Deterrence, Firearm Arrests, and Subsequent Shootings: A Micro-Level Spatio-Temporal Analysis

Abstract

Do police firearm arrests reduce later shootings in nearby locations and in the days immediately following the arrest? This question is examined at a more detailed level than in previous work in order to better describe the spatio-temporal dynamics linking these two event types. All firearm arrests (n = 5,687) and shootings (n = 5,870) in Philadelphia from 2004 to 2007 were analyzed using a modified close-pair method. Following a firearm arrest shootings declined significantly, 28–47% up to a couple of blocks away. These significant declines, however, lasted for just a few days. Overall, results suggest police firearm suppression effects occur, may extend up to two blocks away from a firearm arrest, but also are short-lived. Potential implications for deterrence are discussed.

Keywords:

• deterrence
• firearm violence
• policing
• space–time clustering

Acknowledgments

The first author was partially funded by Grant # 2009-IJ-CX-0026 from the National Institute of Justice, Ralph B. Taylor, Principal Investigator. Materials included here represent neither the opinions nor the official policies of Temple University, the Department of Justice, the National Institute of Justice, or the Philadelphia Police Department. The authors would like to thank Alison LaLond Wyant, Evan Benshetler, Travis Taniguchi, John M. MacDonald, the editor, and three thoughtful reviewers for helpful comments on earlier drafts, and the Philadelphia Police Department for data access.

Notes

1. Although some research has highlighted the importance of sentencing enhancements as part of effective law enforcement initiatives to decrease violent crime (e.g. Rosenfeld, Fornango, & Baumer, 2005), arrests alone may still elicit a deterrent effect. It is certainly true that an arrest does not always lead to a conviction and additional sanctions, but the arrest itself arguably constitutes a type of sanction. Further, the arrest follows soonest after a shooting. Survey results of an adult population in three states revealed that most people did not differentiate in terms of the unpleasantness of being arrested, convicted, or incarcerated (Tittle, 1973 as cited in Tittle & Rowe, 1974). Arrests, therefore, with or even without additional sanctions might function as punishments for and potential deterrents to criminal activity. Additionally, the use of arrest data may be more desirable than focusing on the certainty and severity of punishment after an arrest. Kohfeld (1983, p. 460) pointed out potential criminals may know if “arrests occur in their neighborhood” but argued “it is unlikely that criminals know either clearance rates or arrest ratios.” The immediacy of a visible arrest is likely to resonate with potential criminals. The arrest creates a heightened threat of apprehension, whereas potential criminals are likely uninformed about the proportion of those arrested who are ultimately convicted, and about the sanctions arrestees receive. Although arrests may be a suitable measure, future research would be served by considering the sanctions attached to arrest, particularly in light of findings that firearm violence might be decreased as a result of sentencing enhancements (e.g. Rosenfeld et al., 2005).

2. Two patrol districts, 77 and 92, cover Philadelphia International Airport and Fairmount Park, the latter a large urban park/green space. Both of these differ significantly from the patrol districts because of land use and the lack of residential population (Philadelphia Police Department, 2007) and were therefore excluded from analyzes.

3. Knox (1964), to identify these interactions in space and time, paired each childhood leukemia case with every other childhood leukemia case; thereby N cases produce N(N – 1)/2 distinct pairs (Townsley et al., 2003). Spatial and temporal distances between all possible pairs were aggregated into a contingency table where rows and columns outlined spatial and temporal intervals. Examining 185 cases of childhood leukemia in the North of England over a 10-year period, Knox considered cases in a pair close in space if the two events were less than one kilometer apart. Cases in a pair were considered close in time if they occurred within 60 days of each other (Knox, 1964). Next, the number of pairs of childhood leukemia cases considered close spatially and temporally were compared to the expected count if the pairs had been distributed randomly in both space and time. Case pairs exceeded what would be expected under the assumption of random spatio-temporal processes generating random patterns. The non-random arrangements, he concluded, arose from contagion processes (Knox, 1964).

4. In order to determine whether the patterns observed were dependent on the cutoff points chosen additional analyses were conducted (not shown) where the spatial and temporal periods were lengthened and shortened.

5. Research has noted that as the distance intervals increase so does the area covered. Stated differently, a distance of 400 feet from an event encompasses a much smaller area than a distance of 800 feet which encompasses an area more than twice as big as the first interval (Rengert, Piquero, & Jones, 1999; Turner, 1969). The equal radius buffers result in different areas with larger areas allowing more opportunities for events to be present. This potential limitation is not problematic as the Monte Carlo simulation takes into account the base rate differences between areas that shape the observed counts by distance.

*p = .01.

6. Ratcliffe and Rengert (2008) found a significant over-representation of shooting incidents within 14 days and 400 feet of previous incidents. In contrast to the Ratcliffe and Rengert (2008) study, the current research considered the temporal order of events and also used finer temporal parameters (2 days); thus, it is not clear if a near-repeat pattern would have been discovered by Ratcliffe and Rengert using smaller time intervals or if the near-repeat patterns would be much more pronounced when they are closer in time to the initiating incident and wane as time passes.

7. Prior research has demonstrated that police and criminal responses to arrests may vary under different community conditions (Klinger, 1997; Kohfeld & Sprague, 1990; Smith, 1986). Therefore, additional analyses were conducted separately by each of the six city police divisions in Philadelphia to explore possible variation in the patterns uncovered across different sections of the city. The varying community composition and violence levels found in each division, however, did not considerably alter the results. Conducting analyses by police divisions potentially introduces the analytic concern of edge effects but performing only a city level analysis may mask variation occurring in different parts of the city. Hence, analyses are conducted across both the entire city and by police divisions. The portion of the analysis that examines whether or not results are replicated across differing parts of the city is in part dependent on the construction of the areas under study. Although the use of different boundaries within the city may result in different findings, the use of very small boundaries is not feasible as the cell frequencies are not large enough to conduct the analysis. Additionally, the use of an event-centered conception of location and multiple micro-spatial increments within these given boundaries may best capture underlying dynamics that may influence offender and police behavior, thus assuaging this concern. However, an important next step for future research is to better determine the spatial complexity of the relationships between firearms arrests and subsequent shootings by better situating the events in smaller and better defined geographies to highlight the social ecological contexts within which these events occurred.