[Paper Review] Is Gun Violence Contagious?

Charles Loeffler, Seth Flaxman|arXiv (Cornell University)|Nov 21, 2016

Crime Patterns and InterventionsSocial Sciences38 references17 citations

TL;DR

Using acoustic gunshot locator data and Bayesian spatiotemporal point process modeling, this study tests whether gun violence spreads contagiously across space and time in Washington, D.C. It finds only minimal diffusion—limited to 126 meters and 10 minutes—suggesting gun violence is better explained by stochastic clustering than by epidemic-like contagion, challenging the validity of contagion-based policy models.

ABSTRACT

Existing theories of gun violence predict stable spatial concentrations and contagious diffusion of gun violence into surrounding areas. Recent empirical studies have reported confirmatory evidence of such spatiotemporal diffusion of gun violence. However, existing tests cannot readily distinguish spatiotemporal clustering from spatiotemporal diffusion. This leaves as an open question whether gun violence actually is contagious or merely clusters in space and time. Compounding this problem, gun violence is subject to considerable measurement error with many nonfatal shootings going unreported to police. Using point process data from an acoustical gunshot locator system and a combination of Bayesian spatiotemporal point process modeling and space/time interaction tests, this paper demonstrates that contemporary urban gun violence does diffuse, but only slightly, suggesting that a disease model for infectious spread of gun violence is a poor fit for the geographically stable and temporally stochastic process observed.

Motivation & Objective

• To determine whether gun violence in urban areas exhibits true contagious diffusion, akin to an epidemic, or merely spatiotemporal clustering.
• To address the limitation of existing space-time interaction tests, which cannot distinguish between clustering and diffusion.
• To overcome measurement error in traditional crime data by using high-resolution acoustic gunshot locator (AGLS) data.
• To model the extent and nature of spatiotemporal dependence in gun violence using Bayesian point process methods.
• To evaluate the empirical validity of contagion-based policy interventions, such as violence interruption programs, by testing their underlying assumptions.

Proposed method

• Employed a Bayesian spatiotemporal point process model to estimate the intensity of gun violence as a function of prior shootings.
• Used space/time interaction tests, including an enhanced version of the Knox test, to detect deviations from complete spatiotemporal randomness.
• Applied a Hawkes process model to quantify the probability that one shooting triggers subsequent shootings within a defined space-time window.
• Calibrated the model using real-time gunshot detection data from Washington, D.C.’s acoustic gunshot locator system (AGLS), which captures nonfatal shootings often missed in police reports.
• Classified shootings as 'triggered' if they fell within the top 13th percentile of excitatory intensity based on the model’s parameter θ.
• Separated the point process into endemic (background) and epidemic (triggered) components to isolate true diffusion from clustering.

Experimental results

Research questions

• RQ1 Does gun violence in urban areas exhibit contagious, epidemic-like diffusion across space and time, or is it better explained by non-diffusing spatial and temporal clustering?
• RQ2 To what extent can conventional space-time interaction tests distinguish between spatiotemporal clustering and true diffusion in gun violence?
• RQ3 How much of gun violence in Washington, D.C. is attributable to prior shootings, and over what spatial and temporal scales does this effect occur?
• RQ4 How does the use of high-resolution acoustic gunshot data improve the detection of gun violence diffusion compared to traditional police-reported data?
• RQ5 What are the implications of minimal diffusion for policy interventions based on the contagion model of violence?

Key findings

• Gun violence in Washington, D.C. exhibits only minimal spatiotemporal diffusion, with a maximum spatial range of 126 meters and a temporal window of 10 minutes after a prior shooting.
• Only 13% of shootings were classified as 'triggered' by previous shootings based on the model’s excitatory intensity parameter θ, indicating that the majority of gun violence is not contagious.
• The observed spatiotemporal clustering is better explained by a stochastic, non-diffusing process than by an epidemic or contagion model.
• The results suggest that most gun violence is spontaneous—resulting from arguments or failed drug transactions—rather than retaliatory or contagious.
• Policy interventions based on the contagion metaphor may be ineffective if they rely on longer response windows, as the window for intervention is extremely narrow.
• The study challenges the use of contagion models in violence interruption programs, as they may misrepresent the true dynamics of urban gun violence.