Objectives. To describe all-outcome injurious shootings by police and compare characteristics of fatal versus nonfatal injurious shootings nationally.
Methods. From July 2021 to April 2023, we manually reviewed publicly available records on all 2015–2020 injurious shootings by US police, identified from Gun Violence Archive. We estimated injury frequency, case fatality rates, and relative odds of death by incident and victim characteristics.
Results. A total of 1769 people were injured annually in shootings by police, 55% fatally. When a shooting injury occurred, odds of fatality were 46% higher following dispatched responses than police-initiated responses. Injuries associated with physically threatening or threat-making behaviors, behavioral health needs, and well-being checks were most frequently fatal. Relative to White victims, Black victims were overrepresented but had 35% lower odds of fatal injury when shot.
Conclusions. This first multiyear, nationwide analysis of injurious shootings by US police suggests that injury disparities are underestimated by fatal shootings alone. Nonpolicing responses to social needs may prevent future injuries.
Public Health Implications. We call for enhanced reporting systems, comprehensive evaluation of emerging reforms, and targeted investment in social services for equitable injury prevention. (Am J Public Health. 2024;114(4):387–397. https://doi.org/10.2105/AJPH.2023.307560)
Firearm injuries are a public health crisis, annually costing 45 000 lives and more than 1 million disability-adjusted life years in the United States alone.1,2 Harms associated with use of firearms are compounded and reinforced by underreporting, inadequate funding for prevention, and other structural inequities.2 Among the most underreported but societally impactful forms of firearm injury are shootings by police, which result in 1000 US fatalities annually1 and likely contribute to worsening public perceptions of policing. According to a Pew Research Center survey, in 2020, just 35% of US adults agreed that police use the right amount of force in all situations, and 34% believed that police treat racial and ethnic groups equally.3 These views have fueled national policy debates about public safety reforms4 and calls for public health action against violence by police,5 but data needed for empirical decision-making are lacking because of persistent gaps in national use-of-force injury surveillance.
Owing to their relative comprehensiveness, inclusion of contextual data, and minimal reporting lag, news media repositories (e.g., Fatal Encounters, Mapping Police Violence, The Guardian’s The Counted, The Washington Post’s Fatal Force, and Gun Violence Archive [GVA]) are currently the best available sources for a national accounting of injuries by police use of deadly force. Alternative data sources include accountability systems of the US Federal Bureau of Investigation (including the recently phased-out Supplemental Homicide Reports, replaced by National Use-of-Force Data Collection), the Centers for Disease Control and Prevention’s (CDC’s) National Vital Statistics System, and CDC’s National Violent Death Reporting System. These 3 federal systems underestimate fatal injuries by police shootings through insufficient agency participation, inconsistencies in cause-of-death designation, and inconsistent state participation, respectively.6–9 Some states and localities maintain accessible databases, but these sources are not nationally representative.10,11 Of all national data sources, only National Use-of-Force Data Collection and GVA document fatal and nonfatal shootings by police.
The various media repositories produce comparable national estimates of fatal injuries from police use of force.7,12 These sources have been used to describe disparities in fatal injuries by age,13,14 race,11,13,15 gender,13 armed status,15,16 mental health status,15 and other characteristics, including US region.13,15,17 In an illustrative analysis by Nix et al. of 2015 shooting fatalities (n = 990), 50% of people killed in police shootings were White, 26% were Black, and 96% were men. The average age of victims was 37 years. Most victims were armed (82% with a deadly weapon or replica gun, 5.5% with a vehicle), and 9% were unarmed. Twenty-five percent involved signs of mental distress or history of mental illness according to reports by journalists and police at the time.18 Although collectively illuminating, by failing to account for nonfatal injuries, fatality studies still likely underestimate the national burden of injury from shootings by on-duty police.
Few studies have examined nonfatal injuries, leaving the frequency and characteristics of these shootings uncertain. In 1 analysis of 11 urban police and sheriffs’ departments with publicly available injury data, fatalities comprised 53% of injurious shootings by police.19 Another analysis of 4 state-mandated databases estimated 56% of people injured in police shootings died.10 A broadly inclusive study describing 2015 GVA-listed “officer involved incidents” (n = 1907) reported 49% of incidents were fatal.20 In these studies, fatal injuries were associated with older victim age,10,20 White versus Black racial identity,10,19 multiple police shooters,19 and nonofficer weapon possession.10,20 Odds of fatality were higher from injuries to victims armed with knives or blunt-force objects, compared with firearms, and lower among vehicularly armed victims (knife: odds ratio [OR] = 2.20; blunt object: OR = 2.33; vehicle: OR = 0.26).20 Unarmed victims had higher odds of survival than armed victims.10 When nonfatal injuries were included, injury disparities most affecting people who are Black were projected to be more severe than when estimated from fatalities alone.10
In sum, open-source data repositories of police use of force are reliable and informative resources that have produced broad understanding of fatal shootings by police nationally. However, fatal shootings may represent little more than half of all injurious shootings by police. To date, no published studies have examined the full and current burden of physical injury from shootings by on-duty US law enforcement officers. The objectives of this exploratory study were to (1) describe total people injured or killed in shootings by police in the United States using an up-to-date, multiyear nationwide data set and (2) compare characteristics of fatal versus nonfatal injurious shootings nationally.
We extracted and compiled data representing incidents and victim characteristics from GVA’s linked articles and other publicly available sources. GVA is a database of fatal and nonfatal US gun violence events, identified from approximately 7500 media, law enforcement, government, and commercial sources daily since 2013.21 Incidents are cataloged by date, location, and gun violence type. Data abstraction occurred from July 2021 to April 2023 for shootings by police occurring January 1, 2015, through December 31, 2020. The abstraction team consisted of 15 students from the Johns Hopkins Bloomberg School of Public Health. All 14 155 incidents designated as “officer involved incidents” were manually reviewed for eligibility and identification of case characteristics at least once. Abstractors received standardized training and a randomly assigned subset of incidents. In addition, a blinded 10% of incidents were repetitively assigned for quality assurance. Median total case assignment was 1100 (range = 460–5525).
Cases were restricted to include only incidents of shots fired by 1 or more law enforcement officers, resulting in injuries to people who were not responding officers. Accidental discharges, policing occupational injuries, injuries by bullet alternatives exclusively (e.g., rubber bullets, taser), shootings without injury, and self-inflicted injuries were excluded. GVA-designated “suicide by cop” shootings (i.e., shootings presumed to have been intentionally provoked) were retained.
Abstracted variables included situational characteristics (e.g., response type, incident type, shooting location, weapon involvement), victim demographics (e.g., gender, age, race, ethnicity), victim characteristics (e.g., housed or unhoused, armed status and weapon type, injury outcome), and a limited set of shooting-officer characteristics (e.g., on- or off-duty status, alone or accompanied, agency affiliation). Abstractors additionally identified and described incidents in which mental or behavioral health conditions were explicitly named in association with the shooting or its initiating incident. These cases were rereviewed and confirmed. Definitions of all abstracted variables are provided in Appendix A (available as a supplement to the online version of this article at https://ajph.org).
Abstractors categorically coded all descriptors using a combination of deductive and inductive techniques, aiming for objective reflection of best-available reporting. Abstractors cross-referenced fatal incidents with Fatal Encounters. Race and ethnicity designations were made when specified by sources or following 2-person concordant review of a published photo, an approximation of socially assigned identities.22 If ambiguous or unreported, abstractors selected “unknown.” All other coding uncertainties were discussed in weekly meetings. Post hoc review of repetitively assigned incidents revealed strong coding consistency; rare discrepancies were resolved through additional source review by the first author.
We calculated counts and proportions for total incidents and injuries, entirely nonfatal incidents versus incidents with at least 1 fatality, and nonfatal injuries versus fatal injuries. We calculated case fatality rates for incident and person characteristics. For each characteristic, we estimated odds of fatal injury outcome from a random-intercept model, in which victims were nested within incidents. We defined reference categories to support intuitive comparisons, based on majority representation (e.g., local police agencies, non-Hispanic White ethno-racial designation, masculine gender), or simplicity of the comparator (e.g., unarmed victim, shooting-related initiating incident). For age, regression models first included only age-specified victims (i.e., excluding “juvenile,” “adult,” or decade-approximated descriptive ages), then categorically examined all victims as “juvenile” (ages 0–17 years) or “adult” (ages ≥ 18 years). In adjusted models, we estimated the effect of each characteristic after accounting for incident-level clustering and holding all other incident and person variables constant. Confidence intervals were calculated based on an α of .05.
Estimates reflect injurious shootings by officers ostensibly acting “in the line of duty,” including shootings by on-duty officers, on- and off-duty officers in a multiple-officer response, off-duty officers acting in an on-duty capacity (e.g., performing investigative activities, identifying oneself as police), and incidents without explicitly reported duty status. In sensitivity analyses, we compared estimates under more restrictive duty-status criteria (i.e., only explicitly on duty or both on and off-duty) and maximally inclusive duty-status criteria (i.e., also off-duty officers working security positions and off-duty officers not acting in a law-enforcement capacity). We performed analyses with Stata version 16.1 (StataCorp LP, College Station, TX) and the melogit command with clustering by incident.
From 2015 to 2020, there were 10 308 incidents of US law enforcement officers shooting their firearms and injuring 1 or more people (Table 1). These incidents resulted in 5874 fatalities and 4741 individuals with nonfatal gunshot injuries, a 55.3% case fatality rate (Table 2). On average, 1769 people were injured annually (979 fatally; 790 nonfatally; Table 3). Examined monthly, injury frequency appeared cyclical but otherwise stable over the 6-year period (Appendix B, available as a supplement to the online version of this article at https://ajph.org).

TABLE 1— Fatal and Nonfatal Injurious Shooting Incidents, by Event Characteristic: United States, 2015‒2020
Incident Characteristic | Nonfatal Injurious Incident, No. | Fatal Incident, No. | % Fatal | Total Injurious Shooting Incidents, No. (%) |
Total | 4467 | 5841 | 56.7 | 10 308 (100) |
Agency type | ||||
Local police | 2867 | 3481 | 54.8 | 6348 (61.6) |
Sheriff’s office | 982 | 1412 | 59.0 | 2394 (23.2) |
State police | 227 | 341 | 60.0 | 568 (5.5) |
National agency | 73 | 103 | 58.5 | 176 (1.7) |
Special jurisdiction | 50 | 28 | 35.9 | 78 (0.8) |
Constable or marshal | 5 | 5 | 50.0 | 10 (0.1) |
Multiple shooting agencies | 209 | 441 | 67.8 | 650 (6.3) |
Unknown | 54 | 30 | 35.7 | 84 (0.8) |
Response type | ||||
On view | 1752 | 1952 | 52.7 | 3704 (35.9) |
Dispatched to 911 call | 2588 | 3783 | 59.4 | 6371 (61.8) |
By subject | 53 | 56 | 51.4 | 109 (1.1) |
Unknown | 74 | 50 | 40.3 | 124 (1.2) |
Incident type | ||||
Shooting | 438 | 525 | 54.5 | 963 (9.3) |
Assault | 155 | 249 | 61.6 | 404 (3.9) |
Disorderly conduct or dispute or disturbance | 162 | 236 | 59.3 | 398 (3.9) |
Domestic incident (disturbance, dispute, or violence) | 566 | 1048 | 64.9 | 1614 (15.7) |
Investigative | 243 | 294 | 54.7 | 537 (5.2) |
Robbery or carjacking | 408 | 398 | 49.4 | 806 (7.8) |
Burglary | 122 | 101 | 45.3 | 223 (2.2) |
Stolen vehicle | 79 | 57 | 41.9 | 136 (1.3) |
Suicidal or behavioral health crisis | 238 | 392 | 62.2 | 630 (6.1) |
Suspicious person or vehicle | 263 | 286 | 52.1 | 549 (5.3) |
Threats | 71 | 144 | 67.0 | 215 (2.1) |
Traffic stop | 789 | 811 | 50.7 | 1600 (15.5) |
Trespassing | 73 | 101 | 58.0 | 174 (1.7) |
Warrant or arrest | 376 | 592 | 61.2 | 968 (9.4) |
Weapon complaint | 203 | 245 | 54.7 | 448 (4.3) |
Well-being check | 54 | 101 | 65.2 | 155 (1.5) |
Othera | 150 | 203 | 57.5 | 353 (3.4) |
Unknown | 77 | 58 | 43.0 | 135 (1.3) |
Note. Includes on duty, both on and off duty, off duty but acting as on duty, and unknown duty status.
a Included within “other” incidents are vehicle collision, fire, hostage, involuntary commitment, pedestrian stop, vandalism, vehicle collision, escaped prisoner responses, immigration-related incidents, disaster responses, evictions, parole checks, dog complaints, fraud, and fare evasion.

TABLE 2— Fatally and Nonfatally Injured Persons, by Event or Person Characteristic: United States, 2015‒2020
Incident or Person Characteristic | Nonfatally Injured, No. | Fatally Injured, No. | % Fatal | Total Injured Persons, No. (%) |
Total | 4741 | 5874 | 55.3 | 10 615 (100) |
Person weapon | ||||
Unarmed | 478 | 477 | 49.9 | 955 (9.0) |
Firearm | 2418 | 3356 | 58.1 | 5774 (54.4) |
BB or replica gun | 154 | 210 | 57.7 | 364 (3.4) |
Total guna | 2572 | 3566 | 58.1 | 6138 (57.8) |
Knife or cutting/stabbing instrument | 491 | 1040 | 67.9 | 1531 (14.4) |
Vehicle | 495 | 279 | 36.0 | 774 (7.3) |
Blunt object | 76 | 123 | 61.8 | 199 (1.9) |
Other | 118 | 121 | 50.6 | 239 (2.3) |
Unknown | 511 | 268 | 34.4 | 779 (7.3) |
Ageb | ||||
Range, y | < 1–93 | 6–91 | ||
Mean of known ages, y (n = 9467; 59.8% fatal) | 33 | 37 | … | 35.4 |
Median of known ages, y (n = 9467; 59.8% fatal) | 30 | 35 | … | 33 |
Total juvenile count (< 18 y) | 212 | 105 | 33.1 | 317 (3.0) |
Total adult count (≥ 18 y) | 4315 | 5733 | 57.1 | 10 048 (94.7) |
Unknown | 214 | 36 | 14.4 | 250 (2.4) |
Gender | ||||
Cisgender man | 4369 | 5613 | 56.2 | 9982 (94.0) |
Cisgender woman | 287 | 248 | 46.4 | 535 (5.0) |
Transgender | 1 | 10 | 90.9 | 11 (0.1) |
Unknown | 84 | 3 | 3.4 | 87 (0.8) |
Race/ethnicity | ||||
Non-Hispanic White | 1106 | 2500 | 69.3 | 3606 (34.0) |
Non-Hispanic Black | 863 | 1363 | 61.2 | 2226 (21.0) |
Hispanic, any race | 424 | 1004 | 70.3 | 1428 (13.5) |
American Indian or Alaska Native | 23 | 105 | 82.0 | 128 (1.2) |
Asian or Pacific Islander | 25 | 106 | 80.9 | 131 (1.2) |
Otherc or multiple | 20 | 20 | 50.0 | 40 (0.4) |
Unknown | 2280 | 776 | 25.4 | 3056 (28.8) |
Unhoused person | 94 | 184 | 66.2 | 278 (2.6) |
Behavioral health involvement, incidentd | 793 | 1611 | 67.0 | 2404 (22.6) |
Note. Includes on duty, both on and off duty, off duty but acting as on duty, and unknown duty status.
a Includes “firearm,” “multiple with firearm,” and “BB or replica gun.” Does not include “service weapon concern,” which was only assessed at the incident level.
b Age was entered as specified, where applicable. Otherwise, age was categorized as juvenile (ages 0–17 years), adult (ages ≥ 18 years), or unknown.
c Includes Middle Eastern-North African.
d Behavioral health incidents include suicidal or self-harming behaviors, substance use, diagnosis of serious mental illness relevant to the incident, disability that may have been misinterpreted as a mental or behavioral health issue, and transportation or response to inpatient behavioral health facility.
Year | Nonfatally Injured, No. | Fatally Injured, No. | % Fatal | Total People Injured, No. |
2015 | 777 | 916 | 54.1 | 1693 |
2016 | 761 | 940 | 55.3 | 1701 |
2017 | 810 | 970 | 54.5 | 1780 |
2018 | 821 | 1026 | 55.5 | 1847 |
2019 | 765 | 990 | 56.4 | 1755 |
2020 | 807 | 1032 | 56.1 | 1839 |
Mean | 790 | 979 | 55.4 | 1769 |
Total | 4741 | 5874 | 55.3 | 10 615 |
In more than half of injurious shooting incidents, a nonofficer was armed with a firearm (56%; n = 5738); 4% involved nonofficer possession of a BB gun or replica gun (n = 403). Combined, 58% of these incidents involved a fatality. Knives were involved in 15% of incidents (n = 1543; 68% fatal), and a vehicle was reportedly weaponized against an officer in 8% of incidents (n = 806; 36% fatal). In another 8%, no weapon was involved (n = 785; 54% fatal). In 1.5% of incidents, a nonofficer reportedly gained control of a service weapon (n = 98) or nearly did so (n = 46; Appendix C, available as a supplement to the online version of this article at https://ajph.org).
Injurious shootings typically involved multiple police responders (81%; Appendix C), most frequently from a local police department (local police: 62%; sheriff’s office: 23%; Table 1). Fourteen percent of incidents with at least 1 fatality occurred during a single-officer response, compared to 18% of nonfatal-injury incidents (Appendix C). Dispatch by emergency services preceded 62% of injurious incidents; officer-initiated encounters preceded 36% of incidents. The most common reasons for police involvement before injurious shootings were traffic stops (16% of incidents, 51% fatal), domestic incidents (16% of incidents, 65% fatal), shots fired (9% of incidents, 55% fatal), and warrants (9% of incidents, 61% fatal). Suicidal crises represented 6% of injurious incidents (62% fatal). Rarer but more frequently fatal injurious shootings included well-being checks (2% of incidents, 65% fatal) and threats (e.g., an armed person verbalizing intent to harm; 2% of incidents, 67% fatal; Table 1).
In victim-level analysis, weapon status, shooting-agency type, response type, and incident type were proportionately similar to incident-level descriptions (Appendix C). Victims’ ages ranged from younger than 1 to 93 years; 95% were adults. Nonfatally injured people tended to be younger than fatally injured (nonfatal median age: 30 years; interquartile range [IQR] = 24–40 years; fatal median age: 35 years; IQR = 27–45 years). Sixty-seven percent of juveniles who were shot were not killed. Men and boys comprised 94% of victims. Race or ethnicity was identified for 71% of victims (n = 7559). When specified, 48% of people were described as non-Hispanic White (n = 3606; 69% fatal), 29% non-Hispanic Black (n = 2226; 61% fatal), and 19% Hispanic of any race (n = 1428; 70% fatal). Seventy-five percent of victims with unknown ethno-racial identities were nonfatally injured (Table 2). Among unarmed victims with assigned race-ethnicity, 40% were non-Hispanic White (n = 282), 35% were non-Hispanic Black (n = 245), and 21% were Hispanic (n = 144; fatal and nonfatal injuries included, data not shown). Nearly 3% of victims (n = 278) were unhoused, of whom 66% were fatally shot. Across incident types, 23% of injured people were shot in incidents involving mental or behavioral health issues (n = 2404; 67% fatal; Table 2). Forty-three percent of unhoused victims were injured in a such incidents (n = 120; data not shown).
Unadjusted logistic regression models suggest that compared with unarmed injured people (n = 955; 9%), odds of a fatal injury were significantly higher for injured people who were armed with a firearm (OR = 1.47; 95% confidence interval [CI] = 1.24, 1.74), BB or replica gun (OR = 1.43; 95% CI = 1.07, 1.91), knife (OR = 2.38; 95% CI = 1.91, 2.97), or blunt-force object (OR = 1.74; 95% CI = 1.20, 2.53). Odds of fatality were lower for injured people armed with a vehicle (OR = 0.52; 95% CI = 0.41, 0.66). Compared with shooting injuries during an officer-initiated interaction, odds of fatality were higher from injuries following dispatched interactions (OR = 1.46; 95% CI = 1.32, 1.63). Compared with injuries from police shootings following an on-view or dispatched “shots-fired” incident, odds of fatality were higher following incidents involving verbal threats (OR = 1.92; 95% CI = 1.32, 2.79), well-being checks (OR = 1.74; 95% CI = 1.14, 2.65), domestic incidents (OR = 1.72; 95% CI = 1.41, 2.11), suicidal or behavioral health crises (OR = 1.52; 95% CI = 1.19, 1.95), assaults (OR = 1.44; 95% CI = 1.08, 1.90), and warrant or arrest attempts (OR = 1.37; 95% CI = 1.11, 1.70). Odds of fatality were lower during traffic stops and other potentially vehicle-involved incidents (e.g., burglaries, robberies, or carjackings, and stolen vehicles). Incidents involving behavioral health concerns had 2.1-times-higher odds of fatal injury than injuries in incidents without such concerns (95% CI = 1.83, 2.45). Injuries from shootings by sheriff’s departments and state police were more likely to be lethal than injuries from shootings by local police departments (Table 4).

TABLE 4— Logistic Regression Models Predicting Odds of Fatal vs Nonfatal Injury: United States, 2015‒2020
Incident or Person Characteristic | OR (95% CI) | AOR (95% CI) |
Officer duty status | ||
On duty (Ref) | 1 | 1 |
On and off duty | 1.43 (0.30, 6.71) | 1.22 (0.28, 5.40) |
Off duty acting as on duty | 0.43 (0.27, 0.70) | 0.82 (0.50, 1.35) |
Unknown | 0.34 (0.21, 0.55) | 1.16 (0.67, 2.02) |
Person weapon | ||
Unarmed (Ref) | 1 | 1 |
Firearm | 1.47 (1.24, 1.74) | 1.37 (1.14, 1.65) |
BB or replica gun | 1.43 (1.07, 1.91) | 1.23 (0.90, 1.67) |
Knife or cutting/stabbing instrument | 2.38 (1.91, 2.97) | 1.92 (1.52, 2.44) |
Vehicle | 0.52 (0.41, 0.66) | 0.55 (0.42, 0.71) |
Blunt object | 1.74 (1.20, 2.53) | 1.43 (0.96, 2.13) |
Other | 1.03 (0.74, 1.43) | 0.82 (0.58, 1.16) |
Unknown | 0.47 (0.37, 0.60) | 0.72 (0.56, 0.93) |
Agency type | ||
Local police (Ref) | 1 | 1 |
Sheriff’s office | 1.24 (1.10, 1.40) | 1.26 (1.11, 1.42) |
State police | 1.32 (1.07, 1.64) | 1.45 (1.16, 1.82) |
National agency | 1.18 (0.82, 1.71) | 1.15 (0.79, 1.69) |
Special jurisdiction | 0.38 (0.21, 0.68) | 0.67 (0.38, 1.19) |
Constable or marshal | 0.66 (0.15, 2.90) | 0.80 (0.16, 3.94) |
Multiple shooting agencies | 1.79 (1.45, 2.22) | 1.57 (1.26, 1.96) |
Unknown | 0.38 (0.22, 0.67) | 1.12 (0.63, 2.01) |
Single officer response | ||
No (Ref) | 1 | 1 |
Yes | 0.69 (0.60, 0.79) | 0.79 (0.68, 0.91) |
Unknown | 0.38 (0.27, 0.52) | 0.75 (0.53, 1.07) |
Age | ||
Where specified (n = 9467)a | 1.03 (1.03, 1.04) | Not included in model |
Adult (≥ 18 y; Ref) | 1 | 1 |
Juvenile (< 18 y) | 0.32 (0.24, 0.44) | 0.63 (0.46, 0.86) |
Unknown | 0.09 (0.06, 0.15) | 0.52 (0.34, 0.81) |
Gender | ||
Cisgender man (Ref) | 1 | 1 |
Cisgender woman | 0.61 (0.48, 0.76) | 0.73 (0.58, 0.92) |
Transgender | 11.11 (1.12, 110.18) | 9.05 (0.89, 91.79) |
Unknown | 0.02 (< 0.01, 0.06) | 0.14 (0.04, 0.49) |
Race/ethnicity | ||
Non-Hispanic White (Ref) | 1 | 1 |
Non-Hispanic Black | 0.65 (0.56, 0.76) | 0.85 (0.74, 0.98) |
Hispanic, any race | 1.07 (0.91, 1.25) | 1.22 (1.04, 1.44) |
American Indian or Alaska Native | 2.29 (1.33, 3.95) | 2.55 (1.48, 4.37) |
Asian or Pacific Islander | 2.12 (1.25, 3.58) | 2.20 (1.32, 3.68) |
Otherb or multiple | 0.38 (0.17, 0.83) | 0.46 (0.22, 0.97) |
Unknown | 0.10 (0.07, 0.15) | 0.14 (0.10, 0.20) |
Unhoused | ||
No or unknown (Ref) | 1 | 1 |
Yes | 0.99 (0.99, 1.00) | 1.00 (1.00, 1.00) |
Response type | ||
On view (Ref) | 1 | 1 |
Dispatched to 911 call | 1.46 (1.32, 1.63) | 1.32 (1.11, 1.56) |
By subject | 1.00 (0.63, 1.59) | 0.76 (0.46, 1.25) |
Unknown | 0.58 (0.37, 0.91) | 1.11 (0.62, 1.98) |
Incident type | ||
Shooting (Ref) | 1 | 1 |
Assault | 1.44 (1.08, 1.90) | 1.18 (0.87, 1.61) |
Disorderly conduct or dispute or disturbance | 1.25 (0.94, 1.65) | 1.24 (0.92, 1.68) |
Domestic incident (disturbance, dispute, or violence) | 1.72 (1.41, 2.11) | 1.43 (1.15, 1.77) |
Investigative | 0.93 (0.73, 1.19) | 1.18 (0.88, 1.58) |
Robbery or carjacking | 0.75 (0.60, 0.94) | 0.92 (0.73, 1.16) |
Burglary | 0.66 (0.46, 0.93) | 0.66 (0.46, 0.95) |
Stolen vehicle | 0.51 (0.33, 0.79) | 0.95 (0.60, 1.50) |
Suicidal or behavioral health crisis | 1.52 (1.19, 1.95) | 1.09 (0.83, 1.43) |
Suspicious person or vehicle | 0.89 (0.70, 1.14) | 1.11 (0.85, 1.46) |
Threats | 1.92 (1.32, 2.79) | 1.89 (1.26, 2.82) |
Traffic stop | 0.81 (0.67, 0.98) | 1.21 (0.95, 1.54) |
Trespassing | 1.13 (0.77, 1.65) | 1.29 (0.86, 1.95) |
Warrant or arrest | 1.37 (1.11, 1.70) | 1.66 (1.26, 2.18) |
Weapon complaint | 1.03 (0.79, 1.34) | 0.91 (0.69, 1.21) |
Well-being check | 1.74 (1.14, 2.65) | 1.32 (0.85, 2.05) |
Otherc | 1.16 (0.87, 1.55) | 1.26 (0.92, 1.72) |
Unknown | 0.60 (0.39, 0.93) | 1.49 (0.86, 2.60) |
Behavioral health‒related incidentd | ||
None (Ref) | 1 | 1 |
Any | 2.12 (1.83, 2.45) | 1.41 (1.22, 1.63) |
Note. AOR = adjusted odds ratio; CI = confidence interval; OR = odds ratio. Odds predicted with 2-level mixed-effects logistic regression models. Includes on duty, both on and off duty, off duty but acting as on duty, and unknown duty status.
a OR represents change in odds of fatality for each additional year of victim age.
b Includes Middle Eastern-North African.
c Included within “other” incidents are fire, hostage, vandalism, vehicle collision, pedestrian stop, involuntary commitment, subject-initiated, escaped prisoner responses, immigration-related incidents, disaster responses, evictions, parole checks, dog complaints, fraud, and fare evasion.
d Behavioral health incidents include suicidal or self-harming behaviors, substance use, diagnosis of serious mental illness relevant to the incident, disability that may have been misinterpreted as a mental or behavioral health issue, and transportation or response to inpatient behavioral health facility.
Demographically, odds of fatality increased by 3% with each year of victim age (95% CI = 1.03, 1.04) and were lower for injured women compared with men (OR = 0.61; 95% CI = 0.48, 0.76). Among people with identified race/ethnicity, odds of fatality were lower among non-Hispanic Black victims (OR = 0.65; 95% CI = 0.56, 0.76) compared with non-Hispanic White victims, and higher among American Indian or Alaska Native victims (OR = 2.29; 95% CI = 1.33, 3.95) and Asian or Pacific Islander victims (OR = 2.12; 95% CI = 1.25, 3.58). In the adjusted model, Hispanic victims were also estimated to have higher odds of fatality (OR = 1.22; 95% CI = 1.04, 1.44), and fewer incident types were statistically significantly associated with fatal injury. All other inferences were unchanged (Table 4).
Estimates calculated from alternative duty-status inclusion criteria varied rarely and minimally from the main analysis. Results based on more restrictive or maximally inclusive on-duty status criteria are provided in Appendix D (available as a supplement to the online version of this article at https://ajph.org).
In this study of 1769 annual injuries from shootings by police over a 6-year period, 45% of injured persons were nonfatally injured, consistent with previous estimates from 4 states’ mandated reporting.10 Compared with estimates drawn from fatal shootings only, victim and incident characteristics were proportionately similar in categorical age, gender, involvement of unarmed victims, and other characteristics.18 However, when nonfatally injured people were included, proportionately more victims were identified as non-Hispanic Black. Case fatality rates varied by incident characteristics. Few would disagree that a fatality is the most severe and irreversible potential outcome of a shooting, but nonfatal injuries are also physically and psychologically impactful. Situations with low case fatality rates are among the most underexamined incidents in previous research on fatal shootings by police.
Incidents with high case fatality rates generally involved complaints of physically threatening or threat-making behaviors (e.g., assaults, verbalized threats, domestic incidents, suicidal and self-harming incidents). Threat perception among police may be amplified by a prominent, and often racialized, emphasis on threat anticipation and officer self-protection in US policing culture and training.23 Absent explicit threats, officers may anticipate increased threat during incidents such as traffic stops or domestic violence episodes, which are more frequently associated with police occupational homicides.24 One potential exception to this pattern in threat-related, more frequently fatal injuries was well-being checks. Well-being checks were 74% more likely to be associated with fatal injury, despite not explicitly or necessarily involving pre-encounter threats of harm. In these cases, the probable involvement of callers and dispatchers may be a source of relayed alarm, prompting readiness for threat perception.25,26 Of all injuries, 61% followed dispatched incidents; these injuries were 1.46-times more likely to be fatal than injuries following on-view responses.
Among injured people, victims identified as non-Hispanic Black comprised 29% of race-identified injured people in this study. This compares to 26% of victims in a single-year sample of fatal shootings and 13% of the total US population.18,27 Injured non-Hispanic Black people had 35% higher odds of surviving than non-Hispanic White injured people. Police may be more apt to fire shots that nonfatally injure people whom they perceive as Black because of biased assumptions of criminality that, in combination with amplified threat perception, may lead to more impulsive, emotional, longer-distance, or otherwise less-accurate shots. Racial disparities in most policing judgments and interactions are well-known.6,28,29 Still, for incidents that may be dismissed as rare, such as shootings by police, underestimating the true scale of injury impact is a further injustice and may obstruct progress toward preventive action and reforms.
Also relatively underexamined in previous research are injuries among people who were unhoused or experiencing symptoms of behavioral health conditions. Unhoused victims comprised nearly 3% of injured people, despite representing just 0.2% of the US population.30 Behavioral health needs were associated with 23% of injured persons; they were twice as likely to die from their injuries as other victims. These represent instances in which not only are “the marginalized … further criminalized” but they are also victimized by a system that is inadequately designed to meaningfully address social needs.26(p771)
Mechanisms for less potentially injurious triaging of social services exist. In 2022, the National Suicide Prevention Lifeline 988 was introduced, yet complementary local systems for improved access to social services without entrenched criminal legal system involvement (e.g., nonpolice mobile units) remain uncommon31 despite strong public health alignment.32 Public support is high for alternative approaches (e.g., diversion to mental health services, police and mental health co-responder models),4 but cost remains a barrier to more widespread implementation.31 Future analysis of incidents at the intersection of dispatched responses and social or behavioral health needs may inform feasibility, design, outreach, and equity-oriented impact analyses of new crisis-support systems.
This study affirmed and expanded upon previous understandings of shootings by police in the United States, providing the first estimate of total injured persons nationally over multiple years. With this larger data set, previously excluded states and relatively rarer incident types could be examined. Still, some limitations exist. First, police perspectives (themselves often reconstructed “observations of observations,”33(p146) which may be subject to recall and social desirability bias) are known to be overrepresented in media accounts of shootings by police.34 To diversify considered narratives, abstractors reviewed multiple sources, including bystander accounts, surveillance videos, legal documents, and articles not linked to the original GVA record. In addition, the study’s inclusion period was defined to allow case details to develop and be represented. Still, some reporting bias is likely. More subjective variables, such as those involving interpretation of intent (e.g., declaring a vehicle weaponized or a service weapon nearly acquired), may be especially subject to dominant narratives and should be interpreted accordingly.
Second, the use of media sources inherently relies on assumptions of newsworthiness, adequate reporting capacity, and resulting news coverage, which may vary by time and place. Consistency with previous studies’ fatal injury estimates is assuring of source validity. Still, nonfatal injuries may be less consistently or less thoroughly reported, leading to nonrandom missingness. Counts of “unspecified” or “unknown” characteristics are signals of underreporting, highlighting continued need for mandatory surveillance of all-outcome shootings by police. The relatively more developed repertoire of open-source repositories for fatalities adds to known information asymmetry. This limitation restricted our ability to precisely calculate national injury disparities. Still, our estimates of fatal and nonfatal injuries, though conservative, are substantial improvements over previous projections of total and subgroup injury burden.
Finally, we only examined injurious shootings in this study; other mechanisms of deadly force exist, and nonfatal shootings without injury were not included. Future research should analyze determinants of survival. This analysis did not account for differences in frequency of policing activities or the unequal distribution of risk in the prerequisite condition of encountering police. Disparities were interpreted on a per-capita basis, but results may not reflect individual risk for injury.
In 2002, American criminologist James Fyfe observed, “ours is a democracy that does not tell us how often we are forcibly injured or killed by the people we pay to protect us.”35(p88) Twenty years later, despite ongoing criticism and controversy surrounding use of deadly force by police, US accountability systems remain persistently inadequate. Nonfatal injurious shootings by police are governed by the same use-of-force policies as fatal shootings and appear similar in frequency and circumstance. However, the historical exclusion of nonfatal injuries from surveillance and research has led to underestimated injury disparities and underexamined shooting incidents, particularly at the margins of policing. Of all injurious shootings by police, incidents involving well-being checks, behavioral health concerns, suicidal crises, and unhoused persons were among the most frequently fatal. Inadequate services for people who are unhoused, insufficient supports for managing mental illness and substance use, and inequitable social and economic protections for minoritized populations are potential areas for priority response.11,36 Evaluations of emerging public safety reforms should monitor fatal and nonfatal shootings by police to assess impact overall and among disproportionately affected groups.
Additional research is needed regarding the role of societal firearm prevalence in shootings by police, characteristics of shootings in rural and other historically underexamined regions, the role of decision-making in single- and multiple-officer responses, and frequency of noninjurious shootings. Researchers and justice advocates would also benefit from analyses of how and for whom publicly known contextual details of police shootings evolve. Finally, improved and sustained investments in reliable data and accountability systems remain essential to the prevention of firearm injuries from armed policing responses.
See also Nix, p.
ACKNOWLEDGMENTS
Funding to support initial data collection and analysis was provided by New Venture Fund and Joyce Foundation. Additional funding was provided by the Robert Wood Johnson Foundation. J. A. Ward acknowledges support from the National Institute of Child Health and Human Development (T32-HD 094687). J. Cepeda acknowledges support from the National Institute on Drug Abuse (K01DA054521).
This project benefited from the commitment and persistence of a team of data abstractors that included Vaishu Bandaru, Ethan Bartlett, Makenzie Bozer, Annie Brown, Kat Catamura, Brooke Dal Santo, Brandon Hardy, Taylor Johnson, Hami Kang, Nick Meyerson, Isabella Parea, Delilah Ponce, Martina Racioppi, and Alissa Zhu. We thank them for their efforts.
Note. Funders had no role in the study design, statistical analyses, interpretation of data or drafting of this article.
CONFLICTS OF INTEREST
The authors declare no actual or potential conflicts of interest.
HUMAN PARTICIPANT PROTECTION
This study was reviewed and approved by the Johns Hopkins Bloomberg School of Public Health institutional review board.