Firearm Prevalence and Homicides of Law EnforcementOfcers in the United StatesDavid I. Swedler, PhD, MPH, Molly M. Simmons, AB, Francesca Dominici, PhD, and David Hemenway, PhDHomicide rates are higher in the United Statesthan in other high-income countries.1---3Al-though the rate of nonfatal assaults in theUnited States is similar to its peer nations, theprevalence ofrearms appears to be a reasonfor this increased homicide rate.4,5The homi-cide rate in the United States varies by regionand by state. When controlling for demo-graphic and economic factors within states,rearm ownership rates and homicide ratesare positively correlated.6,7For the majority of working Americans,being murdered on the job, known as occupa-tional homicide, has never been a major con-cern. In addition, like homicide rates generally,occupational homicide rates have been declin-ing nationwide for the past 2 decades.8How-ever, although occupational homicides are rare,they do constitute a high percentage of fataloccupational injuries; in 2012, occupationalhomicides accounted for 17% of fatal occupa-tional injuries.9Law enforcement ofcers(LEOs) have an occupational homicide rate 3times the national average, and it is thesecond leading cause of occupational mor-tality for this group, behind motor vehiclecollisions.10,11It is possible that homicides of LEOs aredriven by criminal offender theory: morefrequent encounters with motivated violentoffenders are the root cause of LEO homiciderates.12In a study of 190 agencies, Kaminskifound that violent crime rate was associatedwith ofcer homicides.13However, LEOs arerecruited, equipped, and trained to encountersuch dangerous situations.14Agency-levelpolicies are also designed to keep ofcers safewhen encountering suspects. Their trainingalso dictates how to avoid escalating hostileencounters into potentially life-threatening sit-uations.15LEOs are often equipped with bodyarmor, greatly increasing the likelihood thatthey will survive being shot.16---18Finally, LEOsin the United States carryrearms, usuallya handgun.19Despite the presence of specialized trainingand equipment, many LEOs are still murderedon the job. More than 90% of these homicidesare committed using arearm.20An assailantmerely needs arearm, rudimentary knowl-edge of howrearms work, and sufcientopportunity to fatally wound any superiorlytrained and equipped ofcer.5A higher pres-ence of guns, then, creates more opportunitiesfor LEO deaths. Small-scale studies have foundthat gun density in cities is correlated with gunhomicide and gun suicide of LEOs in theUnited States.21An international comparison ofLEO mortality in New York City and London,United Kingdom, found that 20 times as manyofcers died from intentional gunshot woundsin New York compared with London, wherepersonal rearm ownership is markedlylower.22A previous study found that takeawayhomicides, a type of homicide where ofcersare killed by their own service weapons, madeup only 10% of LEO homicides, meaning thatthe majority of LEOhomicides were committedwith a privately owned gun.20We also knowthat state laws have the potential to increase ordecreaserearm ownership by private citi-zens23andrearm homicides caused by pri-vately owned weapons.24---26We examined therelationship between staterearm ownershiprates and LEO occupational homicide rates,rst controlling for the state violent crime rate,and then adding other state-level factors knownto be associated with homicide rates in thegeneral population. We hypothesized thatrearm ownership would be positivelycorrelated with homicide rates of LEOs.METHODSAs part of its Uniform Crime Reporting(UCR) program, the Federal Bureau of Investi-gation (FBI) annually reports all homicides ofLEOs in the United States.19To be includedas a fatality in the reporting, an individual mustbe aduly sworn [ofcer] feloniously or acci-dentally killed or assaulted in the line ofduty.19(p109)Thisline of duty designationdescribes on- or off-duty LEOs acting in anofcial capacity (i.e., acting as if they would inObjectives. In the United States, state rearm ownership has been correlatedwithhomiciderates.Morethan90%ofhomicidesoflawenforcementofcers(LEOs) arecommittedwithrearms. Weexaminedtherelationshipbetweenstate rearm ownership rates and LEO occupational homicide rates.Methods. We obtained the number LEOs killed from 1996 to 2010 from a FederalBureauofInvestigation(FBI)database. Wecalculatedhomicideratesperstateasthe number of ofcers killed per number of LEOs per state, obtained from anotherFBI database. We obtained the mean household rearm ownership for each statefrom the Behavioral Risk Factor Surveillance System.Results. Using Poisson regression and controlling for factors known to affecthomiciderates, weassociatedrearmownershipwiththehomicideratesforLEOs (incidence rate ratio=1.044; P=.005); our results were supported bycross-sectional and longitudinal sensitivity analyses. LEO homicide rates were3timeshigherinstateswithhighrearmownershipcomparedwithstateswithlowrearmownership.Conclusions. High public gun ownership is a risk for occupational mortality forLEOs inthe UnitedStates. States couldconsider methods for reducingrearmownershipas a way to reduce occupational deaths of LEOs. (AmJ Public Health. Published onlineahead of print August 13, 2015: e1e7. doi:10.2105/AJPH.2015.302749)RESEARCH AND PRACTICEPublishedonlineaheadofprintAugust13, 2015| AmericanJournal ofPublicHealth Swedleretal. | PeerReviewed | ResearchandPractice | e1their ofcial duties as an LEO). This denitionincludes local, county, state, college or univer-sity, tribal, and federal agencies that are com-posed of police ofcers, sheriffs and deputies,highway patrol ofcers, marshals, and specialagents; probation ofcers, correctional ofcers,jailers, and prison ofcials are excluded.11,19We used the UCR because we focused onsworn ofcers who conduct policing activitiesin public spaces.11The UCR database is moreconservative in estimating occupational fatali-ties and homicides compared with other na-tional surveillance systems11; however, the UCRdata are meant to serve as a census of allhomicides of LEOs in the country, so everytime an ofcer homicide is identied, it must bereported to the UCR.19We downloaded thedata as Supplemental Homicide Reports, whichare made available through the National Archiveof Criminal Justice Data.27We collected thehomicide counts annually for each state and theDistrict of Columbia from1996 through 2010.To calculate homicide rates, we obtainedannual estimates of LEOs employed in eachstate from another FBI database. The PoliceEmployee database collects information on alllaw enforcement agencies in the United States,including the number of ofcers employedeach year.27We totaled the number of swornofcers employed in each agency in each statefor each year. We calculated the homicide ratesas the number of occupational homicides per10 000 LEOs annually for each state. Becausehomicides were rare events at the individualstate-year level, we combined homicides andnumber of ofcers per state across the studyperiod. Using Poisson regression, our outcomeof interest was the homicide count per statewith the number of ofcers (per 10 000) as theoffset term. In essence, the LEO homicide rateper 10000 ofcers during the 15-year studyperiod (1996---2010) was the dependent vari-able in the main analysis.The main independent variable was statehouseholdrearm ownership rate. Althoughmany factors affect the lethality ofrearmassaults, such as intent, location of injury, andcaliber,16,28,29we did not consider the effects ofthese factors on LEO homicides. A commonlyused direct measure of publicrearm owner-ship comes from the Behavioral Risk FactorSurveillance System (BRFSS).30,31This annualnationwide survey collects data over a widerange of health topics. Questions on householdrearm ownership were only included in theBRFSS in 2001, 2002, and 2004. To calculatethe staterearm ownership rate, we averagedthe prevalence ofrearm ownership from theBRFSS across the 3 surveys. Mean householdrearm ownership rate (calculated from therates in 2001, 2002, and 2004 BRFSS data)was the primary independent variable in themain regression analysis. Our second indepen-dent variable of interest was state violent crimerates. Violent crime rates for each state-yearwere obtained from the FBIs annual Crime inthe United States reports.32Because the BRFSSrearm data were onlyavailable for 3 years, we also collected a proxyofrearm ownership. The Web-based InjuryStatistics Query and Reporting System reportsdata on completed suicides in the United States.Firearm suicides as a percent of all suicides hasbeen used as an annual measure ofrearmownership rate.6,31,33This term is known asrearm suicides/suicides (FS/S). We used theseannual data as sensitivity analyses, described inthe following section, to examine the longitu-dinal effects ofrearm ownership on LEOhomicide rates during the study period.Data AnalysisWe calculated descriptive statistics forhomicide counts and rates per 10 000 LEOs.We calculated homicide counts and rates forstates over the entire 15-year study period andfor individual state-years. We compared LEOhomicides rates in the highestrearm preva-lence states with those in the states with thelowestrearm prevalence. To do this compar-ison, we selected from the highestrearmprevalence states and the lowestrearm prev-alence states until the number of LEO-yearswas approximately equal.34,35Because thelow-prevalence states were typically morehighly populated and had many more ofcersthan the high-prevalence states, the nal 2-by-2analysis had 8 low-prevalence states and 23high-prevalence states, covering approximately2.75 million LEO-years per group.Homicide rates followed a count distribu-tion. A likelihood ratio a test for both Poissonand negative binomial data conrmed thatthe datat a Poisson distribution.36Firearmownership rates and select covariates wereregressed on the homicide rate using Poissonregression with the number of LEOs (per10 000) as the offset term.37Wet thefollowing Poisson model:1log EYs log ns bxscovariates;where Ysis the number of LEOs homicideswithin each state, nsis the number of LEOswithin each state, xsis the dependent variableof interest (state-specicrearm ownership orthe proxy), and b is the log relative rate of LEOhomicide rate for a unit change inrearmownership. For our initial regression, the vio-lent crime rate was the only covariate includedin the model withrearm ownership; we re-ferred to this as model 1. We then ran a modelincluding selected covariates in Table 1 usinga method described in the following section;we referred to this as model 2.We averaged the variables in Table 1 acrossthe entire study period. Werst analyzed thecorrelation of each covariate to LEO homiciderates with a Spearman q, listed in Table 1. Werst selected any covariates with an absoluteq of 0.300 or greater. Second, we excluded anycovariates that were highly correlated withother included covariates from thenal model.All of the covariate correlations can be found asa supplement to the online version of thisarticle at http://www.ajph.org. All data wereanalyzed using Stata version 12.1 (Stata Corp,College Station, TX).Selection of Covariates for RegressionAnalysisBecause little research has been conductedon state-level factors affecting homicides ofLEOs, we looked to the general homicideliterature for potential control variables. Siegelet al. have produced a series of studies exam-ining the role ofrearm ownership on differenthomicide outcomes at the state level.6,38,39From these previous studies, we selected thefollowing state-level variables as potential con-founders of the association betweenrearmownership and the LEO homicide rate: prop-erty crime rates, percentage of the populationaged 15 to 29 years, percentage of the pop-ulation that is African American, percentage ofthe population that is Hispanic, poverty rate,median income, educational attainment level,alcohol consumption, divorce rate,; incomeinequality, and percentage of the populationin urban areas (urbanicity).RESEARCH AND PRACTICEe2| ResearchandPractice | PeerReviewed | Swedleretal. AmericanJournal ofPublicHealth | PublishedonlineaheadofprintAugust13,2015Table 1 lists the variables collected aspotential controls in the regression analysis, thevariable denitions, and the source for eachvariable. We collected all data for each stateannually, aside from the income inequality andurbanicity data, which were only available for1999 and 2000, respectively.Sensitivity AnalysesWe tested the robustness of ourndingswith various sensitivity tests. First, the FS/Sproxy measure ofrearm ownership rate wassubstituted as the independent variable ofinterest in a Poisson regression. Second, homi-cide rates were calculated as the number ofLEO homicides per 1000000 state population,rather than per LEO-years. Third, althoughrearms were used in more than 90% of LEOhomicides, we examined the association of staterearm ownership rates andrearm homicidesof LEOs. These analyses were all cross sectionalacross the entire study period.Because a cross-sectional analysis presentsa lower level of evidence, we conducted atime-series sensitivity analysis by dividing thestudy period into three 5-year time blocks:1996 to 2000, 2001 to 2005, and 2006 to2010. These 5-year divisions in a time-seriesanalysis allowed us to draw stronger conclu-sions concerning the relationship betweenrearm ownership and LEO homicides withoutinappropriately increasing the variance of FS/S.This longitudinal analysis also allowed us thechance to observe the relationship betweenrearm ownership rates and LEO homicidesas they changed over time. The longitudinalmultilevel Poisson model usedxed effectswithin states for rearmownership rates and theselected covariates, as was used in previousanalyses of the relationship betweenrearmownership rates and general homicide rates.6,38We averaged each covariate across the respec-tive 5-year blocks.RESULTSThere were 782 homicides of LEOs overthe study period, 716 (92%) of which werecommitted withrearms. Handguns were usedin 515 homicides (72% ofrearm homicides,or 66% of all homicides). States averaged morethan 15 LEO homicides over the study period,or approximately 1 per year. States averaged15861 LEOs employed per year, or 237915LEO-years over the whole study period. Per10000 LEOs, homicide rates were 0.76 and0.85 for the entire study period and annually,respectively. Table 2 lists the mean, SD, andrange for homicide counts and rates for theentire study period and for each year. Californiaand Texas were the only states to experienceat least 1 homicide in every year of the study.The states with the most homicides wereCalifornia (n =77), Texas (n =70), Florida(n =39), Georgia (n =36), and North Carolina(n =33). Iowa, Maine, Vermont, and WyomingTABLE 1Variables Considered as Possible Covariates for Regression Analysis: United States, 19962010Covariate Denition SourceSpearman q for LEOHomicide Rate SelectedViolent crime rate Per 100 000 population Federal Bureau of Investigation 0.390 YesProperty crime rate Per 100 000 population Federal Bureau of Investigation 0.457 NoPopulation age 1529 y Percentage Census Bureau 0.391 YesAfrican American population Percentage Census Bureau 0.300 YesHispanic population Percentage Census Bureau 0.120 NoPoverty rate Percentage of population living in poverty Census Bureau 0.448 NoMedian income Tens of thousands 2012 US$ Census Bureau 0.360 YesEducation level Percentage of adult population with college degree Census Bureau 0.278 NoAlcohol consumption Per capita alcohol consumption, gallons of ethanol National Institute on Alcohol Abuse and Alcoholism 0.070 NoDivorce rate Per 1000 population National Center for Health Statistics 0.301 YesIncome inequality Gini coefcient (1999 only) Census Bureau 0.136 NoUrbanicity Percentage of state population in urban areas andurban centers (2000 only)Statistical Abstract of the United States 0.178 NoNote. LEO = law enforcement ofcer.TABLE 2Rates and Counts for Homicides and Firearm Homicides of Law EnforcementOfcers per State for Total Study Period and Single Years: United States, 19962010Counts Rate per 10 000 LEOsVariables Mean (SD) Range Mean (SD) RangeTotal study periodHomicide 15.1 (15.67) 074 0.78 (0.63) 04.3Firearm homicide 14.0 (14.9) 072 0.68 (0.50) 03.0Annual dataHomicide 1.0 (1.5) 010 0.85 (1.7) 016.6Firearm homicide 0.9 (1.4) 09 0.75 (1.5) 013.9Note. LEO = law enforcement ofcer.RESEARCH AND PRACTICEPublishedonlineaheadofprintAugust13, 2015| AmericanJournal ofPublicHealth Swedleretal. | PeerReviewed | ResearchandPractice | e3experienced zero homicides of LEOs duringthe study period. Figure 1 displays states byLEO homicide rate quintile.Firearm OwnershipUsing the average of the 2001, 2002, and2004 BRFSS data as the source ofrearmownership, the meanrearm ownership ratewas 38.3% (SD=14.5%). Average ownershipranged from 4.8% (District of Columbia) to62.0% (Wyoming). The estimate of meanrearm ownership rates from FS/S for these3 years was highly correlated with the meanBRFSS estimate (Pearson correlation =0.861).Figure 2 displaysrearm ownership rates fromthe BRFSS by state quintiles. The state-levelPearson correlation betweenrearm owner-ship and violent crime rates was0.368.Table 3 compares states with the lowestrearm ownership rates with those states withthe highest ownership rates. For the 15-yearstudy period, both the low and high gun statescovered approximately 2.75 million LEO-years. The incident rate for LEO homicidesin the high gun states was 3.11 times greater(95% condence interval =2.42, 4.03) than inthe low gun states.Firearm Ownership and Homicides of LawEnforcement OfcersUsing Poisson regression forrearm owner-ship and violent crime rates on LEO homiciderates (model 1), the incidence rate ratio (IRR)for increasingrearm ownership rate by 1%was 1.035 (P =.011), and the IRR for theviolent crime rate was 1.002 (P =.002). Formodel 2, in addition torearm ownership andthe violent crime rate, the variables enteredinto the regression model were the percentageof the population ages 15 to 29 years, medianincome, divorce rate, and the percent of thepopulation that is African American. In multi-variable regression with the 6 covariates se-lected in Table 1, the adjusted IRR forrearmownership was 1.041 (P =.009), and the IRRfor the violent crime rate was 1.002 (P =.073;Table 4). Putting this result into context, be-cause states averaged 237915 LEOs over thestudy period, a 10% increase inrearm own-ership rate would result in 10 additionalLEO homicides over 15 years. Table 4 displaysthe results for model 1 and for model 2, inwhich therearm ownership rate was the onlystatistically signicant covariate.Sensitivity analyses supported thisnding.The full results of the sensitivity analyses arepresented as data available as a supplement tothe online version of this article at http://www.ajph.org. The signicant results of the respectivemodel 2 are presented here. Using the FS/Srearm ownership proxy, the IRR forrearmownership on LEO homicide rate was 1.050(95% CI =1.008, 1.095). A second sensitivityanalysis considered LEO homicide rate asa function of state population. No covariateswere signicantly associated with LEO homi-cides rates per state population. Becausere-arms were used in 92% of homicides ofLEOs, the effect ofrearm ownership rate onrearm-only homicides of LEOs was similarto the main analysis (IRR=1.036, 95%CI =1.002, 1.071). Dividing the study periodinto 5-year blocks and conducting a longitudi-nal analysis resulted in a similar correlation forrearmownership as seen in the cross-sectionalanalyses (adjusted IRR=1.054; 95%CI =1.030, 1.078) state violent crime rates werealso signicantly associated with LEO homiciderates in the time-series analysis (IRR=1.002;95% CI =1.001, 1.003). In the sensitivity ana-lyses, no other covariates were associated withLEO homicide rates with a single exception:median income was positively correlated with ahigher homicide rate in the time-series analysis(IRR=1.429; 95% CI =1.112, 1.836).DISCUSSIONOur study found that the occupationalhomicide rate for LEOs was positively correlatedwithrearm ownership rate. We found cleardifferences in LEO homicide rates betweenstates with low and highrearm ownership.District ofColumbiaQuintileBottom 20%20%40%40%60%60%80%Top 20%FIGURE 1Quintiles of law enforcement ofcer homicide rates: United States, 19962010.District ofColumbiaQuintileBottom 20%20%40%40%60%60%80%Top 20%FIGURE 2Quintiles of rearm ownership: United States, 19962010.RESEARCH AND PRACTICEe4| ResearchandPractice | PeerReviewed | Swedleretal. AmericanJournal ofPublicHealth | PublishedonlineaheadofprintAugust13,2015Exposure to guns is an occupational riskfactor for LEOs. In our analysis, we controlledfor state violent crime rates, which indicatedthat the higher risk of LEO homicide victimi-zation in high gun states was not simplybecause these ofcers more frequently en-countered violent situations. Siegel et al. foundthat the violent crime rate was associated withthe homicide rate of the general population6;we also found that the violent crime rate wasassociated with homicides of LEOs, but this lostits statistical signicance when we includedother covariates in the regression models.High rates of LEO homicide victimizationappeared to be caused by more frequent en-counters with violent criminals and by morefrequently encountering situations where pri-vately ownedrearms were present. For exam-ple, we knowthat ofcers were often killed uponarriving at a residence for a domestic distur-bance call.40,41Encountering more domesticdisturbance calls along with an increased likeli-hood that the household would contain rearmsundoubtedly increased the homicide risk forLEOs. Thus, domestic disturbance calls in stateswith higher householdrearm ownership ratespresent increased opportunities where LEOswould encounter a potentially deadly situation.Previous research indicated that the rate ofprivate gun ownership23,42and homicidescommitted with privately owned weapons weremodiable through changes to state laws.24,25,43Our research suggests that there is the poten-tial to reduce LEO homicide rates throughchanges to state laws.When selecting variables for inclusion inmultivariate regression, we referenced previ-ous literature investigating the relationshipbetween state-level gun ownership rates andoverall state homicide rates. We did this be-cause we could not locate any studies thatexamined state-level risks for LEO homicidesspecically. Covington et al. analyzed the risksfor ofcer assaults using logistic regression;however, this analysis was restricted to a singleagency and only covered offender, ofcer, andsituational variables.44Kaminski et al. con-ducted a series of analyses using methodssimilar to our study; however, their level ofanalysis was census blocks or law enforcementagencies, not states.13,45In determining theappropriate parsimonious model, some vari-ables that had previously been associated withhomicide rates were excluded. For example,level of urbanization within a state was pre-viously found to be a confounding variablewhen assessing staterearm ownership andgun homicides.38However, this predictor hadvery low correlation with the outcome in ourstudy (Spearman q =0.178). Ourndingsindicate the need for further research intostate-level risk factors for LEO homicides.When we examined geographic variation ofLEO homicide rates andrearm ownershiprates in Figures 1 and 2, respectively, somestates fell in the same quintiles in both graphs,but the relationships were not perfect.Alabama, Alaska, Arkansas, Mississippi, andMontana were in the top quintile for LEOhomicides andrearm ownership, whereasConnecticut, Massachusetts, New Jersey, NewYork, and Rhode Island were in the lowestquintile in bothgures. These correlationswithin the top and bottom quintiles ofrearmownership supported thendings of Table 3.However, not all states supported a positivecorrelation between the 2 variables. For ex-ample, Wyoming had the highestrearmownership rate and zero LEO homicides,whereas the District of Columbia had thelowestrearm ownership rate and was inthe highest quintile for homicide rates. Itappeared that both the state levels ofrearmprevalence and violent crimes affected the LEOhomicide rate; however, further research isnecessary to identify other state-level predic-tors of LEO homicide rates.LimitationsOur study had various limitations. The mainstatistical analysis used cross-sectional data.Although a single cross-sectional analysis couldnot produce a conclusive link betweenrearmownership and the LEO homicide rate, thefact that multiple sensitivity analyses producedsimilarndings supported this link. Thetime-series analysis allowed us to test for thepresence of reverse causality in ourndings.For example, if a state experienced a highnumber of LEO homicides early in the studyperiod, which then led state residents to obtainTABLE 3Homicides of Law Enforcement Ofcers in the 8 States With the Lowest andthe 23 States With the Highest Household Firearm Ownership Rates: United States,19962010Characteristics Low Gun States High Gun StatesAverage household rearm ownership levels, % 13.5 52.0LEOs employed 19962010, no. 2 777 567 2 759 590Total homicides 19962010, no. 85 263Homicide rate per 10 000 LEOs 0.31 0.95Note. LEO = law enforcement ofcer. Low gun states: CT, DC, HI, IL, MA, NJ, NY, RI. High gun states: AL, AK, AR, IA, ID, KS, KY,LA, MN, MO, MS, MT, NC, ND, OK, SC, SD, TN, UT, VT, WI, WV, WY.TABLE 4Multivariate Poisson Regression of Firearm Ownership and Law EnforcementHomicide Rates: United States, 19962010Variable Model 1, IRR (95% CI) Model 2, IRR (95% CI)Firearm ownership 1.035 (1.008, 1.063) 1.041 (1.001, 1.072)Violent crime rate 1.002 (1.001, 1.003) 1.002 (1.000, 1.004)% population aged 1529 yMedian income 1.374 (0.876, 2.155)Divorce rate 1.070 (0.717, 1.599)% population African American 1.008 (0.968, 1.049)Note. CI = condence interval; IRR = incidence rate ratio. Model 1 includes only rearm ownership rate and violent crimerates. Model 2 adds in select covariates.RESEARCH AND PRACTICEPublishedonlineaheadofprintAugust13, 2015| AmericanJournal ofPublicHealth Swedleretal. | PeerReviewed | ResearchandPractice | e5rearms for personal protection, the time-seriesanalysis might not reect thendings of thecross-sectional analyses. However, the time-seriesanalysis did support the relationship betweenrearm ownership rates and LEO homicide ratesthroughout the study period; we did not believethat LEO homicide rates were the cause of theincrease in privately owned rearms. No study ofthe reasons for privaterearm ownership citedpolice deaths as an important cause for privateindividuals obtainingrearms.There was no gold standard for annualrearm ownership rates. Although the BRFSSsurvey is viewed as the best available measure,and as a reasonably accurate measure ofrearm ownership,30,31questions concerninghouseholdrearm ownership were only in-cluded for 3 years on the survey. Althoughusing FS/S estimates from the Centers forDisease Control and Prevention allowed for anannual estimate, the data were heteroscedastic:states with lower population had more year-to-year variation in the FS/S estimate comparedwith larger states. Despite the increased vari-ance for FS/S in smaller states, we foundgood correlation between the Centers forDisease Control and Prevention and BRFSSestimates ofrearm ownership (Pearson cor-relation =0.86). Although neither the BRFSSnor FS/S was a perfect measure ofrearmprevalence, the consistent results using eithermeasure and their close correlation addedvalidity to their use in our study.The FBI uses a very strict denition of LEOsfor inclusion in its UCR reports.19Other datasources, such as the Bureau of Labor StatisticsCensus of Fatal Occupational Injuries and theNational Law Enforcement Memorial Fundinclude a broader denition of LEOs.11How-ever, the FBI data bestt the purpose of ourstudy, which was to assess the risk of homicideto sworn ofcers in the line of duty. Thegeneral validity of the UCR system wasassessed by Lynch and Jarvis46and Pizzaro andZeoli.47Although these reviews found over-reporting in larger cities and some missing datafor general homicides, there was no directassessment of the validity of homicides ofLEOs. As of the late 1980s, UCR reportingrelied on what is called the National Incident-Based Reporting System.19Because occupa-tional homicides of LEOs are important events,it was not likely that a local law enforcementagency would fail to report the murder of oneof its own ofcers to the FBI.ConclusionsThe occupational homicide rate of lawenforcement ofcers in the United Stateswas positively associated with the staterearmownership rate, with a 10% increase inrearmownership correlated to 10 additional ofcerhomicides over the 15-year study period. Theseresults controlled for the main factor likely toaffect the LEOhomicide rate (i.e., the state violentcrime rate) and other factors expected to affecthomicide rates in the general population. Higherlevels of privaterearm ownership likely in-creased the frequency with which ofcers facedpotentially life-threatening situations on the job.LEOs working in states with higher levels ofgun ownership faced a greater likelihood ofbeing shot and killed on the job compared withtheir peers in states with lower gun ownership.The differences were large. Ofcers in thehigh-gun states had 3 times the likelihood ofbeing killed compared with low-gun states.Higher levels of civilian gun ownershipappeared to be a signicant risk factor for thehomicide of LEOs. jAbout the AuthorsAt the time of this study, David I. Swedler was withEnvironmental Health Sciences, Harvard School of PublicHealth, Harvard University, Boston, MA, and Center forInjury Epidemiology, Liberty Mutual Research Institute forSafety, Hopkinton, MA. Molly M. Simmons is with the JohnsHopkins Center for Injury Research and Policy, JohnsHopkins Bloomberg School of Public Health,Baltimore,MD. Francesca Dominici is with the Department of Bio-statistics, Harvard School of Public Health. David Hemenwayis with the Department of Health Policy and Management,Harvard School of Public Health.Correspondence should be sent to David I. Swedler,Division of Environmental and Occupational Health Sciences,2121 W Taylor St, 556 SPHW (MC 922), Chicago IL,60612 (e-mail: Swedler@uic.edu). Reprints can be orderedat http://www.ajph.org by clicking theReprints link.This article was accepted April 29, 2015.ContributorsD. I. Swedler and D. Hemenway conceptualized thestudy. D. Swedler, F. Dominici, and D. Hemenwaydesigned the analysis. D. I. Swedler and M. M. Simmonscollected and analyzed the data and wrote the article. Allauthors contributed to editing and approved thenaldraft of the article.AcknowledgmentsPortions of these data will be presented at the AmericanPublic Health Associations Annual Meeting; October31---November 4, 2015; Chicago, IL.Human Participant ProtectionAll data were publicly available and free to download;thus, the Harvard School of Public Health institutionalreview board deemed this study to be non---humanparticipant research.References1. Krug EG, Powell KE, Dahlberg LL. Firearm-relateddeaths in the United States and 35 other high- and upper-middle-income countries. Int J Epidemiol. 1998;27(2):214---221.2. Mercy JA, Krug EG, Dahlberg LL, Zwi AB. Violenceand health: the United States in a global perspective. Am JPublic Health. 2003;93(2):256---261.3. Richardson EG, Hemenway D. Homicide, suicide,and unintentional rearm fatality: comparing the UnitedStates with other high-income countries, 2003. J Trauma.2011;70(1):238---243.4. Hemenway D, Miller M. Firearm availability andhomicide rates across 26 high-income countries. JTrauma. 2000;49(6):985---988.5. Felson RB, Berg MT, Rogers ML. Bringing a gun to agunght: armed adversaries and violence across nations.Soc Sci Res. 2014;47:79---90.6. Siegel M, Ross CS, King C III. The relationshipbetween gun ownership andrearm homicide rates inthe United States, 1981---2010. Am J Public Health.2013;103(11):2098---2105.7. Hepburn LM, Hemenway D. Firearm availabilityand homicide: a review of the literature. Aggress ViolentBehav. 2004;9(4):417---440.8. Hendricks SA, Jenkins EL, Anderson KR. Trends inworkplace homicides in the US, 1993---2002: a decade ofdecline. Am J Ind Med. 2007;50(4):316---325.9. Bureau of Labor Statistics. Census of Fatal Occupa-tional Injuries, 2012. Washington, DC: Department ofLabor; 2014.10. Tiesman HM, Hendricks SA, Bell JL, Amandus HA.Eleven years of occupational mortality in law enforce-ment: the Census of Fatal Occupational Injuries, 1992-2002. Am J Ind Med. 2010;53(9):940---949.11. Tiesman HM, Swedler DI, Konda S, Pollack KM.Fatal occupational injuries among US law enforcementofcers: a comparison of national surveillance systems.Am J Ind Med. 2013;56(6):693---700.12. Clarke RV. Situational Crime Prevention: SuccessfulCase Studies. Guilderland, NY: Harrow and Heston;1997.13. Kaminski RJ. Murder of Police Ofcers. New York,NY: LFB Scholarly Publishing LLC; 2004:205.14. Pinizzotto AJ, Davis EF, Miller CE III. The deadlymix: ofcers, offenders and the circumstances thatbring them together. FBI Law Enforcement Bull. 2007;76(1):1---10.15. Willits DW, Nowacki JS. Police organisation anddeadly force: an examination of variation across large andsmall cities. Policing Soc. 2014;24(1):63---80.16. National Institute for Justice. Ballistic Resistance ofBody Armor. Washington, DC: Ofce of Justice Programs;2008.17. Jena AB, Sun EC, Prasad V. Does the declininglethality of gunshot injuries mask a rising epidemic of gunviolence in the United States? J Gen Intern Med. 2014;29(7):1065---1069.RESEARCH AND PRACTICEe6| ResearchandPractice | PeerReviewed | Swedleretal. AmericanJournal ofPublicHealth | PublishedonlineaheadofprintAugust13,201518. LaTourrette T. The life-saving effectiveness of bodyarmor for police ofcers. J Occup Environ Hyg. 2010;7(10):557---562.19. US Department of Justice. Uniform Crime ReportingHandbook. Clarkburg, WV: Federal Bureau of Investiga-tion; 2004.20. Swedler DI, Kercher C, Simmons MM, Pollack KM.Occupational homicide of law enforcement ofcers in theUS, 1996---2010. Inj Prev. 2014;20(1):35---40.21. Lester D. The police as victims: the role of guns inthe murder of police. Psychol Rep. 1987;60(2):366.22. Kyriacou DN, Monkkonen EH, Peek-Asa C, et al.Police deaths in New York and London during thetwentieth century. Inj Prev. 2006;12(4):219---224.23. Lee AD, Frandsen RJ, Lauver GA, Naglich D.Background Checks for FirearmTransfers, 2010-StatisticalTables. Washington, DC: Department of Justice; 2013.24. Webster DW, Vernick JS, Hepburn LM. Effects ofMarylands law banning Saturday night special handgunson homicides. Am J Epidemiol. 2002;155(5):406---412.25. Webster D, Crifasi CK, Vernick JS. Effects of therepeal of Missouris handgun purchaser licensing law onhomicides. J Urban Health. 2014;91(2):293---302.26. Loftin C, McDowall D, Wiersema B, Cottey TJ.Effects of restrictive licensing of handguns on homicideand suicide in the District of Columbia. N Engl J Med.1991;325(23):1615---1620.27. National Archive of Criminal Justice Data. NAJCD:Uniform Crime Reporting Program Resource Guide. AnnArbor, MI: National Archive of Criminal Justice Data;2014.28. Molina DK, Dimaio VJ. Trends inrearm usage inhomicides and suicides in Bexar County Texas from1982 to 2004. Am J Forensic Med Pathol. 2008;29(4):281---284.29. Beaman V, Annest JL, Mercy JA, Kresnow M,Pollock DA. Lethality ofrearm-related injuries in theUnited States population. Ann Emerg Med. 2000;35(3):258---266.30. Azrael D, Cook PJ, Miller M. State and localprevalence ofrearms ownership measurement, struc-ture, and trends. J Quant Criminol. 2004;20(1):43---62.31. Miller M, Azrael D, Hemenway D. Rates of house-holdrearm ownership and homicide across US regionsand states, 1988---1997. Am J Public Health. 2002;92(12):1988---1993.32. US Department of Justice. Crime in the United States,2010. Clarksburg, WV: Federal Bureau of Investigation;2011.33. Centers for Disease Control and Prevention. Web-Based Injury Statistics Query and Reporting System. Atlanta,GA: Centers for Disease Control and Prevention; 2014.34. Miller M, Lippmann SJ, Azrael D, Hemenway D.Householdrearm ownership and rates of suicide acrossthe 50 United States. J Trauma. 2007;62(4):1029---1034.35. Miller M, Barber C, White RA, Azrael D. Firearmsand suicide in the United States: is risk independent ofunderlying suicidal behavior? Am J Epidemiol. 2013;178(6):946---955.36. Hilbe JM. Negative Binomial Regression. 2nd ed.Cambridge, UK: Cambridge University Press; 2013.37. Vittinghoff E, Glidden DV, Shiboski SC, McCullochCE. Regression Methods in Biostatistics: Linear, Logistic,Survival, and Repeated Measures Models. New York, NY:Springer Science+Business Media, Inc.; 2005.38. Siegel M, Negussie Y, Vanture S, Pleskunas J, RossCS, King C III. The relationship between gun ownershipand stranger and nonstrangerrearm homicide rates inthe United States, 1981---2010. Am J Public Health.2014;104(10):1912---1919.39. Siegel M, Ross CS, King C. Examining the relation-ship between the prevalence of guns and homicide ratesin the USA using a new and improved state-level gunownership proxy. Inj Prev. 2014;20(6):424---426.40. Kercher C, Swedler DI, Pollack KM, Webster DW.Homicides of law enforcement ofcers responding todomestic disturbance calls. Inj Prev. 2013;19(5):331---335.41. Smith SG, Fowler KA, Niolon PH. Intimate partnerhomicide and corollary victims in 16 states: NationalViolent Death Reporting System, 2003---2009. Am JPublic Health. 2014;104(3):461---466.42. Webster DW, Wintemute GJ. Effects of policiesdesigned to keeprearms from high-risk individuals.Annu Rev Public Health. 2015;36:21---37.43. Webster D, Crifasi CK, Vernick JS. Erratum to:effects of the repeal of Missouris handgun purchaserlicensing law on homicides. J Urban Health. 2014;91(3):598---601.44. Covington MW, Huff-Corzine L, Corzine J. Batteredpolice: risk factors for violence against law enforcementofcers. Violence Vict. 2014;29(1):34---52.45. Kaminski RJ, Jefferis E, Gu J. Community correlatesof serious assaults on police. Police Q. 2003;6(2):119---149.46. Lynch JP, Jarvis JP. Missing data and imputations inthe Uniform Crime Reports and the effects on nationalestimates. J Contemp Crim Justice. 2008;24(1):69---85.47. Pizarro JM, Zeoli AM. An assessment of the qualityof homicide data in the Supplementary Homicide Re-ports: a research note. Justice Q. 2013;30(4):711---731.RESEARCH AND PRACTICEPublishedonlineaheadofprintAugust13, 2015| AmericanJournal ofPublicHealth Swedleretal. | PeerReviewed | ResearchandPractice | e7