Abstract
The study assesses the extent to which gun assaults are clustered in space and time using crime data from Houston, Texas. The analysis examines patterns of gun assaults at the city-level as well as more localized levels in order to understand the spatial distribution of near-repeats within the city. Consistent with prior research, the city-level analysis shows significant and meaningful near-repeat patterns. The localized analysis indicates that the risk of near-repeats is not evenly distributed across space within the city, but is concentrated among a small portion of incidents and four relatively small spatial clusters. In addition, an examination of crime types, locations, and gang involvement shows slight differences between gun assaults with and without near-repeat follow-up shootings.
Keywords
Among the most straightforward, policy-relevant contributions of criminology is research that finds crime is concentrated among a relatively small number of individuals and places (Farrell and Pease 1993; Hindelang, Gottfredson, and Garofalo 1978; Sampson and Lauritsen 1994; Sherman, Gartin, and Buerger 1989). Popular criminal justice policies in the United States are grounded in this research finding, including habitual offender sentencing laws (e.g., “three-strikes” laws), hot spots policing, and police focus on repeat offenders. Successful crime prevention strategies designed around repeat burglary victimizations have been used in the United Kingdom (Chenery, Holt, and Pease 1997; Forrester, Chatterton, and Pease 1988). These strategies all share the assumption that it is efficient to focus limited justice system resources on the small groups of individuals and places that generate disproportionately large numbers of crimes.
Advancements in environmental criminology have extended research on the spatial concentrations of crime by integrating a temporal component. This research has identified repeat and near-repeat crime phenomena whereby a victimized location and its immediate surroundings face an increased risk of a subsequent crime for a short period of time. An analogy is the spread of communicable diseases whereby people exhibit symptoms soon after exposure to an agent. Shortly after “exposure” to a burglary, locations within a relatively small area are more susceptible to experiencing this same event (Johnson and Bowers 2004a).
Empirical evidence of repeat and near-repeat patterns of victimization has been generated mainly through studies of residential burglary; researchers are beginning to examine this phenomenon among gun assaults and develop theoretical explanations (Bullock and Tilley 2002; Ratcliffe and Rengert 2008). Despite nearly three decades of focused research on gun-related violence, many questions remain unanswered, including, for example, the factors most closely associated with the risk of experiencing a gun victimization. Spatial–temporal analyses of gun-related problems hold the potential to improve understandings of the nature of gun assaults and to also generate policy-relevant information that can be used to direct limited criminal justice system and crime prevention resources.
The current study utilizes a set of spatial and temporal-referenced crime data from Houston, Texas to assess the extent to which gun assaults are clustered in space and time. This large city in the south-central part of the United States provides a contrast to evidence obtained using data from Philadelphia (Ratcliffe and Rengert 2008). The analysis advances current knowledge by identifying near-repeat incidents and then distinguishing between the initiator events and the follow-up events. This allows for the measurement of differences between these events and aids in understanding the potential for interventions to prevent follow-ups. The analysis also makes a contribution by examining patterns at the city-level and more localized levels. A localized analysis allows for a description of the basic characteristics of near-repeats and their distribution within the city. Findings show there is a near-repeat phenomenon that is concentrated in four clusters within the city. There appears to be moderate overlap between concentrations of initiator and noninitiator shootings.
Literature Review
The burdens of crime fall disproportionately hard on specific locations and people. But this general description of crime concentrations does not accurately characterize the more precise nature of this phenomenon. It is not just that some people and places are uniquely susceptible to crime victimization, some are prone to multiple, repeated victimizations. It appears that small groups of repeat victims drive patterns that generate some crime hot spots (Townsley, Homel, and Chaseling 2000; Trickett et al. 1992). According to Skogan, 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 1996:3). Trickett et al. (1992) explain that “if high crime rates occur because of repeat victimization, crime prevention should correspondingly focus on preventing people who have already been victimized from being victimized again,” which is more efficient than generalized approaches because of the ability to focus on relatively small portions of “people and places” (81). Research also shows that the risks of a subsequent victimization following an initiating event are not just confined to a precise address or an individual, risks can be transmitted to surrounding locations. These subsequent crimes that occur within a relatively short time period and in a relatively close distance to the initiating event are known as near-repeats (Morgan 2001).
Evidence of repeat and near-repeat victimization phenomena comes primarily from research on burglaries outside of the United States (Bernasco 2008; Bowers and Johnson 2005; Johnson and Bowers 2004a; Sagovsky and Johnson 2007; Townsley, Homel, and Chaseling 2000, 2003), although research has started to examine patterns in the United States (Grubesic and Mack 2008; Johnson et al. 2007). 1 A focus on burglaries is justified on grounds that these offenses are well-reported, there is a clear address, burglary targets are locations which can be subjected to interventions, and burglaries are plentiful enough to study (Townsley et al. 2000). Patterns of burglary show a near-repeat phenomenon whereby nearby locations run increased risks of experiencing a burglary for a short period of time after an initiating burglary (Bernasco 2008; Bowers and Johnson 2005; Johnson et al. 2007; Johnson and Bowers 2004a; Townsley et al. 2003). Johnson and Bowers (2004a), for example, find that “a burglary event is a predictor of significantly elevated rates of burglary within 1-2 months and within a range of up to 300-400 meters of a burgled home” (250). In other words, the location faces an elevated risk of a subsequent burglary that spans a distance of 300 to 400 meters for a 1 to 2 month period of time. Table 1 shows that spatial and temporal bands for communicable risk vary across different locations.
Time Frame and Distance of the Communicable Risk Identified Through Studies of Near-Repeat Burglaries
More localized analysis of space–time clustering reveals patterns that may be hidden in spatially aggregated analyses. An analysis of burglaries in Merseyside County (UK) uncovered a significant space–time clustering within 1 to 2 months after and between 300 and 400 meters of a previously victimized house (Johnson and Bowers 2004a). However, when the burglary data from Merseyside were disaggregated into 118 wards, not all wards exhibited significant clustering (Bowers and Johnson 2005). In other words, clustering at the ward-level exhibited various spatial and temporal bands. When burglaries were measured at the street-level, houses next to a victimized location were found to be “at a substantially heightened risk relative to those located further away, particularly within one week of an initial burglary” (Bowers and Johnson 2005:67). Compared with disadvantaged areas, affluent areas in Merseyside were found to suffer more near-repeats (or have more significant space–time clustering) than disadvantaged areas. On the other hand, disadvantaged locations tended to suffer more repeat burglaries (Bowers and Johnson 2005).
Two mechanisms have been proposed to explain repeat burglary victimizations: flag and boost hypotheses (Bowers and Johnson 2004; Johnson 2008; Johnson, Summers, and Pease 2009; Pease 1998; Sagovsky and Johnson 2007; Tseloni and Pease 2003). “Flag” indicates that a place is likely to be repeatedly victimized due to static characteristics that signify the location is a suitable target (Pease 1998). For instance, a residential property may be located next to walking paths that people at risk for offending routinely use, thus making this an attractive target because of its proximity to points of entry and escape. The initial and repeated crimes may be independent but appear related due to enduring features that make the location susceptible. In contrast, the “boost” mechanism explains that an initial crime itself increases the chances of additional victimizations because the initial crime changes something about the location to increase future victimization risk. For example, after a household is successfully burglarized the offender has learned that the location is a suitable target which makes it more attractive than nearby, and even similar targets (see Johnson et al. 2009). Thus, the “boost” explanation is compatible with the possibility that an initial crime and the repeated crime against the same target are committed by the same offender. Bernasco (2008) and Johnson et al. (2009) report evidence to support the boost explanation; burglaries close in space and time are frequently the work of the same offender. Johnson (2008) reports findings from a simulation experiment that suggest that both “flag” and “boost” have explanatory power. Theoretically, it would seem easier to explain patterns of some property crimes, like burglaries and thefts, because the targets are physical, immobile objects that posses generally stable characteristics. Violent, interpersonal crimes are qualitatively different because targets and offenders both move through time and space. Thus, explaining patterns of convergence that generate spatial–temporal clustering becomes more complicated.
Near-Repeat Shootings
The practical implications of understanding the spatial–temporal patterns of property crimes can be easily extended to personal crimes, like assaults. If research documents clear patterns of risk that fit a near-repeat phenomenon then agencies are in a position to focus responses after initiating events and analysts can begin to consider the value of prospective hot spot analysis (Johnson and Bowers 2004a). Two published studies have identified the existence of repeat and near-repeat patterns in gun assaults. Bullock and Tilley (2002) studied patterns of shootings in Manchester, United Kingdom between 1997 and 2000. As expected, the data revealed strong evidence of a repeat phenomenon among individuals. They estimated the overall risk of being shot to be .01 per 1,000 population. The risk jumped to 80 per 1,000 population following a gunshot victimization and then to 180 per 1,000 among people shot two times.
Ratcliffe and Rengert (2008) examined the spatial and temporal patterns of shootings (aggravated gun assaults and gun homicides) in Philadelphia between August 2003 and September 2005. The analysis utilized a temporal parameter of 2 weeks that was based on police investigators’ expertise about the timing of retaliation and escalation. The choice of a 400-foot spatial parameter was practical because it corresponds to the length of a city block in Philadelphia. Given these parameters, the spatial–temporal patterns of gun assaults differed significantly from what was expected under the assumption that gun assaults are distributed randomly. Ratcliffe and Rengert (2008) explained that “the risk of a repeat shooting within 2 weeks and one block of a previous event was elevated 33 percent over the normal background risk of a shooting” (71). Ratcliffe and Rengert (2008) also found that the concentration of near-repeat shootings across police sectors differed from the concentrations of all shootings. In practical terms this means it may be possible to more precisely focus interventions beyond an approach that simply targets areas with the highest volume of shootings.
The Current Study
The current study has two primary purposes. The first is to determine whether a near-repeat gun assault phenomenon exists in Houston, a large city in the south-central part of the United States. Houston offers an important contrast to evidence derived from Philadelphia data. The second purpose is to describe the basic characteristics of initiator shootings and formally assess the degree to which near-repeat gun assaults cluster within the city. The study is unable to test theories that predict near-repeat crime patterns due to an inability to measure core theoretical concepts. For example, it is not possible to measure the extent to which near-repeat shootings are substantively connected through, for instance, retaliation. In addition, variables that measure the characteristics of near-repeat locations are not available. Thus, it is not possible to assess the degree to which physical decay, as an indicator of broken windows, or active drug corners are associated with near-repeat locations. The evidence provided here contributes to the developing understanding of spatial–temporal concentrations of gun violence which can be used to advance explanations and refine future theoretical tests. In addition, the evidence has the potential to make crime prevention interventions more efficient and successful.
Setting and Data
The study setting is Houston, Texas, the fourth-largest city in the United States and largest city in the state. The 2005 to 2007 American Community Survey (ACS) estimates the city has a population of 2.03 million within 600 square miles (1,600 km2). The city’s population density is 3,391 persons per square miles. Houston is the seat of Harris County and the economic center of the sixth-largest metropolitan area in the United States with a population of 5.7 million. The city’s population is about 54.7 percent White, 24.6 percent Black, 5.4 percent Asian, and 41.7 percent Hispanic or Latino of any race (U.S. Census Bureau 2007a). The setting offers a contrast to Philadelphia, where a near-repeat shooting pattern has been identified (Ratcliffe and Rengert 2008). According to the 2005 to 2007 ACS estimates, Philadelphia has a population of 1.45 million within 135 square miles (349.6 km2). The population density is substantially greater than Houston, with 10,730 persons per square miles (U.S. Census Bureau 2007b). Given population density differences, space–time clustering of crime events in Houston may differ from those in Philadelphia. According to 2007 Uniform Crime Report (UCR) data, there were 21,180 violent crimes in Philadelphia and 24,564 in Houston (Federal Bureau of Investigation 2008). 3
This project uses an official crime dataset provided by the Houston Police Department (HPD) that contains information about the date, time, and location of incidents known to police. The 6,764 shootings examined in this study include aggravated assaults with a firearm, justifiable homicides with a firearm, and murders with a firearm between January 2005 and December 2006. Table 2 presents descriptive statistics for this sample of shootings. Over 90 percent of the cases are aggravated assaults and 8 percent are criminal homicides. More than half of firearm assaults occurred at a house or apartment and just under one-third occurred in open area locations, including streets, parks, and fields. About 17 percent occurred in business locations, which includes hotels, stores, shopping malls, commercial lots, and construction sites. The data include codes to indicate whether the incident is believed to be gang-related and about 7 percent are coded this way.
Characteristics of Shooting Incidents, January 2005 to December 2006 (n = 6,764)
Analytic Plan
Citywide analysis
The analysis begins with a citywide examination of gun assaults using a revised Knox test (Knox 1963, 1964) and the near-repeat calculator tool (Ratcliffe 2008; Ratcliffe and Rengert 2008). The spatial and temporal distances between incidents are measured for each possible incident-pair to create four groups for the pairs: close in space and close in time; close in space and not close in time; not close in space and close in time; not close in space and not close in time. The Knox test is used to evaluate whether the number of incident-pairs that are both “close” in space and time is significantly larger than what is expected if shootings were randomly distributed in space and time across the entire city.
As a technique first developed in epidemiology, there are two types of Knox tests (Knox 1963, 1964; Townsley et al. 2003). Criminological research on the near-repeat phenomenon largely utilizes the original Knox test (Bowers and Johnson 2005; Johnson et al. 2007; Johnson and Bowers 2004a, 2004b; Townsley et al. 2003). In the original Knox test, researchers use discretion to select a single cutoff point for space (d) and for time (t). Hence, the original Knox test divides both space and time into two categories, from 0 to d and over d; from 0 to t and over t. Then a 2 × 2 contingency table can be formed, where each cell represents the pairs falling into the corresponding temporal and spatial categories. A limitation of the original Knox test is that it includes two time categories and two spatial categories (Townsley et al. 2003). The revised test partitions “both the space and time dimensions into a number of categories” (Knox 1963; Townsley et al. 2003:621) and may reduce biases that result from the selection of space and time cut points. The contingency table includes greater numbers of space and time categories: from 0 to d1, d1 to d2, d2 to d3, and so on; from 0 to t1, t1 to t2, t2 to t3, and so on. Hence, it is possible to examine the significance of space–time clustering given many possible combinations of spatial and temporal bands. The space–time clustering identified in the data is compared against the null-hypothesis of a random distribution of incidents.
A Monte Carlo simulation approach is used to test the null hypothesis of a random space–time distribution (Grubesic and Mack 2008; Johnson et al. 2009; Ratcliffe and Rengert 2008). After randomly permuting the event times, a new contingency table is generated. In other words, all observed events are held constant in their locations, but the times of the events are randomly shuffled. Such a permutation is repeated many times, and all derived expected values can form a distribution assuming the null hypothesis of no spatial–temporal relationship. Then it is possible to determine the number of times that the observed number of events exceeds the expected number of events (for any combination of spatial and temporal bands). For example, if 999 simulations are run and the actual number of shootings in a specific space–time bands exceeds the simulated values 990 times, the significant level is 1 – 990/(999 + 1) = .01. That is, there is significant space–time clustering in this specific space–time band at the .01 level.
Local level analysis
The analysis also examines more localized clustering of near-repeat shootings to determine whether near-repeat shootings are unevenly distributed across the city. It is based on an examination of each shooting site and takes into account the information on the timing of initial gun assaults and nearby follow-ups. The local-level analysis used here explicitly identifies which is the initiator and which is the follow-up gun assault. 4 Thus, it is possible to gain a more precise understanding of the interactions among shootings. The local-level analysis expands on Ratcliffe and Rengert’s (2008) analysis that utilized police sectors because it allows for a description of initiator and follow-up shootings, and it examines geographic clusters of events within and across police boundaries. This disaggregated analysis also responds to the possibility that a citywide analysis will mask significant patterns at local-levels.
In order to understand the characteristics of near-repeat shootings, it is necessary to place them into categories according to how they are part of the near-repeat. The analysis distinguishes between shootings that are not spatially or temporally “near” to any other shootings and those that are “near” (the determination of “near” is discussed below). Incidents not spatially and temporally near to other shootings are referred to as “noninitiators” because they have not spawned a subsequent shooting event. Incidents that are spatially and temporally near to other shootings are “near-repeats.” Within near-repeats an incident can occur first or second in time. Those that occur first are referred to as “initiators” and those that occur later in time are referred to as “follow-ups.”
The localized analysis begins by delineating a zone around each shooting, a focal shooting, to identify any spatially near gun assaults. That is, if a gun assault is within a predetermined Manhattan distance of the focal shooting (but is not on the same location of the focal shooting), this gun assault will be considered as a spatially near gun assault. Then it is determined which nearby gun assaults occurred within a specified temporal period, such as 14 days, of the focal event. Any event which occurred outside of the temporal window is discarded from additional analyses. In other words, only shootings which meet both criteria will be considered near-repeats of focal shootings. The near repeat calculator is used to conduct a point-level analysis instead of spatially aggregating to a police-defined spatial aggregate. When the near-repeats are identified the analysis can determine whether initiator shootings in near-repeats tend to cluster within areas of the city as compared to the gun assaults without near-repeats. In addition, it will be possible to describe the basic characteristics of both initiator and follow-up shootings.
Results
Citywide Analysis of Near-Repeat Gun Assaults
The near-repeat calculator developed by Ratcliffe (2008) is used to conduct the citywide analysis of shootings. This software combines the revised Knox test and Monte Carlo simulation process to detect near-repeat crime phenomenon. Between January 2005 and December 2006, 6,764 firearm assaults occurred in Houston. However, 47 incidents were missing X-Y coordinates. Thus, 6,717 incidents will be used to examine the near-repeat shooting phenomenon. 5 These incidents occurred in 5,011 locations (distinct pairs of X-Y coordinates). Among these 6,717 incidents, 4,209 (63 percent) are associated with unique locations while 2,508 (37 percent) incidents occurred in 802 location. That is, each of these 802 locations suffered more than one shooting incident, but not necessarily within a small period of time following an initiating event. 6
In determining the spatial band to use for the current analysis, it was discovered that city-block sizes are variable across Houston, a geographically large city. In the absence of an empirically or theoretically defensible spatial bandwidth, we explored different bandwidth parameters. In this analysis eight different spatial bandwidths are applied. 7 Meanwhile, 14 days is adopted as a temporal bandwidth because previous near-repeat analyses have identified this as an important temporal bandwidth (Bowers and Johnson 2005; Ratcliffe and Rengert 2008).
In near-repeat research, spatial distances are measured with different methods, such as a Euclidean distance (Bernasco 2008; Johnson and Bowers 2004a; Townsley et al. 2003), a Manhattan distance (Ratcliffe and Rengert 2008), and the number of doors apart (Bowers and Johnson 2005). A Euclidean distance is the straight line distance between two points, while a Manhattan distance is between two points following a grid-like path. A Manhattan distance approximates the actual distance required to travel between two points in a city (Chainey and Ratcliffe 2005; Ratcliffe and Rengert 2008). The current analysis conducts 16 near-repeat analyses by combining these two distance calculations with the eight spatial bandwidths. For each analysis, 999 Monte Carlo simulations are conducted. All 16 analyses identify meaningful near-repeat patterns, regardless of the spatial band and distance type used.
Table 3 presents the results of one of the 16 analyses, which uses a 400-foot spatial bandwidth, a 14-day temporal bandwidth, and Manhattan distances. This table shows the significance level and observed over mean expected frequency across all the spatial–temporal bands. The three significance levels of clustering are based on a pseudo p-value. 8 While pseudo p-values indicate whether a cell is significant, the observed over mean expected frequency represents how meaningful and important a cell is. More specifically, the value in each cell is the ratio between the number of observed space–time pairs and the average expected pairs in the corresponding spatial–temporal band. Larger values indicate greater differences between an observed risk level and the risk level determined under the assumption of space–time randomness. For example, the 1.35 value in the top left portion of the table is interpreted to mean that once a location experiences a gun assault, the chance of a second one taking place within 1 to 400 feet and within the next 14 days is 35 percent greater than if there were no discernible pattern. These findings reveal a clear near-repeat and repeat pattern of gun assaults. In addition, there is a clear spatial and temporal decaying pattern of ratios.
Observed Over Mean Expected Frequencies and Significance Levels
* p < .05.
** p < .01.
*** p < .001.
Caution must be used when interpreting the first row of Table 3 which shows a clear pattern of repeat shootings; shootings in which the location is the exact same. An examination of these repeat incidents shows that 68 percent occurred at apartment complex addresses that lack apartment unit numbers. Thus, these repeat shootings occurred at the same general address, but may have taken place at different locations within the complex. Though these incidents may be near-repeat shootings, the lack of more detailed location information does not make it possible to accurately classify them as near-repeats. We do not modify these addresses by combining them with the next spatial band because this would correct measurement error for an unknown number of cases and may introduce new measurement error. Due to the lack of precise location information within these apartment complexes, our discussion defines follow-up shootings as those which occur within 1 to 400 feet and up to 14 days after the focal shooting.
Local-Level Analysis of Near-Repeat Gun Assaults
The analysis of all shootings across the entire city identified significant and meaningful near-repeat patterns of shootings. To better understand these incidents the local analysis describes the frequency and nature of initiating and follow-up shooting events in near-repeat sets and then compares the spatial distribution of shootings with near-repeats and shootings without near-repeats to determine whether their distributions across the city are distinct. The phrase “near-repeat set” is used to describe a group of two or more shootings that are close in both space and time. Near-repeat shootings were identified using the 1 to 400-foot spatial bandwidth, 14-day temporal bandwidth, and Manhattan distances, since these parameters revealed meaningful near-repeat patterns across the city. 9 Once these near-repeat sets of incidents were identified they could be described in terms of their basic characteristics, like premise and crime type.
Counting and categorizing the number of shootings that are part of near-repeat sets is not necessarily straightforward. By definition, one incident in any near-repeat set (or a near-repeat pair) must have occurred before any subsequent follow-ups. In a near-repeat set, it is possible for an initiator to produce more than one near-repeat follow-up shooting. For instance, one shooting can be followed by two near-repeats. At the same time, an incident can also act as the follow-up event for more than one initiator. In this instance, one shooting may occur less than 400 feet from and sooner than 14 days after two previous shootings. A single incident may also act as a follow-up and an initiator shooting if there are three or more incidents in a set of near-repeats. These situations are what Townsley (2007) refers to as “near-repeat chains.” The method of categorizing and counting events as initiators and/or follow-up shootings used here is somewhat crude because detailed information about the substance of the shootings is not available. In other words, we do not actually understand whether one event is related in a substantive way to other nearby shootings. Detailed information about each incident would allow for a determination of whether incidents are connected by more than simply time and place.
Table 4 presents information about the manner in which shootings are involved in near-repeat sets. This table summarizes the number of shooting events in each of four different categories. The first category includes shootings that are not part of any near-repeat set. This category includes nearly 95 percent of shootings (n = 6,351) and shows that large crime prevention results should not be expected from programs that treat all shootings as having the same potential to generate near-repeats. Only 366 distinct shooting incidents in the current sample are part of a near-repeat shooting pair or set. Table 4 shows that 174 incidents served as an initiator and another 174 served as a follow-up. In only 18 cases could a shooting be categorized as both a follow-up to a previous shooting and as the initiator of a follow-up shooting.
Shootings in Near-Repeat Sets Categorized as Initiators and Follow-Ups
Note: Counts are determined by space–time thresholds of 14 days and 400 feet. Percentages are out of the total number of shootings classified as being part of a near-repeat set or pair (N = 366).
Table 5 reports the characteristics of the initiator shootings. Less than 3 percent of all shootings (n = 173) generated a single near-repeat and fewer than 1 percent (n = 19) were linked to more than one near-repeat follow-up. An examination of crime types, premise types, and gang involvement shows slight differences between shootings with and without near-repeats. Business locations (3.8 percent) appear slightly more likely to generate follow-ups than homes (2.7 percent) and open areas (2.6 percent), although this difference is not statistically significant. In addition, gang-linked shootings appear slightly more likely than nongang incidents to generate follow-up shootings (3.9 percent compared to 2.8 percent), but this difference is also not statistically significant.
Characteristics of Near-Repeat Initiator Shootings in Near-Repeat Sets/Pairs
Table 6 presents information on the follow-up shootings in near-repeat sets. Similar to initiator shootings, the large majority of shootings (n = 6,525, 97 percent) were not linked to an initiator. Murders (5.3 percent) and justifiable homicides (4.3 percent) appear more likely to be linked to an initiator shooting than aggravated assaults (2.5 percent). The difference between aggravated assaults and murders is statistically significant; a significance test of the difference between justifiable homicides and assaults is not possible because of low cell counts. In addition, the percentage of murders that is a follow-up (Table 6) is greater than the percentage that is an initiator (Table 5). Together this evidence suggests a possible escalation of violence for near-repeats sets. Across Tables 3 and 4 there are no other seemingly meaningful differences.
Characteristics of Follow-Up Shootings in Near-Repeat Sets/Pairs
A visual inspection of the distribution of events suggests that these initiators are not evenly distributed across the city. In order to more systematically examine this distribution the Spatial and Temporal Analysis of Crime (STAC) component of CrimeStat (Levine 2007) was used to determine if distinct clusters of initiator shootings existed. 10 STAC identified four distinct clusters of initiator shooting events (Figure 1 ). Two clusters are located in southwest Houston, a third is located in the south-central portion of the city, and a fourth, dense cluster is in the northwest. In order to understand how the spatial distribution of initiator shootings compares to the distribution of shootings without near-repeats, STAC was used to identify clusters of shootings without near-repeats. STAC identified three clusters of nonnear-repeat shootings.
Clusters of initiator shootings and non-near repeat shootings.
Table 7 presents descriptive information on the seven clusters of shootings. All seven clusters cover multiple police beats and none is contained within a single police beat boundary. Despite this, the shooting densities appear quite high, especially those that are not linked to follow-ups. This implies that, even apart from the near-repeat phenomenon, police practitioners and violence prevention specialists have the opportunity to focus their work geographically. Clusters A, B, and C contain what appears to be high numbers of shootings per square mile. Table 7 also illustrates that 87 of the initiator shootings occurred in just four clusters. Cluster 2 appears to stand out from the other groups of initiator shootings that were identified. This cluster includes 17 initiator shootings that occurred in an area that was less than one square mile. In addition, this cluster is spatially apart from all other clusters (see Figure 1). This cluster appears unique in that a cycle of retaliation may be occurring or this may be a location that frequently brings together individuals, such as gang members, who are motivated to use firearms. This cluster should be investigated in greater detail to better understand the dynamics and implications for practice.
Clusters of Initiator and Noninitiator Shooting Events
Conclusions
Studies have recently identified a near-repeat phenomenon of burglaries and shootings. A location and nearby locations remain at a statistically elevated risk for a subsequent crime during a relatively short period of time. This finding holds practical implications for focusing crime prevention work. The current study contributes to the near-repeat phenomenon literature by examining a sample of geo-referenced gun assaults from Houston, Texas. The study describes how gun assaults are clustered in space and time at both city and local levels and reports on the basic characteristics of shooting incidents in near-repeat sets.
Like previous studies, the citywide analysis in Houston determined that gun assaults are linked to an increased risk of a subsequent gun assault within a relatively small spatial band for a short period of time. More specifically, once a location experiences a shooting, the chance of a second shooting taking place within 1 to 400 feet and within the next 14 days is 35 percent greater than if there was no discernible pattern. The findings also reveal a repeat phenomenon that is being driven by shootings at apartment complexes. The patterns of results show a clear spatial and temporal decaying pattern of the near-repeat risk level.
The analysis makes an advancement by examining more localized levels in order to describe basic characteristics of near-repeat gun assaults and their distribution within the city. Even though the city-level analysis shows significant and meaningful near-repeat patterns, the risk of near-repeats appears unevenly distributed in space. The local-level analysis not only considers the proximity in time but also the timing of initial gun assaults and nearby follow-ups. Thus, we can know, for each incident, how many times it is an initiator, follow-up, or both. The sample of 6,717 gun assaults includes only 366 (5.4 percent) that are part of near-repeat sets. In other words, a seemingly small portion of all gun assaults are responsible for the significant near-repeat pattern.
An examination of crime types, premise types, and gang involvement shows slight differences between shootings with and without near-repeats. Business locations appear slightly more likely to generate follow-ups than homes and open areas. Gang-related shootings also appear slightly more likely to generate follow-up shootings. There are few differences between shootings with and without initiators. The percentage of follow-ups that is a murder is greater than assault and is greater than the percentage of murders that is an initiator. This evidence suggests a possible escalation of violence for near-repeats sets.
A visual inspection of the spatial distribution of initiator shootings shows that, as expected, these events are not evenly distributed across the city. An analysis of the spatial distribution of initiator shootings and noninitiator shootings shows the importance of distinguishing the two types of incidents. The two groups of shootings show different clustering and that the clusters are not subsumed by each other. For example, the dense cluster of 17 near-repeat initiators in the northwest part of the city is not proximate to clusters of nonnear-repeat shootings. Finding variation in near-repeat initiator shootings across the city suggests that the deployment of police and crime prevention resources should focus primarily on the small number of particularly troublesome places. A challenge for future research will be to determine why some small set of locations faces unique, elevated risks for near-repeat gun assaults. Such an understanding would have practical value and would likely promote theoretical development.
Near-repeat research is appealing because of the practical implications of findings; the potential for prospective hot spotting (Bowers, Johnson, and Pease 2004; Johnson and Bowers 2004a) is intriguing. While the citywide analysis identified a near-repeat phenomenon, the pattern is driven by only about 5 percent of gun assaults in the city. The importance of preventing gun assaults cannot be minimized, but it is also important to recognize that the volume of total gun assaults that can be prevented because they are part of near-repeat sets which is not large (Ratcliffe and Rengert 2008). In Houston, 192 follow-up shootings were identified, which is a relatively small number compared to the volume of over 6,700 total shootings during the 2005 to 2006 time period. More detailed information about near-repeat shootings may reveal patterns that permit better predictions of when and where these sets of events are likely to unfold. This understanding may be required before meaningful crime prevention results can be expected.
One aspect of the phenomenon identified is the uneven distribution of near-repeat shootings across Houston. Thus, it does not make practical sense to treat each shooting as having the same potential to generate follow-ups. Shootings with near-repeats were found to be concentrated in four distinct clusters with one clustering appearing unique (see Figure 1). This distinct location should draw the attention of service providers, including the police and prevention specialists, during the immediate aftermath of a shooting. A timely intervention consisting of community-based programs may fundamentally reduce, rather than temporarily deter, the risk for a series of shootings to take place. Chicago Ceasefire offers a promising example of a comprehensive, community-based intervention designed to reduce shootings and killings. Through street-level outreach, public education, and conflict mediation, Ceasefire emphasizes work with the communities and individuals at elevated risks for becoming involved in shootings as offenders and victims (Skogan et al. 2008).
A methodological explanation for the seemingly small number of shootings that are part of near-repeat sets concerns the 400-foot spatial parameter and the decision to only examine shootings in the 1 to 400-feet range. The 400-foot spatial parameter is smaller than the 400-meter parameter that has been used in previous research on burglaries (see Table 1). A 400-meter parameter roughly equates to 1,300 feet. Expanding the definition of a near-repeat by using parameters that include 0 to 28 days and 1 to 1,200 feet of an initiator shooting generates 1,905 distinct shootings; expanding the spatial parameter to 1,600 feet produces 2,650 distinct shootings. This illustrates that the crime prevention possibilities expand greatly when the definition of a near-repeat is widened. It will be important for researchers and practitioners to identify spatial and temporal parameters that can most effectively guide interventions. The selection of these parameters may be informed by the availability of police resources and the nature of a local context. For instance, it may make more sense to use smaller spatial parameters when population densities are relatively high.
In addition to assisting in the direction of crime prevention resources, studying near-repeat shooting patterns also has the potential to advance theoretical understandings of violence. The patterns uncovered about space and time concentrations (see Table 3 and Figure 1) show the value of examining the qualities of space that may generate outbreaks of violence. For example, Table 7 suggests that nearly 2,000 nonnear-repeat shootings are spatially concentrated in three distinct clusters. A question not addressed in this study asks about the enduring and temporary physical and social conditions that exist in these areas that make them statistically more susceptible to gun violence. 11 Identifying the conditions that exist in these locations can aid not only theoretical development but also in the design of interventions. One valuable way to begin this examination is by explicitly focusing on the most policy-relevant variables that might be at play. For example, an analysis might integrate information about where Hurricane Katrina evacuees settled in the city, which may have upset a social balance in gang turfs and drug markets. Anecdotal information obtained through conversations with Houston police officers suggested this phenomenon caused at least some set of additional violent encounters and shootings to occur.
The study suffers from some limitations that must be recognized. The study is exploratory by its nature and is unable to offer theoretical tests or explanations for near-repeat mechanisms. Even though the descriptive characteristics suggest the possibility of a retaliation process and the escalation of violence, the analysis is unable to confirm this process. This limitation is persistent in near-repeat research and may be overcome with an examination of detailed information about near-repeat cases. Characterizations of near-repeat violent incidents can be improved through a more precise determination of exactly how incidents are related. Understanding whether incidents are related because they are part of an ongoing gang or drug-related conflict or whether incidents are substantively unrelated can advance theoretical explanations and offer practical information about why incidents are clustered at a particular location. These different scenarios would likely generate different sets of crime prevention responses.
Future research should begin to examine more detailed components of near-repeat incidents through, for instance, an examination of police report narratives and interviews with police personnel knowledgeable about cases and parties involved in the incidents (see Klofas and Hipple 2006). Examining the basic details of near-repeat shooting incidents in the dataset analyzed here is beyond the scope of the current study but a description of one set of near-repeats is illustrative of the value to be gained from such an analysis. Among 6,717 shootings, there is one unique shooting which was preceded by three and followed by three. This incident is connected to the largest number of shooting incidents within 14 days and between 1 and 400 feet. The focal case is a murder with a firearm, which happened in an apartment complex in early 2006. Three aggravated gun assaults occurred 8 days, 7 days, and 2 days prior to the focal case, all in the same apartment complex. The three following aggravated gun assaults occurred 1 day, 1 day, and 12 days after the focal incident. In total, five of the seven shootings took place in the same apartment complex and two occurred in two additional, but different apartment complexes. These apartment complexes are located in a triangular shaped area near the intersection of two major highways. Additional details of these cases would permit a determination of whether and how many of these are part of a series of connected shootings, which would aid in understanding the dynamics of this localized problem.
In addition, the study is unable to identify the precise locations of some repeat incidents. For instance, several incidents in large apartment complexes are listed as occurring at the same address but no apartment unit number is available. An improved understanding of problematic, repeat locations within large apartment complexes could be gained through a detailed examination of incident reports and interviews with residents, police officers, and social workers familiar with these locations. In a broad sense, future research must begin to describe exactly how near-repeat violent events are substantively connected and offer accounts for these observed patterns. The presumption is that with this additional knowledge, prediction can be improved.
Footnotes
The authors declared no potential conflicts of interests with respect to the authorship and/or publication of this article.
The authors received no financial support for the research and/or authorship of this article.