Firearm violence occurred more frequently in US cities in 2020 compared with previous years.1 Two major events of 2020 may explain this increase: enactment of containment policies to reduce the spread of COVID-19 and a national reckoning with systemic racism, including widespread protests sparked by the police killing of George Floyd.1 This study evaluated independent associations between COVID-19 containment policies and the killing of George Floyd on firearm violence in 1 US city, Philadelphia, Pennsylvania.
Methods
This study accessed data from the Philadelphia Police Department’s registry of shooting victims from January 1, 2016, through November 26, 2020.2 This registry is updated daily and includes all individuals shot and/or killed with a firearm as a result of interpersonal violence. There were no changes in data collection policies or practices in 2020. Compared with trauma center records, the police registry contains approximately twice the number of individuals shot with a firearm.3 Changes in counts of individuals shot per week were examined following 3 time points: the enactment of Philadelphia’s first COVID-19 containment policy (closure of nonessential businesses; March 16, 2020), the killing of George Floyd (May 25, 2020), and the partial lifting of containment policies (June 26, 2020).1
A time-series analysis was conducted using autoregressive integrated moving average (ARIMA) models. The dependent measure, a count of individuals shot per week, was log transformed to ensure normal distribution. Interruptions were specified the weeks of March 16, May 25, and June 26, 2020. We identified the ARIMA model by systematically testing local and 52-week seasonally lagged autoregressive, difference, and moving average terms to remove trends and account for temporal dependencies. Three transfer functions assessed associations between the interruption variables and the dependent measure, capturing abrupt permanent relationships (stepped change), gradual permanent relationships (asymptotic growth curve), and abrupt temporary relationships (immediate increase followed by gradual decline).4 We selected the best-fitting model based on ARIMA terms and transfer functions with 2-sided P < .05, autocorrelation and partial autocorrelation functions within 95% CI bands (calculated using the Bartlett formula), and low values for both the Ljung-Box Q statistic at 24 lags and the Akaike information criterion.4 Statistical models were estimated using SCA Workbench, version 6.3 (Scientific Computing Associates Corp). The Temple University institutional review board determined that this study was not human subjects research and did not require approval.
Results
During the 256 weeks included in the study, 7159 individuals were shot in Philadelphia, which breaks down to a mean of 24.9 (SD, 7.1) individuals shot per week during the 219 weeks before enactment of COVID-19 containment policies and a mean of 46.4 (SD, 13.5) individuals shot per week in the 37 weeks after containment policy enactment. The ARIMA model that best fit this time series included local autoregressive and moving average terms and a gradual permanent association beginning the week COVID-19 containment policies were enacted. The killing of George Floyd and the partial lifting of containment policies were not independently associated with any changes in shooting incidence. The Figure shows the observed and model-predicted incidence of individuals shot per week during the study period.
Discussion
These analyses provide evidence of a significant and sustained increase in firearm violence in Philadelphia following the enactment of COVID-19 containment policies. Counts of individuals shot per week continued to increase during protests following the killing of George Floyd and remained high during the partial lifting of containment policies until the end of the study period.
Limitations of this study include that it was restricted to 1 US city and may not be generalizable to other contexts. While Philadelphia police data are likely more complete than other available registries, the data do not include self-inflicted shootings; therefore, no conclusions may be reached about these types of shootings.3,5 Moreover, police data are not collected with the primary purpose of epidemiologic surveillance, and missing or inaccurate reports could bias the results presented.
Social and economic distress may account for the observed associations. In Philadelphia, firearm violence incidence has been empirically tied to poverty and structural place-based economic disinvestment.6 The sustained nature of the increase in firearm violence observed in this study may be related to longer-term effects of COVID-19 containment policies, including intensifying unemployment and poverty, particularly in lower-income Philadelphia communities where shootings are most concentrated.
