Using Social Networks to Predict Gun Violence in Chicago

• Bookmarks: 41


As gun violence takes hundreds of lives each year in Chicago, academics, journalists and policy makers continue to debate the merits of viewing gun violence as a public health issue. While organizations like the American Public Health Association advocate in favor, others remain skeptical of extending the term “epidemic” beyond communicable diseases. However, researchers have discovered that modeling these events in a process similar to the spread of a blood-borne pathogen like HIV significantly improves the predictive capability of previous analyses of gun violence.

In a recent article published in JAMA, Ben Green, Thibaut Horel, and Andrew V. Papachristos analyzed social networks of individuals involved in over 10,000 incidents of violent crime in Chicago between 2006 and 2014. The researchers found that social contagion, or the spread of an effect or behavior through contact with peers, accounted for 63 percent of episodes of gun violence. That is, almost two thirds of victims of gun violence were socially connected to at least one other person in their network who was also a victim of gun violence. To measure social connection, the researchers relied on arrest records, which may in reality underestimate this effect.

On average, a perpetrator of gun violence in their data set was shot 125 days after their co-offending peer, or infector. However, the median infection time was only 83 days, which suggests that the majority of related episodes actually take place more quickly. To understand how many people were connected to a given incidence of gun violence, the researchers tracked cascades, or strings of events, where the victims had been arrested together on a previous offense. The researchers found 4,107 cascades throughout the 8-year sample. The length of each cascade varied from some instances involving only one person, to a particular cascade which involved 469 separate events.

In previous models, the researchers predicted the likelihood that an individual will perpetrate gun violence using qualities like the neighborhood someone lives in and other demographic information. This is akin to how epidemiologists model the spread of airborne diseases. By adding co-offending as a proxy for social networks, the combined model utilized in this research paired elements of social contagion and demographics, and identified over half of the individuals who were predicted to fall victim to gun violence each day.

The research restricted its focus to the largest connected component of the network, so the results might overstate the predictive ability of a larger model because it excluded gunshot subjects who were not as deeply connected within the set of co-offenders. Since the initial data set included 16,399 individuals who appeared in CPD arrest records related to gun violence, while the final sample consisted of 11,123 individuals, this raises the concern that this model only works among densely connected networks. Eleven thousand individuals still provides a large sample, so the results are interesting with regard to this subset, but perhaps not generalizable to a larger population. Additionally, the researchers are careful to point out that there are still many more questions than answers about the complex social networks of gun violence. In particular, they do not know why some individuals in the social network never became gunshot subjects.

Despite its focus on a relatively small fraction of the population, this research provides actionable insight into addressing gun violence as an epidemiologist. Green et al. advocate for violence prevention efforts that consider the complex social dynamics of gun violence. Furthermore, the researchers encourage an expansion in thinking about gun violence prevention. Instead of a framework that only predicts which individuals will commit firearm related crimes, they advocate a framework that also predicts which individuals are likely to become victims of gun violence, hoping to prevent such events instead of reacting to ones that have already occurred.

Article source: Green, Ben, Thibaut Horel, and Andrew V. Papachristos. “Modeling Contagion Through Social Networks to Explain and Predict Gunshot Violence in Chicago, 2006 to 2014.” JAMA Intern Med. (2017).

Featured photo: cc/(carlballou, photo ID: 482691137, from iStock by Getty Images)

278 views
bookmark icon