Gun violence is spatially concentrated within cities in the U.S. in the most socially disadvantaged communities [1]. Black and Hispanic men suffer higher rates of gun violence compared to other minority populations [14]. The racial inequality in gun violence victimization rates are also associated with areas of concentrated disadvantage, reflecting higher spatial concentrations of poverty, unemployment, joblessness, family disruption, and geographic isolation linked to the enduring legacy of system racism in racial residential segregation and urban disinvestment [16,5]. However, poverty levels and demographics alone do not explain the high concentration of gun violence observed in certain small geographies. Even within the poorest neighborhoods the majority of blocks have no shootings in a given year [1]. The rates of gun violence increased significantly during the 2020-2021 pandemic and the increase was concentrated in neighborhoods with higher poverty levels [17]. These findings suggest that during epidemic periods of gun violence it is important to examine the subset of places with the most potential volatility in generating violence.
In this paper, we examine the extent to which the surge in shooting victimization during the pandemic in Philadelphia, New York, and Los Angeles occurred in concentrated gun violence "hot spots," and whether the relationship between gun violence in places was disparate by race and ethnicity. In this descriptive analysis we quantify the variability of shootings by place before (2016-2019) and during the pandemic (2020-2021) and how it varies by race and ethnicity of victims. In particular, we delineate how the intensity of gun violence in particular places impacted the racial and ethnic disparity in gun violence victimization rates. This analysis provides an important step for thinking about prevention approaches to reduce the burden of gun violence in cities.
We analyze open source data on shooting events from Philadelphia 1 , Los Angeles 23 , and New York 4 . Each event is associated with a date and time, along with the latitude and longitude of the location. Events without a location were removed from the analysis (652 events were removed from Philadelphia). Overall the data consists of 6,200 events in Los Angeles, 7,568 events in New York, and 9,409 events in Philadelphia across 2016-2021. Data also contains the race/ethnicity and age of the shooting victim. We focus on Black, Hispanic/Latino and white individuals due to small sample sizes of other racial/ethnic groups in the data. We merge shooting event data with American Community Survey (2015-2019) data on race/ethnicity, percent of income below the poverty line and percent unemployed at the census block group level. The block group is the lowest level of population enumeration in the census that provides demographic estimates.
We use two methods to assess the association of race, crime concentration, and the increase in gun violence during the pandemic. In the first approach, we rank census block groups by aggregate shooting incident counts during the pre-pandemic period 2016-2019. We define "hot spot" census block groups to be those in the top decile (10%) of block groups. We then fit Poisson regressions on yearly shooting incident counts per block group, disaggregated by race/ethnicity, with indicator variables for pre-pandemic shooting decile and time period (2016-2019 vs. 2020-2021). In the second approach, we measure inequality in the distribution of shootings using a Poisson-Gamma estimate of the spatial gini index of shootings in census block groups that corrects for small sample size [13]. We compare the gini index disaggregated by race/ethnicity in the pre/post pandemic time periods.
Figure 1 displays the trend in shootings by race and ethnicity over time in Philadelphia, Los Angeles and New York. There is a clear increase in shootings in 2020-2021 that was greatest for Black victims, followed by Hispanic and white victims.
Table 1 shows the distribution of shooting victim race/ethnicity relative to the general population during the pre (2016-2019) an post (2020-2021) pandemic time periods. Victimization rate per population was highest for Black individuals and second highest for Hispanic individuals. For example, in New York, 70% of shooting victims were Black, despite comprising 22% of the population. In contrast, 3% of shooting victims were white, relative to representing 32% of the population. Victimization among Hispanic individuals more closely mirrors the population. These trends were consistent before and during the COVID-19 pandemic. The patterns of increase also does not change substantially by age, which is consistent with research that shows criminal offending and victimization by age tends to be similar across time periods [6,8] Next we examine the extent to which the increase in gun violence observed during the pandemic was concentrated in gun violence "hot spots". Figure 2 displays excess shootings during 2020-2021 relative to the expected shootings from the Poisson regression (with pandemic indicator set to equal zero) in gun violence hot spots vs. lower decile census block groups. Here we observe that the gun violence increase was disproportionately concentrated in hot spots. For example, in Los Angeles there were 288 additional shootings (compared to 2016-2019 levels) where the victim was Black in the top decile, compared to 124 additional shootings where the victims was Black across deciles 1-9. Gun violence was also disproportionately concentrated in the top decile of census block groups in Philadelphia and New York, where 36% (Philadelphia) and 47% (New York) of the increase in shootings observed during the period 2020-2021 occurred in the top decile of census block groups. Further details of the Poisson regression are contained in the Appendix. Figure 3 shows a map the location of gun violence hot spots as defined by the top decile of census block groups during the pre-and post-pandemic periods. There was significant overlap of block groups in the top decile across 2016-2019 and 2020-2021, representing a 51% overlap in Philadelphia, 54% in Los Angeles, and 64% in New York. In 2020-2021, the top decile of census block groups accounted for 44% of shootings in Philadelphia, 57% of shootings in Los Angeles and 74% of shootings in New York. These shooting hot spots had greater concentrations of Black and Hispanic individuals and disproportionately more victims of the same race and ethnicity (Table 2). Table 3 displays the demographic distribution of victims in the lowest 9 deciles of census block groups ranked by shootings. The fraction of the population identifying as white is larger in these census block groups compared to the top decile. However, the fraction of shooting victims was largely Black and Hispanic.
Consistent with prior research, poverty and economic disadvantage alone do not explain the concentration of shootings during the pandemic. To illustrate this point further, we measure inequality in the distribution of shootings using a Poisson-Gamma estimate of the spatial gini index of shootings. The gini index ranges from 0 (total equality) to 1 (total inequality). In 2020-2021, the gini index of shootings was 0.6, 0.7 and 0.8 in Philadelphia, Los Angeles and New York respectively. For comparison, we also ranked census block groups by poverty and unemployment indices and computed the gini index of shootings. While the poverty index explains some percentage of the concentration of shootings (gini index of .3-.5 across cities), there remains a significant concentration of shootings unexplained by poverty that is consistent across time and cities. It is important to note that poverty likely changed in dynamic ways with the COVID pandemic that we cannot capture with census measures.
While gun violence surged in Philadelphia, New York, and Los Angeles in 2020-2021, much of this surge was confined to a small fraction of places. These findings should not be a surprise. Research going back more than a century consistently demonstrates that crime is spatially concentrated into a small share of city blocks [3,18,19,13]. We observed a concentration by census block groups, but recognize that the concentration of shootings is even greater at more micro units like street segments or addresses. It is well-known that crime and gun violence coincide with other chronic social problems such as poverty and negative health outcomes [20], producing "concentrated disadvantage" that is often correlated with race [15]. What we demonstrate here is that the concentration of gun violence victimization by race-ethnicity is multiplicative, or compounding when gun violence rates surged during the pandemic. The top census block groups of gun violence in Philadelphia, Los Angeles, and New York, already have a disproportionate number of Black and Hispanic residents, relative to those cities as a whole, exposing minorities to a higher baseline risk of victimization. Yet, even within these "hot spots," Black and Hispanic residents experience a disproportionate risk of victimization relative to white residents in those same hot spots. Gun violence is first spatially concentrated and then demographically concentrated, reflecting enduring legacies of racial inequalities in American society [16]. For example, in Philadelphia in 2020-2021, a resident of a top decile hot spot was 6.6 times more likely to be Black than white (see Table 2). Compared to the city as a whole, we expect there to be more Black victimization because there are more Black individuals living in top gun violence hot spots. Yet the victims of shootings in those same top decile hot spots were actually 13.8 times more likely to be Black than white, more than two-times greater than expected based on spatial concentration alone.
Inexplicably, empirical facts like those reported above, are often lost (or ignored) when researchers, the media or the public stop to consider what to do about gun violence in these "hot spots" [4]. Since gun violence is concentrated in space, it makes sense that police and other public safety resources should be concentrated in those areas where gun violence is the most prevalent, especially during a period of surging gun violence [18,19]. Place-based approaches in hot spots that disrupt the routine activities of individuals at risk for committing acts of gun violence include more direct deployment of police to these areas, more effective management of problematic bars, and restrictions on time when alcohol is sold at alcohol outlets [18,9,7].
The concentration of gun violence within hot spots suggests there should be a more focused effort at the delivery of police and public safety services in collaboration with community members in economically disadvantaged minority neighborhoods to reduce gun violence hot spots. Braga and Weisburd [2] suggest that addressing community problems is especially important in "minority neighborhoods where residents have long suffered from elevated crime problems and historically poor police service" (p. 5). In addition to place-based efforts that focus on disrupting routine social activities that lead to gun violence, more effort should be directed towards making structural improvements to the environments of gun violence hot spots. Research evidence shows that changing environmental aspects of places where gun violence concentrates helps to reduce serious crime and gun violence without simply displacing it to nearby areas [10]. Such changes include cleaning up vacant lots, remediating abandoned housing, and improving street lighting.
These recommendations would not be contested if we were talking about the delivery of resources that provide public safety benefits to disadvantaged communities suffering from higher rates of gun violence. Targeted delivery would be hardly controversial because the focus is the provision of benefits with few obvious downside risks. The difference with targeting efforts to reduce gun violence in hot spots is that most short-term prevention tactics are blunt, retrospective, focused on offenders, and prone to abuse of civil liberties. Here there are potential real costs that coincide with the potential benefits [12]. And those costs and benefits are often hard to link causally. Police activity in gun violence hot spots that focuses on actual criminal behavior of individuals instead of loose heuristics of suspicion can help reduce gun violence in the short-term while minimizing racial disparities in who is stopped and questioned by the police [11].
The results presented here suggest that we really need to consider the problem in two parts: (1) Where and among whom is gun violence most concentrated? and (2) What are the most effective, fair and just tools that can be brought to bear preventing firearm victimization? A place-based problem solving approach that engages the police, municipal services, and community-based organizations to identify gun violence hot spots, and target for preventative interventions that communities desire, would be a particularly useful approach to attempt. Place-based approaches to addressing public safety offer some guidance for how to reduce gun violence in hot spots and racial disparities in shooting victimization.
Tables 4-6 include estimates from a Poisson regression modelfoot_0 with robust standard errors clustered at block group level. Regressions are run separately for each racial/ethnic group and include indicator predictor variables for decile and time period. Tables 456show the number of shootings per year in the top decile compared to bottom deciles overall and by race/ethnic group in Philadelphia, New York and Los Angeles. We compare the estimated number of shootings in 2020-2021 using only the location that were in pre-pandemic (2016-2019) top decile with those in the actual top decile. The results show that the estimated number of shootings increases proportionally more in the top decile of 2020-2021, and that this increase is being driven by a higher rate of victimization for Black and Hispanic individuals. For example, there were an estimated 653 The mean age of shooting victims was 29 in Philadelphia, 31 in Los Angeles, and 25-44 in New York (the most frequent age category of victims) (Figure 5). The age distribution of victims from 2016-2019 vs 2020-2021 is consistent across racial/ethnic groups and shows only a small increase in the average age for Hispanic victims.
We compared Poisson and negative binomial models. Both models yielded similar predictions (correlation > 99.7%, mean absolute error < .05) and we used the simpler model.