Neighborhood Racial and Economic Segregation and Disparities in Violence During the COVID-19 Pandemic

Abstract

Objectives. To describe associations between neighborhood racial and economic segregation and violence during the COVID-19 pandemic.

Methods. For 13 US cities, we obtained zip code–level data on 5 violence outcomes from March through July 2018 through 2020. Using negative binomial regressions and marginal contrasts, we estimated differences between quintiles of racial, economic, and racialized economic segregation using the Index of Concentration at the Extremes as a measure of neighborhood privilege (1) in 2020 and (2) relative to 2018 through 2019 (difference-in-differences).

Results. In 2020, violence was higher in less-privileged neighborhoods than in the most privileged. For example, if all zip codes were in the least privileged versus most privileged quintile of racialized economic segregation, we estimated 146.2 additional aggravated assaults (95% confidence interval = 112.4, 205.8) per zip code on average across cities. Differences over time in less-privileged zip codes were greater than differences over time in the most privileged for firearm violence, aggravated assault, and homicide.

Conclusions. Marginalized communities endure endemically high levels of violence. The events of 2020 exacerbated disparities in several forms of violence.

Public Health Implications. To reduce violence and related disparities, immediate and long-term investments in low-income neighborhoods of color are warranted. (Am J Public Health. 2022;112(1):144–153. https://doi.org/10.2105/AJPH.2021.306540)

In many places in the United States, interpersonal violence increased during the COVID-19 pandemic to levels not seen in recent history.¹ News reports suggest that this increase was unequally distributed across racial/ethnic groups,² where disparities are already stark.³

The pandemic has exacerbated social and structural conditions that contribute to violence and associated racial/ethnic inequities, including economic and housing instability,^4,5 lack of access to resources and support services,^6,7 and neighborhood social disorganization.^8,9 Simultaneously, political violence by White supremacists and violence against Black people at the hands of the state have spurred national outrage, despair, and trauma.¹⁰

Although research has demonstrated increases in violence in US cities during the COVID-19 pandemic,^1,11 no studies have, to our knowledge, documented among whom or where—in cities—the burden of violence was highest. Existing aggregate estimates likely mask substantial variation by sociodemographics and place. We are not aware of comprehensive recent data on the characteristics of individuals injured by violence, but, because of pervasive racial and socioeconomic residential segregation,¹² detailed geographic estimates of health outcomes can provide information on who is affected and social and environmental conditions that might contribute to risk.

We drew on place-based social measures to document disparities in violence during the first months of the pandemic. We constructed measures of racial, economic, and racialized economic segregation (using the Index of Concentration at the Extremes [ICE]), comparing rates of firearm violence and other violent crime (homicide, aggravated assault, robbery, and rape) in zip codes in 13 major US cities. We had 2 aims. First, we examined cross-sectional differences between zip codes in 2020 to determine where the burden of violence was highest during the first months of the pandemic. Second, we examined zip code differences in 2020 relative to such differences in previous years (i.e., difference-in-differences) to identify differential change over time between less versus more privileged zip codes, isolating the unique contribution of the pandemic context.

METHODS

We selected 13 major US cities that represent a geographic and sociopolitical range and made data on crime during our study period publicly available through open data portals: Baltimore, Maryland; Boston, Massachusetts; Chicago, Illinois; Cincinnati, Ohio; Dallas, Texas; Denver, Colorado; Detroit, Michigan; Los Angeles, California; Milwaukee, Wisconsin; Philadelphia, Pennsylvania; Phoenix, Arizona; San Francisco, California; and Seattle, Washington. The study period was March through July in 2018, 2019, and 2020.

Data

Outcomes

We examined 5 outcomes: intentional, interpersonal firearm violence (hereafter “firearm violence”) and the Federal Bureau of Investigation’s Uniform Crime Reporting Part I violent crime offenses: criminal homicide, rape, robbery, and aggravated assault.

We downloaded data on police-reported crime incidents (except for firearm violence) from open data portals (Table A, available as a supplement to the online version of this article at http://www.ajph.org for data sources). We coded crimes to maximize comparability across cities, although classifications varied somewhat (e.g., some cities reported more detailed categories, and some made exclusions; Table A). We accounted for between-city differences with city fixed effects (“Analysis” section). Zip codes were the smallest geographic unit provided in the data. We report counts of incidents rather than victims (although multiple victims may be involved in a single incident) because data were consistently reported only at the incident level.

Firearm violence data came from the Gun Violence Archive, a real-time repository for firearm violence incidents compiled from approximately 7500 news outlets and other public sources;¹³ most cities’ data portals did not include information on firearm-involved crime. Each incident record in the Gun Violence Archive includes basic descriptive information, the location, and the number of people injured or killed. We geocoded incidents to obtain latitude and longitude. We included all incidents of intentional, interpersonal firearm violence in which at least 1 person was injured or killed.

We assigned incidents of firearm violence and police-reported crimes to zip codes and summed counts from March through July in each year (2018, 2019, 2020). We excluded incidents that did not correspond to Zip Code Tabulation Areas (ZCTA; see “Exposures” section) in our cities, did not contain valid geographic information, or were in ZCTAs with no population. We excluded cities from analyses that were missing crime data or had fewer than 5 incidents annually for that outcome citywide: San Francisco and Seattle for homicide; Boston, Dallas, Denver, Detroit, Milwaukee, and Seattle for rape.

Exposures

We used the ICE to measure racial and economic spatial segregation. The ICE quantifies the extent to which individuals in a neighborhood (here, zip code) are concentrated in the extremes of the distribution of socioeconomic characteristics (here, race and income). Unlike other commonly used measures, the ICE is meaningful at the neighborhood level and provides information on the direction, not just magnitude, of spatial concentration.¹⁴ For example, the Dissimilarity Index compares a larger geographic unit (e.g., city) to smaller geographic units (e.g., zip codes), returning a single metric of segregation for the larger unit as a whole.

We used the formula ICE_i = (A_i – P_i)/T_i, where A_i, P_i, and T_i reflect the number of people in neighborhood i belonging to the most privileged group, the least privileged group, and the total population, respectively. The ICE thus ranges from −1 (all residents belong to the least privileged group) to +1 (all residents belong to the most privileged group). The ICE has been previously used in research on zip code–level health disparities.¹⁵

Similar to previous work,¹⁴ we computed 3 versions of the ICE: (1) income, comparing households with incomes of $100 000 or greater (most privileged) to households with incomes of $24 999 or less (least privileged); (2) race, comparing White people of all ethnicities (most privileged) to Black people of all ethnicities (least privileged), with race reflecting socially constructed hierarchies and risk for exposure to racism;¹⁶ and (3) race–income, comparing White households with incomes of $100 000 or greater (most privileged) to Black households with incomes of $24 999 or less (least privileged). The race–income measure avoids collinearity problems that arise from including separate measures for each in 1 model.¹⁴ We adapted code from the Public Health Disparities Geocoding Project¹⁷ to obtain these data and population data from the American Community Survey (2015–2019 estimates) for ZCTAs, which are stable geographic units defined by the US Census Bureau designed to reflect US Postal Service zip code boundaries.

ZCTAs (and zip codes) do not nest neatly in cities; based on visual inspection of the geographic overlap, we excluded a ZCTA if more than 75% of its land area was outside the city. Although our outcome data sources included incidents that occurred slightly outside the city bounds, we made this restriction to minimize potential bias from unobserved missing data (i.e., inconsistent measurement of crimes outside the city, which may be correlated with neighborhood characteristics).

We binned ICE estimates into quintiles, as done previously,^14,15 for each city separately to avoid extrapolating beyond the data (i.e., ensuring all cities included zip codes in all quintiles) and because our interest is within-city variation. City-specific cutoffs are shown in Tables B through D (available as a supplement to the online version of this article at http://www.ajph.org).

Analysis

First, we described the total number of incidents per outcome (i.e., homicide, rape, robbery, aggravated assault, and firearm violence) from March through July in each year, along with the rate per population.

Second, we quantified cross-sectional differences in outcomes between ICE quintiles in 2020, with the most privileged quintile (Q5) as the referent. These estimates describe whether violence during the pandemic was disproportionately concentrated in less-privileged zip codes in a city compared with the most privileged zip codes. We used negative binomial regression models (which provided better fit than Poisson), including the log of the zip code population as an offset to adjust for population size. Models included city fixed effects, so we made all comparisons in, rather than between, cities. The exposure was a categorical variable for ICE quintiles (with separate models for each ICE measure: race, income, and race–income), and the outcome was counts of incidents from March through July in 2020 (with separate models for each outcome). We included zip code median age and percentage male as covariates.

Using the fitted models, we then estimated the marginal difference in the number of incidents associated with ICE quintiles by predicting counts under each level of the exposure, holding other variables (including population size) at their observed levels and taking the average difference.¹⁸ Sometimes called “standardization” or “g-computation,” this approach involves estimating conditional associations (adjusted for covariates), which are then used to generate marginal estimates of the expected outcome, standardized to the population’s covariate distribution.¹⁹ This allows us to estimate associations on the additive scale, which is most relevant for understanding public health impacts.²⁰

Third, we estimated change in violence over time (2020 vs 2018–2019) between quintiles of ICE measures. These difference-in-differences estimates describe whether violence increased (or decreased) disproportionately during the pandemic versus before the pandemic in less-privileged zip codes compared with the most privileged zip codes. We used negative binomial regression models and the marginal estimation approach we have described, but we included an interaction between time (an indicator equal to 1 if the year was 2020 and 0 if 2018 or 2019) and the categorical ICE measure wherein the interaction terms and associated marginal contrasts correspond to difference-in-differences estimates. That is, the interaction reflects the difference over time in the least privileged zip codes minus the difference over time in the most privileged. Because this approach differences out stable characteristics of place, we did not include median age or percentage male.

The use of marginal contrasts to estimate differences is advantageous because relative measures (e.g., ratios) do not account for baseline rates, whereas absolute measures do. For example, the public health implications of doubling a rate per 100 000 from 5 to 10 over time are different from doubling a rate from 1 to 2, yet the rate ratio for both is 2. Relative measures of association can therefore obscure potentially important public health effects of the exposure for communities in which the outcome is more common.

We calculated confidence intervals (CIs) for all estimates with bias-corrected clustered bootstraps with 500 iterations. We performed analyses in R version 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria) and Stata version 15.1 (StataCorp, College Station, TX).

Additional Analyses

We conducted 3 additional analyses. First, we computed ICE race/ethnicity measures comparing non-Latinx White people to all people of color because our main analysis, which included Latinx White individuals and non-Latinx White individuals together in the most privileged group, may be attenuated toward the null, given disparities in violence and associated risk factors between these groups. These measures compared non-Latinx White people (most privileged) to people of color (least privileged), and non-Latinx White households with incomes of $100 000 or more (most privileged) to households of people of color with incomes of $24 999 or less (least privileged). Second, we estimated associations for firearm violence injuries (nonfatal and fatal), as opposed to incidents. (We had consistent data on numbers of victims for this outcome only.) Third, we excluded cities with fewer than 20 events annually for an outcome citywide; this resulted in the additional exclusion of San Francisco for rape.

RESULTS

We excluded 3178 incidents. Exclusions ranged from 0% for many outcomes to 10% for firearm violence in Denver and averaged 2.7% of incidents (interquartile range = 1.0%–3.7%). Of excluded incidents, 34.5% did not contain valid geographic information, 0.7% did not correspond to a ZCTA, 0.9% were in ZCTAs with no population, and 63.9% were in ZCTAs that fell outside city boundaries.

Descriptive

The correlation between ICE measures was strong: 0.7 for ICE race and ICE income and 0.9 for ICE income and ICE race–income.

Aggravated assault was the most common and homicide the least common crime (Table 1). Overall during the pandemic, firearm violence increased 29.3% (from 15.0 per 100 000 population in 2018–2019 to 19.4 per 100 000 in 2020); assault increased 4.0% (from 198.7 to 206.6 per 100 000); homicide increased 27.7% (from 6.0 to 7.6 per 100 000); robbery decreased 23.3% (from 112.15 to 86.0 per 100 000); and rape decreased 31.4% (from 19.4 to 13.3 per 100 000).

TABLE 1—

Description of Violence Outcomes in 13 US Cities: March–July 2018, 2019, and 2020

Outcome and Year	Total No. Incidents	Total No. Incidents per 100 000 Population	Total No. Zip Codes
Firearm violence
2018	2526	14.6	520
2019	2667	15.4	520
2020	3356	19.4	520
Aggravated assault
2018	34 229	197.7	520
2019	34 567	199.7	520
2020	35 763	206.6	520
Homicidea
2018	909	5.8	465
2019	957	6.1	465
2020	1195	7.6	465
Robbery
2018	20 004	115.5	520
2019	18 829	108.8	520
2020	14 894	86.0	520
Rapeb
2018	2397	19.6	335
2019	2346	19.2	335
2020	1631	13.3	335

Note. The 13 US cities were Baltimore, MD; Boston, MA; Chicago, IL; Cincinnati, OH; Dallas, TX; Denver, CO; Detroit, MI; Los Angeles, CA; Milwaukee, WI; Philadelphia, PA; Phoenix, AZ; San Francisco, CA; and Seattle, WA.

^a Seattle and San Francisco were excluded because of missing outcome data or low counts (< 5 citywide).

^b Boston, Dallas, Denver, Detroit, Milwaukee, and Seattle were excluded because of missing outcome data or low counts (< 5 citywide).

On average, zip code median age was 35.6 years (SD = 5.5) and 50% of zip code populations were male (SD = 5%).

Difference Across Quintiles in 2020

For every outcome, in multivariable regression models controlling for median age and percentage male, less-privileged neighborhoods experienced a higher burden of violence in 2020 than the most privileged neighborhoods (findings were essentially unchanged in unadjusted models). Results were consistent across all ICE measures (race, income, and race–income). For example, we estimated that, on average across cities, if all zip codes were in the least privileged quintile (Q1) of ICE race–income, there would be approximately 14.1 more firearm violence incidents (95% CI = 5.9, 31.8; Figure 1a), 146.2 more aggravated assaults (95% CI = 112.4, 205.8; Figure 2a), and 4.9 more homicides (95% CI = 2.7, 9.0; Figure 3a) per zip code, than if all zip codes were in the most privileged quintile (Q5). These averages represent a range because cities vary in their baseline rates, and they should be interpreted in context. For example, the total number of aggravated assaults in March through July 2020 per 100 000 population ranged from 93.8 in Dallas (1416 assaults) to 822.6 in Milwaukee (5721 assaults). Results for robbery and rape show the same pattern (Figures A and B, panel A, available as a supplement to the online version of this article at http://www.ajph.org).

FIGURE 1—

Association Between Zip Code Index of Concentration at the Extremes (ICE) Income, Race, and Race–Income Measures and Firearm Violence in 13 US Cities, March–July by (a) Difference Across ICE Quintiles in 2020, and (b) Difference in 2020 Relative to 2018–2019 Across ICE Quintiles

Note. CI = confidence interval; Diff = difference in count of incidents. The most privileged quintile (Q5) is the referent. Results in part a reflect cross-sectional differences between quintiles in 2020. Results in part b reflect difference-in-differences estimates of change over time (2020 vs 2018–2019) between quintiles.

FIGURE 2—

Association Between Zip Code Index of Concentration at the Extremes (ICE) Income, Race, and Race–Income Measures and Aggravated Assault in 13 US Cities, March–July by (a) Difference Across ICE Quintiles in 2020, and (b) Difference in 2020 Relative to 2018–2019 Across ICE Quintiles

Note. CI = confidence interval; Diff = difference in count of incidents. The most privileged quintile (Q5) is the referent. Results in part a reflect cross-sectional differences between quintiles in 2020. Results in part b reflect difference-in-differences estimates of change over time (2020 vs 2018–2019) between quintiles.

FIGURE 3—

Association Between Zip Code Index of Concentration at the Extremes (ICE) Income, Race, and Race–Income Measures and Homicide in 11 US Cities, March–July by (a) Difference Across ICE Quintiles in 2020, and (b) Difference in 2020 Relative to 2018–2019 Across ICE Quintiles

Note. CI = confidence interval; Diff = difference in count of incidents. The most privileged quintile (Q5) is the referent. Results in part a reflect cross-sectional differences between quintiles in 2020. Results in part b reflect difference-in-differences estimates of change over time (2020 vs 2018–2019) between quintiles. San Francisco and Seattle were excluded because of missing outcome data or low counts (< 5 citywide).

Relative Differences Over Time

Disparities in change over time between quintiles were not consistent across outcomes. We found larger increases from 2018 through 2019 to 2020 in less-privileged quintiles than in the most privileged quintile for 3 outcomes: firearm violence, homicide, and aggravated assault (difference-in-differences estimates shown in Figures 1–3, part b). Relative to 2018 through 2019, we estimated an increase of 2.3 firearm violence incidents (95% CI = 0.5, 6.7; Figure 1b), 22.5 aggravated assaults (95% CI = 5.0, 44.7; Figure 2b), and 0.9 homicides (95% CI = 0.1, 2.2; Figure 3b) per zip code on average across cities associated with the least privileged versus the most privileged quintile of ICE race–income in 2020. These results were similar across race, income, and race–income ICE measures. Robbery and rape exhibited relative decreases in less-privileged zip codes for some or all ICE measures (Figures A and B, panel b).

Additional Analyses

Results for ICE race/ethnicity and race/ethnicity–income measures that compared non-Latinx White people to all persons of color were similar to those from the main analyses (Figures C–G, available as a supplement to the online version of this article at http://www.ajph.org).

Results for firearm violence injuries (Figure H, available as a supplement to the online version of this article at http://www.ajph.org) were consistent with results for firearm violence incidents.

Excluding cities with fewer than 20 events annually citywide did not change the results (not shown).

DISCUSSION

In this study of 13 large US cities, we quantified place-based social disparities in violence (1) cross-sectionally during the first months of the COVID-19 pandemic, and (2) over time relative to years past. Zip codes with higher concentrations of low-income households and higher concentrations of either Black people or all people of color experienced substantially higher rates of violence from March through July 2020 than did zip codes with higher concentrations of high-income households and White people. For firearm violence, aggravated assault, and homicide, inequities increased during the pandemic. Our findings are consistent with previous studies documenting stark racial and socioeconomic differences in violence²¹ and with news media reports showing a disproportionate rise in violence in marginalized communities during the pandemic.²

Future research should examine the factors driving this disproportionate increase. For example, what impact has the closure of schools and community organizations had on violence in segregated and disinvested communities? The fact that homicide and assault, but not robbery and rape, rose disproportionately in disadvantaged neighborhoods suggests that these communities experienced disparate exposure to conditions during the pandemic that uniquely affected risk for these types of violence.

Robbery and rape decreased disproportionately in less-privileged zip codes (where they were already higher) compared with the most privileged zip codes. These were the only 2 outcomes that declined overall in 2020, so results may reflect the fact that relative declines of the same magnitude correspond to greater absolute declines in areas with higher baseline rates (exploratory analyses of rate ratios showed no significant disproportionate changes over time). The overall decline in robbery and rape may be real or an artifact of changes in reporting. For example, shelter-in-place orders may have reduced stranger rape or limited victims’ ability or willingness to report intimate partner rape. Again, research is needed to understand the drivers of these trends.

Our results have 2 main implications. First, given the substantial and growing burden of violence in low-income neighborhoods of color, there is a need for focused violence prevention strategies that address the unique challenges of the pandemic. Recent federal commitments to invest in community violence prevention are encouraging,²² and existing empirical evidence supports a number of actionable steps.²³ For example, reducing economic and housing instability (e.g., through temporary financial assistance and eviction bans during the pandemic) and improving the physical environment and outdoor green spaces (which can promote collective efficacy while minimizing risks of coronavirus transmission) may curb or reduce violence.²³

A growing body of literature also documents the success of focused interventions that engage individuals most affected by violence, such as Advance Peace and Cure Violence.²⁴ Chronic underfunding and the pandemic have brought challenges to implementation, including physical-distancing requirements, disruptions in client engagement, interruptions in service and resource provisions, and additional demands on outreach workers,²⁵ but preliminary evidence suggests that such programs have adapted.²⁵ This resilience, combined with appropriate investment and resources (including funding staff, designating them essential workers, and providing personal protective equipment during the pandemic), indicates that these interventions may be especially critical for reducing violence in the current context. Similarly, comprehensive, trauma-informed care for those injured by violence, including hospital-based violence intervention programs, may help interrupt the cycle of violence,²⁶ although these programs too have been hampered by the pandemic.²⁷ Support for those indirectly exposed to violence is also warranted.

Second, our results affirm well-documented associations between the spatial concentration of violence and low-income and marginalized racial/ethnic groups²⁸ and challenge policymakers and those interested in improving public health to examine the historical and contemporary processes that contribute to these enduring inequities. For example, neighborhoods “redlined” by the Homeowner’s Loan Corporation in the 1930s (i.e., neighborhoods with large Black and immigrant populations) experience higher rates of firearm violence today than do neighborhoods deemed most desirable.²⁹ This past de jure segregation³⁰ may be related to present-day violence via impacts on education, transportation, jobs, income and wealth, and the built environment.³¹

Mass incarceration of Black men, another example of structural racism in the United States,³² and its direct and indirect consequences (e.g., on educational and employment opportunities) have contributed to family disruption and, in turn, potentially elevated the risk of violence, particularly among youths.³³ The socioecological factors underlying these disparities are not unique to violence, and there is growing consensus that racism is a public health crisis³⁴ with pervasive impacts. Additional research on the embodiment of structural racism may contribute to our conceptualization of health disparities and the advancement of policies that promote equity and justice. Continued surveillance of health disparities will be critical in these efforts.

Limitations

Exposures and outcomes may be measured with error. For example, crime data rely on police reports, which do not capture all incidents; changes in incident reporting over time could bias results but likely toward the null, given strains between police and low-income communities of color during the pandemic. Coding systems varied slightly across cities (e.g., Dallas excluded incidents if the victim or suspect was younger than 17 years), but these differences are unlikely to influence results because we focused on within-city comparisons. Gun Violence Archive data are based, in part, on news reports and other public sources and may, therefore, fail to capture all instances of firearm violence or accurately record their circumstances. However, the Gun Violence Archive is the most comprehensive, real-time source of data on firearm violence to our knowledge, and it has been shown to correlate well with official firearm homicide data from the Centers for Disease Control and Prevention³⁵ (no such data on nonfatal firearm injury exist).

Place-based social measures are from 2015 to 2019 (the most recent estimates available), whereas outcomes were measured in 2018 through 2020. There is also likely to be some degree of mismatch between ZCTAs and zip codes. This is reflected, in part, by missing data (2.7% of incidents were missing, on average). Zip codes are relatively small geographic areas, and some of our estimates are accompanied by wide CIs. We based our selection of 13 large US cities on convenience in accessing data; the selection is not nationally representative and may not generalize to other cities. Lastly, our data only go through July 2020, but the pandemic continues. Future research should examine whether continued consequences of the pandemic lead to even wider variation across categories of advantage.

Public Health Implications

Evidence on the social distribution of violence is critical to guide equitable prevention and response measures. We show that the least privileged communities endure endemically high levels of violence and that the COVID-19 pandemic and the events of 2020 had inequitable impacts, exacerbating disparities in some of the most severe forms of violence. Immediate mitigation strategies that address the unique challenges of the pandemic and long-term investments in antiracist programs and policies that address economic inequality are warranted.