the actual numbers of incidents that are part of the repeat and near-repeat phenomenon may not be meaningful (Ratcliffe & Rengert, 2008; Wells et al., 2011). Thus, it is important for analysts to understand the magnitude of the repeat and near-repeat phenomenon. The frequency of events may vary across crime types, like shootings and burglaries of homes and vehicles. In some cases, these frequencies may be too low to warrant the devotion of resources toward a new intervention.
Second, combining together information from a variety of sources can allow crime analysts to provide a more complete understanding of the repeat or near-repeat problem (Rosenbaum, 2006). For the problem-solving process to be effective, it is necessary to identify the mechanisms that are driving repeat and near-repeat phenomenon. It may be the case that these patterns are caused by different mechanisms at work. Police are in a better position to respond when these mechanism are clearly understood.
The research described here is also important for understanding hot spots policing as it has been implemented in one large police agency. In the first 10 months of its operation, the CRU was relying on traditional police tactics, including proactive patrols, vehicle and pedestrian stops, and arrests, to generate deterrence across a variety of crime problems and contexts within the city. For the most part, one-size-fits-all tactics were being used. This approach to implementing hot spots policing may be fundamentally limited because the conditions that are primarily responsible for generating relatively dense clusters of crime problems may remain unchanged. A challenge is for researchers and practitioners to collaborate on understanding how hot spots policing is practiced and defining the challenges that inhibit the most promising practices, such as problem solving or systematic follow-up, from being adequately implemented (see Maguire, 2010).
Footnotes
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
The author(s) received no financial support for the research, authorship, and/or publication of this article.
1.
The near-repeat calculator requires that each shooting incident has at least an x-coordinate, a y-coordinate, and a date value.
2.
Business locations include hotels, stores, shopping malls, commercial lots, restaurants, and construction sites. Open area locations include streets, parks, and fields. HPD data include codes that indicate whether an offense is gang or domestic related.
3.
Districts 21 and 23 are in the Airport Division, and District 24 is in the Kingwood Division. Before the intervention, 1 shooting occurred in District 21, 1 in District 23, and 17 in District 24. During the intervention, there were no shootings in Districts 21 and 23 and 10 shootings in District 24.
4.
The six spatial bands include the same location, 1 to 400 ft, 401 to 800 ft, 801 to 1,200 ft, 1,201 to 1,600 ft, and more than 1,600 ft. The 5 temporal bands include 0 to 14 days, 15 to 28 days, 29 to 42 days, 43 to 56 days, and more than 56 days.