The recent rise in homicide: An analysis of weekly mortality data, United States, 2018–2022

Abstract

Background:

From 2019 to 2020, homicide showed its largest single-year increase in modern US history. While many have cited the COVID-19 pandemic or police killing of George Floyd as initiating the rise, there has been limited systematic investigation of how the timing of the increase corresponded with these key events. We investigated trends in firearm and non-firearm homicide across sociodemographic and geographic groups to clarify the timing and nature of the recent increase.

Methods:

We conducted a descriptive epidemiologic study using National Vital Statistics System weekly mortality data from January 2018 to December 2022, US. We seasonally adjusted and smoothed weekly firearm and non-firearm homicide data, quantifying changes in relation to key event dates for the COVID-19 pandemic, killing of George Floyd, and 2020 national election. We disaggregated trends by sociodemographic and geographic characteristics.

Results:

Between January 2018 and December 2022, firearm homicide increased by 54% while non-firearm homicide was stable. The increase in firearm homicide started by October 2019 and stabilized by November 2020; 28% of the eventual increase had already occurred by the time COVID-19 was declared a national emergency. All sociodemographic and geographic groups experienced large recent increases in firearm homicide.

Conclusions:

The magnitude and timing of the recent increase in homicide has been previously understated and obscured by crude data and seasonal patterns. Existing theories, including the COVID-19 pandemic, fall short in explaining the historic surge, which is specific to firearm homicide, started in late 2019, and affected all persons and places across the US.

INTRODUCTION

From January 2019 to January 2020, the homicide rate in the United States increased by 35%, the largest single-year increase in modern history.^1–4 By 2021, nearly 26,000 lives were lost to homicide and over 80% of these violent deaths were caused by firearms.³ The unprecedented increase in fatal violence has been widely attributed to the COVID-19 pandemic,⁵ with theories pointing toward health and social disruptions, heightened stressors (e.g., economic, social, and psychological), and a surge in firearm purchases during the pandemic as possible reasons.^6–10 Others have proposed theories relating to the social and political unrest following the police killing of George Floyd in May 2020, which may have eroded police-community relations and prompted possible de-policing efforts.^11,12 In addition, this period coincided with a deeply divisive presidential election campaign that generated one of the highest voter turnouts in 40 years.^13,14

While some of these theories have received limited scientific investigation,^6,11,12 a basic understanding of the timing, magnitude, and distribution of the recent increase in homicide is lacking. These descriptive epidemiologic facts are fundamental to understanding the recent rise in fatal violence and can shed critical insights on the plausibility of the commonly cited explanations.^15–17 For example, if the COVID-19 pandemic was the main driver, then the increase would not be anticipated until after the pandemic widely impacted lives of Americans.¹⁸ Alternatively, if social unrest following the police killing of George Floyd drove increased violence, then the differential racial impacts of this event may have led to differential trends among Black communities or urban areas.^19,20

There has been little systematic evaluation of the extent to which the timing of the recent homicide increase corresponded with the timing of events that have been hypothesized as incendiary incidents. It is also unclear if specific sociodemographic groups and localities disproportionately bore the burden of additional violent deaths. In this study, we provide a detailed analysis of the timing of recent changes in national homicides using recently released weekly mortality data from January 2018 to December 2022. We describe how rising homicides mapped onto key events, as well as describe these changes by sociodemographic and geographic characteristics. Detailing the exact timing of the change in trend from a historically stable baseline can help to rule out – or put constraints on – the portion of the increase that can plausibly be attributed to any given event.

METHODS

This descriptive analysis modeled trends in weekly mortality data between January 2018 and December 2022 to isolate and characterize the recent increase in homicide in the US. The University of Michigan Institutional Review Board approved this study and waived informed consent (HUM00246809), as the study was not human participant research. We followed recommendations set out in the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline.

Data Sources

We used mortality information from the National Vital Statistics System (NVSS) data files. These data are compiled from standardized death certificates filed by various jurisdictions throughout the US and are maintained centrally by the National Center for Health Statistics (NCHS). Medical examiners or coroners are required to investigate homicides and typically populate the information provided in homicide death certificates.²¹ The NCHS then uses this information, as well as information on cause of death, to classify deaths using International Classification of Diseases (ICD) codes. These data are the most reliable source of national homicide data for the US, capturing over 99% of all deaths.²²

Recently Released Mortality Data by Week

For our main analyses, we obtained the most up-to-date complete and final weekly mortality data from the Provisional Multiple Cause of Death database,²³ which we accessed via Centers for Disease Control and Prevention Wide-ranging Online Data for Epidemiologic Research (CDC WONDER; downloaded July 20, 2024). At the time of analysis, the provisional database compromised of final weekly data for the years 2018 to 2022, provisional data for 2023, and provisional and partial data for 2024.²³ We restricted our analyses to only include the final weekly data from the provisional database (2018–2022). We identified underlying causes of death for firearm and non-firearm homicide using the injury intent and mechanism fields provided in CDC WONDER, which are based on the ICD-10 codes for firearm homicide (U01.4, X93-X95) and non-firearm homicide (U01.0-U01.3, U01.5-U01.9, U02, X85-X92, X96-Y09, Y87.1). We extracted the finest grained temporal information available of Morbidity and Mortality Weekly Report (MMWR) week. MMWR weeks vary slightly from calendar year weeks as MMWR weeks run Sunday through Saturday and, while the first MMWR week always includes January 1^st, it can begin before January 1^st. The first and last MMWR Week can therefore include dates from the preceding or next calendar year, and some years have 53 MMWR weeks.²³ We defined MMWR week by the first day of the week. As a result, our main study period spanned from December 31, 2017 (the earliest available MMWR week) to December 25, 2022 (the latest available and final MMWR week).

Historical Mortality Data by Month

For estimating our seasonal adjustment factors (see Statistical Analyses), we extracted a longer time series of monthly data from 2011 to 2017 from CDC WONDER. For this time period, month was the finest grained available temporal interval.²⁴ For supplementary analyses that aimed to contextualize the magnitude of the recent increase in homicide, we also obtained restricted-access micro-files via special request to the NCHS to obtain monthly historical data from the same source (from 1973 through to 2022) as data before 1999 were not publicly available (see eAppendix; http://links.lww.com/EDE/C201 for more details).²⁵

Statistical Analysis

We defined the critical period as July 2019 (MMWR week July 7, 2019) to December 2020 (MMWR December 27, 2020), based on visual inspection of when the homicide increase occurred (Figure 1) and comparison of year-over-year increases based on the various year definitions (i.e., month beginning, see eTable 1; http://links.lww.com/EDE/C201). This critical period was excluded when computing the seasonal adjustment factor.

Figure 1.

Weekly counts of firearm and non-firearm homicide, December 30, 2017, to December 25, 2022. Solid vertical lines represent the critical period (July 2019 to December 2020).

Separately for firearm and non-firearm homicides, we computed a seasonal adjustment factor as the estimated homicide risk ratio for each MMWR week relative to the annual risk averaged over 10 years of data (e.g., a MMWR week that covers July 4 to July 10 has 1.14 times more firearm homicides than average).^26,27 In computing this, we used weekly data when available (since 2018) and monthly data back to July 2011. We then computed seasonally adjusted firearm homicides for a given MMWR week as the actual firearm homicides in that week divided by that week’s seasonal adjustment factor. Next, to reduce noise but maintain the accuracy of the data, we smoothed the seasonally adjusted time series, comparing various semiparametric and nonparametric techniques for modeling the seasonally adjusted data as a function of time.²⁸ We used leave-one-out cross-validation to select the model that minimized root mean squared error in the critical period, with our final selected smoother being a kernel with tricube weighting distribution and a bandwidth of 18 weeks. More details are available in the eAppendix, including eFigure 1; http://links.lww.com/EDE/C201. All analyses were performed in R version 4.3.1, and all analytical code is publicly available on the Open Science Framework.²⁹

Key Event Dates

We identified dates relating to the three primary events of COVID-19, George Floyd, and the 2020 national election, which occurred during the critical period (July 2019 – December 2020). We focused on these events as these represent commonly proposed theories for explaining the recent increase in homicide. For to COVID-19, these included the issuing of the first travel ban from China (January 31, 2020), the declaration of COVID-19 as a national emergency (March 13, 2020), widespread shutdowns of schools and businesses (March 15, 2020), and the administration of the first vaccine (December 14, 2020).³⁰ For George Floyd’s killing, these included his death (May 25, 2020) and the peak of widespread protests (June 6, 2020).³¹ For the 2020 national election, these included the first presidential primary in Iowa (February 3, 2020) and Election Day (November 3, 2020). We also included the beginning date of each quarter during the critical period to provide additional benchmarks at even intervals throughout the critical period.

To quantify the increase in firearm homicide for these key event dates and quarters, we calculated three summary measures relative to a historical baseline: i) risk of firearm homicide relative to baseline, estimated as the modeled number of seasonally adjusted firearm homicides divided by the baseline number; ii) percent change from baseline that had occurred by each date relative to the maximum elevation in risk; and iii) the cumulative number of additional seasonally adjusted deaths relative to the baseline. We defined the historical baseline as the seasonally adjusted weekly homicide count averaged between July 2018 and December 2019. We selected this 18-month period because it represents a stable baseline, which is derived using the same weekly data as our primary analyses and accounts for seasonal effects by matching on the months included in the critical period (July 2019 to December 2020). In a sensitivity analysis, we derive an alternative baseline using a longer 3-year time series from the monthly data to check the robustness of our primary baseline definition (see eAppendix; http://links.lww.com/EDE/C201 for methodologic details).

Comparing Increases in Firearm Homicide by Sociodemographic and Geographic Groups

In subgroup analyses, we crudely quantified the magnitude of the recent increase in firearm homicides by sociodemographic and geographic groups for comparison purposes. We identified two seasonally matched 18-month periods that represented the most stable periods before (July 2018 through December 2019) and after (July 2020 through December 2021) the period of steepest increase in firearm homicide to estimate differences in the total eventual increase by sociodemographic and geographic groups (see Figure 1 and 2). We conducted a sensitivity analysis of these time periods – comparing relative increases in firearm homicide from January 2018 through June 2019 to January 2020 through June 2021. These 18-month periods circumvent suppressed counts for important subgroups. Sociodemographic groups included sex (male; female), age group (<18y; 18–34y; 35–54y; 55y+), and race and ethnicity (White, not Hispanic or Latino; Black or African American, not Hispanic or Latino; Asian, not Hispanic or Latino; Native Hawaiian or Other Pacific Islander, not Hispanic or Latino; American Indian or Alaska Native, not Hispanic or Latino; more than one race, not Hispanic or Latino; all races, Hispanic or Latino). Geographic groups, based on county of residence, include county urbanicity (large central urban, fringe urban (suburban), other/rural),²³ Census Division, and state politics based on Democratic share of the two-party presidential vote (red, purple, blue) in 2020.³²

Figure 2.

Smoothed seasonally adjusted weekly firearm homicides compared to a historical baseline, December 30, 2017, to December 25, 2022. Solid vertical lines represent the critical period (July 2019 to December 2020). Smoothed using the best fitting model (a tricube kernel with an 18-week bandwidth) and includes model estimates for 95% confidence intervals. The baseline (horizontal dashed line) was defined as the average seasonally adjusted weekly homicide count between July 2018 and December 2019 (272 firearm homicides). Key event dates are represented by dashed vertical lines for: the declaration of COVID-19 as a national emergency and widespread shutdowns (March 13, 2020), George Floyd’s death (May 25, 2020), and the national Election Day (November 3, 2020).

RESULTS

Between December 31, 2017, and December 25, 2022, there were 113,431 homicides, 78% of which involved a firearm, an average of 339 firearm homicides and 96 non-firearm homicides per week (eTable 1; http://links.lww.com/EDE/C201). Weekly non-firearm homicides were stable, but firearm homicides dramatically increased during this period (Figure 1 and eFigure 2; http://links.lww.com/EDE/C201). After accounting for a strong seasonal pattern (eFigure 2; http://links.lww.com/EDE/C201) – with lows in February and peaks in July – Figure 2 highlights that the deviation from the previously stable baseline preceded 2020. Firearm homicide began to increase steadily from October 2019 through March 2020, then increased rapidly from April through June 2020 (eFigure 3; http://links.lww.com/EDE/C201), peaking in risk a during the last week of October 2020 – the week before the national Election Day (Figure 2). The timing of the onset and peak of the increase in firearm homicide broadly tracks a surge in firearm purchasing, as measured by firearm background checks, which started to increase in mid-2019 (eFigure 4; http://links.lww.com/EDE/C201).

At the peak of the increase in risk in late October 2020, the weekly firearm homicide risk was 1.54 (95% CI:1.49–1.58) times higher than baseline (Figure 2). Of this total increase in risk, over a quarter (28%) had already occurred by the time COVID-19 was officially declared a national emergency and widespread closures began. Nearly two-thirds (65%) of the increase in risk had occurred by the time of George Floyd’s death, and nearly three-quarters (72%) had occurred before the most widespread protests following his death. By the time of the national Election Day, there had been an estimated 4,241 additional firearm homicides relative to the expectation under the baseline rate (Table 1). The general timing and magnitude of these changes were robust to different modeling specifications (eFigure 1; http://links.lww.com/EDE/C201) and alternative baseline definitions (eTable 2; http://links.lww.com/EDE/C201).

Table 1.

Increase in seasonally adjusted firearm homicide, relative to historical baseline, as realized by key dates.

Quarters and key event datesa	Date	Relative riskb (95% CI)	% increase relative to maximum increase in riskc	Cumulative additional deaths (95% CI)d
Start of Q4 2019	October 1, 2019	1.02 (0.97 to 1.07)	4	69 (−58 to 196)
Start of Q1 2020	January 1, 2020	1.09 (1.04 to 1.14)	17	285 (92 to 478)
COVID-19: First travel ban from China	January 31, 2020	1.10 (1.05 to 1.15)	19	359 (147 to 570)
National election: First presidential primary (in Iowa)	February 3, 2020	1.10 (1.06 to 1.15)	19	395 (179 to 611)
COVID-19: National emergency and widespread shutdowns	March 14, 2020e	1.15 (1.10 to 1.20)	28	533 (295 to 771)
Start of Q2 2020	April 1, 2020	1.17 (1.13 to 1.22)	32	684 (432 to 936)
George Floyd: Death	May 25, 2020	1.35 (1.30 to 1.40)	65	1062 (779 to 1345)
George Floyd: Peak of widespread protests	June 6, 2020	1.39 (1.35 to 1.44)	72	1338 (1045 to 1631)
Start of Q3 2020	July 1, 2020	1.45 (1.41 to 1.50)	83	1830 (1519 to 2142)
Start of Q4 2020	October 1, 2020	1.52 (1.47 to 1.56)	96	3477 (3106 to 3847)
National election: Election Day	November 3, 2020	1.53 (1.49 to 1.58)	98	4241 (3849 to 4634)
COVID-19: First vaccine administered	December 14, 2020	1.49 (1.45 to 1.54)	91	5057 (4639 to 5476)

^aKey dates are the start of each quarter (Q) and the exact dates of three key events during the critical period for COVID-19, George Floyd, and the 2020 national election.

^bRelative risk (RR) measures the adjusted weekly count of firearm homicide relative to the baseline of 272 homicides per week based on weekly seasonally adjusted firearm homicides from July 2018 through to December 2019 (see eTable 2 for alternative baseline).

^cMaximum increase in risk relative to baseline was 1.54, which occurred in late October 2020. Percent (%) increase relative to maximum is thus calculated as ((RR-1.00)/(1.54–1.00))*100.

^dCumulative additional deaths are computed as the sum of seasonally adjusted weekly firearm homicides minus average weekly seasonally adjusted firearm homicides during the baseline (July 2018 through to December 2019) with the cumulative sums starting the week beginning July 7, 2019, and going through the week beginning that contains the key date. Corresponding 95% confidence intervals (CI) were calculated based on the natural log scale for the ratio of counts.

^eNational emergency declaration occurred on March 13, 2020, and widespread shutdowns of schools and businesses began on March 15, 2020.

Between July 2020 to December 2021, there were 1.5 times the number of firearm homicides as occurred 2 years prior (July 2018 to December 2019), which corresponds to an increase of more than 10,500 firearms (Table 2). An increase of this magnitude represents a considerable historical outlier in both relative risk and count difference (eFigure 5; http://links.lww.com/EDE/C201). In the 45 years prior to 2020, the largest relative risk calculated over a comparable timeframe was 1.3; the largest historical increase in absolute terms was 4,738 firearm homicides (eFigure 5; http://links.lww.com/EDE/C201).

Table 2.

Increase in firearm homicide risk for July 2020 through December 2021 relative to July 2018 through December 2019, by sociodemographic and geographic group.

Sociodemographic or geographic group	Number of deaths July to December 2019	Number of deaths July 2020 to December 2021	Count differencea	Relative risk (95% CI)b
National total	21501	32031	10530	1.49 (1.46 to 1.52)
Sex
Female	3414	5023	1609	1.47 (1.41 to 1.54)
Male	18087	27008	8921	1.49 (1.46 to 1.52)
Race, ethnicity
White, Not Hispanic or Latino	4756	6189	1433	1.30 (1.25 to 1.35)
Black or African American, Not Hispanic or Latino	12520	19611	7091	1.57 (1.54 to 1.61)
Asian, Not Hispanic or Latino	259	293	34	1.13 (0.96 to 1.34)
Native Hawaiian or Other Pacific Islander, Not Hispanic or Latino	46	59	13	1.28 (0.87 to 1.88)
American Indian or Alaska Native, Not Hispanic or Latino	230	292	62	1.27 (1.07 to 1.51)
More than one race, Not Hispanic or Latino	252	373	121	1.48 (1.26 to 1.74)
All races, Hispanic or Latino	3403	5166	1763	1.52 (1.46 to 1.59)
Age group (y)
<18	1368	2336	968	1.71 (1.60 to 1.83)
18–34	12177	18196	6019	1.49 (1.46 to 1.52)
35–54	5964	9010	3046	1.51 (1.46 to 1.56)
55+	1990	2489	499	1.25 (1.18 to 1.33)
Urbanicity
Large central urban	9286	14384	5098	1.55 (1.51 to 1.59)
Large fringe urban (suburban)	3745	5488	1743	1.47 (1.41 to 1.53)
Other/rural	8470	12159	3689	1.44 (1.40 to 1.48)
Census Region, Division
West, Pacific	2388	3531	1143	1.48 (1.41 to 1.56)
West, Mountain	1249	1859	610	1.49 (1.39 to 1.60)
Midwest, West North Central	1278	1727	449	1.35 (1.26 to 1.45)
Midwest, East North Central	3512	5685	2173	1.62 (1.55 to 1.69)
South, East South Central	2266	3479	1213	1.54 (1.46 to 1.62)
South, West South Central	3282	4974	1692	1.52 (1.45 to 1.59)
South Atlantic	5544	7618	2074	1.37 (1.32 to 1.42)
Northeast, Middle Atlantic	1672	2721	1049	1.63 (1.53 to 1.73)
Northeast, New England	310	437	127	1.41 (1.22 to 1.63)
Politics
Red	7151	10332	3181	1.44 (1.40 to 1.48)
Purple	6443	9520	3077	1.48 (1.43 to 1.53)
Blue	5844	8919	3075	1.53 (1.48 to 1.58)

^aCount difference was defined as number of firearm homicides in July 2020 to December 2021 minus firearm homicides in July 2018 to December 2019.

^bRelative risk (RR), or relative increase, was defined as the ratio of total firearm homicides in July 2020 to December 2021 divided by the total firearm homicides in July 2018 to December 2019. Corresponding 95% confidence intervals (CIs) were calculated by identifying the 95% confidence limits on the natural log scale and exponentiating the upper and lower limits to convert back to the RR scale.⁵⁰

This unprecedented recent increase was experienced among all sociodemographic and geographic subgroups, with risks relative to 2 years prior ranging from 1.13 to 1.71 (Table 2). While relative increases were similar among males and females, there was variation by race, ethnicity, and age group. Relative increases were more pronounced among non-Hispanic Black persons (relative risk (RR)=1.57 [95%CI:1.54–1.61]) and Hispanics of all races (RR=1.52 [95%CI:1.46 to 1.59]) than among non-Hispanic White persons (RR=1.30 [95%CI:1.25–1.35]) and non-Hispanic Asian persons (RR=1.13 [95%CI:0.96–1.34]). Youth and children (under age 18) experienced the largest increase in this period (RR=1.71 [95%CI:1.60–1.83]), almost three times higher than the increase experienced among persons aged 55 and over (RR=1.25 [95%CI:1.18–1.33]). Relative increases in firearm homicide were notably uniform across the US, with little variation by different geographic groups, including county urbanicity, Census Division, and state politics. The pattern of subgroup differences in relative increases was similar irrespective of the 18-month periods compared (eTable 3; http://links.lww.com/EDE/C201).

DISCUSSION

Several reports have characterized recent trends in firearm homicide as a sharp 35% increase from 2019 to 2020 that increased again in 2021, then decreased in 2022.^1,3 This characterization, however, is based on crude annual or monthly mortality rates that did not appropriately account for a strong seasonal pattern in firearm homicides that cyclically troughed in February and peaked in July each year. As a result, the magnitude of the recent increase in homicide, that is specific to firearm homicide, has been understated and dynamics of recent shift in trends obscured.

This descriptive epidemiologic analysis exploits the granular temporality available in recently released weekly US mortality data, and uses 10 years of monthly data to adjust for seasonality and nonparametric smoothing techniques to isolate the recent increase in firearm homicide from seasonal patterns and random noise. In doing so, we are able to show that late-2019 to late-2020 was an even deadlier year than previous reports suggest, with 54% more firearm homicides than expected based on historical trends.^1,3

Using finer-grained mortality data allowed us to examine how firearm homicide evolved relative to key events of 2020 that have been proposed as inciting incidents. We show that the rise in firearm homicide risk began in October 2019, several months before COVID-19 had impacted the lives of US citizens. While trends steepened after COVID-19 was declared a national emergency and following George Floyd’s death and widespread protests, much of the eventual increase in firearm homicide had already occurred by the time of those events. COVID-19 and the death of George Floyd might therefore have exacerbated an existing increasing trend, but neither can explain why firearm homicide started to deviate towards the end of 2019 from a baseline that has been remarkably stable during the 4 years prior.

The timing of the run-up in firearm homicide risk – starting in October 2019 and ending by November 2020 – coincides with the campaigning for the 2020 national election (Joe Biden versus Donald Trump).¹³ Previous evidence has also found that nationwide stress peaked around the time of this election, with increased reports of depression and anxiety, mental health visits, and prescription drug use.^33,34 However, given that homicide occurs under various circumstances and has multiple causes, the recent national increase in firearm homicide is unlikely to be explained by one singular factor,³⁵ such as the 2020 national election, especially as we found that increases were markedly similar in states across the political ideological spectrum. Further research is needed to examine the specific impacts of divisive elections (e.g., the 2020 national election), and the role of political factors (e.g., political legitimacy, polarization, government distrust),^13,36 in shaping population trends in violence.

Our analysis also shows that the recent increase in lethal violence was specific to firearms.³⁷ The specificity of the increase to firearms suggests that established social determinants of violence, such as income inequality, economic opportunity, illegal drug activity, social capital, and health policies, cannot adequately explain the recent trend.^38,39 Changes in these root causes of violence would typically be expected to affect both firearm and non-firearm homicide, not to result in an increase that is specific to one mechanism of injury. A surge in firearm purchases, that was already evident in the early months of 2020,^6,13,40 is one possible mediator that is specific to the mechanism of injury (eFigure 4; http://links.lww.com/EDE/C201). Understanding the reasons motivating increasing numbers of Americans to purchase firearms, including the reasons motivating new profiles of first-time firearm owners (e.g., women, historically marginalized communities),^40,41 may be key to explaining the surge in firearm homicide. Recent studies suggest that nationwide increases in firearm purchasing behaviors were too related to the COVID-19 pandemic due to heightened fears of violence,^42,43 and prior studies show that political elections can also influence firearm purchasing behaviors due to fear of stricter gun control (cf. the “Obama” effect in 2008).⁴⁴ Closer examinations of the interrelationship between the recent increase in firearm purchases and the recent increase in firearm homicide are needed to isolate the role of firearm availability– even when regulated and legally purchased – in contributing to the recent excess in violent deaths, which were specific to firearms.⁶

One of our most striking findings is that the recent rise in firearm homicides similarly affected all geographic areas. We also saw large increases across all sociodemographic groups, although the magnitude of the increase was particularly stark for children and adolescents and Black and Hispanic/Latino Americans, compounding existing social disparities in violent victimisation.^37,45,46 Plausible theories must be able to explain these two features of the increase in firearm homicide: that all places and people were affected, but not all people were affected equally. While universal factors appear to be contributing to an overall increase in firearm homicide, irrespective of person and place, other specific factors may also be at play, resulting in the observed concentrated increases among certain populations. For example, Black and Hispanic/Latino Americans experienced a disproportionate burden of COVID-19 morbidity and mortality because of persistent structural inequalities,⁴⁷ which may have also placed these populations at heightened risk of firearm homicide victimization.

This study is subject to several limitations. First, complete mortality data were only publicly available at weekly intervals from 2018 through 2022. Second, discontinuity in the methodology for deriving monthly population estimates (2010 to 2019 intercensal estimates versus 2020–2022 postcensal estimates⁴⁸) meant that it was not possible to identify a consistent monthly time series for population estimates. Although our analyses are limited by not adjusting for population changes from 2018 to 2022, we anticipate that this would minimally impact our findings as population growth was at a historical low (0.1%) during these critical years due to reduced fertility, increase mortality, and reduced net migration.⁴⁹ Third, our primary baseline definition is slightly upward biased because it includes months at the end of 2019, which already show elevated levels of firearm homicide. While our sensitivity analysis using an alternative baseline shows that our findings are robust, the inclusion of these months means that our estimates of the timing and magnitude of the increase in firearm homicide are conservative. Fourth, the study design cannot causally attribute a portion of the increase in firearm homicide to any of the key events – only quantify the portion of the increase that cannot be attributed to each key event since a portion of the increase occurred prior to every key event date. Finally, recent research and media reports suggest that police-reported firearm violence did not increase equally across or within cities,^5,50 but we found little variation in relative increases in medically reported firearm homicide by the cruder geographic groups of urbanicity, Census Division, and state politics. More research is needed to determine whether the cruder groupings used in our analyses are obscuring important geographic differences, such as at the neighborhood level,^5,50 or whether medically reported firearm homicide differs from police-report firearm violence.

The findings from this descriptive epidemiologic analysis cannot identify what caused the historic surge in firearm homicide. However, by clarifying the timing of the surge as starting from October 2019 up until November 2020 and showing that all segments of the population were affected, we can rule out effects of COVID-19 and the police killing of George Floyd as responsible for the change in trend. Plausible theories for explaining the unprecedented increase in fatal violence must explain why the increase was specific to firearms, the timing of initial deviation from a previously stable trend, and how all peoples and places were affected across the United States. Given that COVID-19 and the police killing of George Floyd do not meet these criteria, it is critical to identify the pre-existing macro-factors that initiated the upward trend in firearm homicide to understand the pervasive risks that continue now that these specific events have passed. Such an understanding is foundational to informing universal prevention strategies that target the root causes, and mechanisms, of firearm violence.