When and where does violence beget violence?

Gun violence is frequently described in the language of epidemics. Yet, few quantitative studies have generated convincing evidence on the most basic question underlying the epidemic model of violence: Does violence at time t beget violence at time t + 1? With a sample of 98 of the 100 largest U.S. cities from 2014 to 2020, we employ an instrumental variable approach developed in (Jacob et al., 2007) that uses weather conditions in a given week to instrument for shootings in the same week. We find that throughout the entire period under study, shootings at week t have a negative or null effect on shootings at week t + 1 within cities. However, in years when cities went through sharp increases in gun violence, the prevalence of shootings in a given week has a strong, positive, causal effect on shootings in the following week. These results suggest that the relationship between current and subsequent violence is not static, but varies across different places and time periods. The results have implications for understanding how violence builds on itself during periods of sharp change.