Firearms Law and Fatal Police Shootings: A Panel Data Analysis
aa r X i v : . [ ec on . GN ] J a n Firearms Law and Fatal Police Shootings:A Panel Data Analysis
Marco Rogna ∗ and Bich Diep Nguyen Faculty of Political Theory - Civic Education, Hanoi National University of Education ∗ Correspondence: Marco Rogna, Department of Economics, Bochum University of Applied Sci-ences, Am Hochschulcampus 1, 44801 Bochum, Germany. Email: [email protected] bstract
Among industrialized countries, U.S. holds two somehow inglorious records:the highest rate of fatal police shootings and the highest rate of deaths relatedto firearms. The latter has been associated with strong diffusion of firearmsownership largely due to loose legislation in several member states. The presentpaper investigates the relation between firearms legislation \ diffusion and thenumber of fatal police shooting episodes using a seven-year panel dataset. Whileour results confirm the negative impact of stricter firearms regulations foundin previous cross-sectional studies, we find that the diffusion of guns ownershiphas no statistically significant effect. Furthermore, regulations pertaining to thesphere of gun owner accountability seem to be the most effective in reducingfatal police shootings. Keywords:
Police shootings; Firearms law; Guns diffusion; Panel data.
J.E.L.
I12; I18. ntroduction
The Unites States (U.S.) is one of the countries with the highest firearms–relateddeath rate, exceptionally high compared to other industrialized countries (Pallin et al.,2019). Furthermore, it is the country with the highest firearms possession rate bycitizens, thanks to loose legislation in several of its member states (Hemenway, 2017).The relation between firearms–related death rate and firearms possession rate hasbeen largely investigated and a causal relation going from the latter to the for-mer seems very plausible, e.g. Krug et al. (1998); Bangalore and Messerli (2013)and Siegel et al. (2013). If this fact has prompted part of the public opinion to askfor stricter guns laws, the constitutional right to posses and bear firearms has beenstrenuously defended by the opposite faction, rendering this one of the most popularand controversial topic in the country.Almost every aspect of the relation between firearms legislation and gun vio-lence has been extensively researched. Besides the mentioned studies that have triedto establish a causal link between firearms–related death rate and firearms posses-sion rate (Krug et al., 1998; Bangalore and Messerli, 2013; Siegel et al., 2013), othershave investigated the effect of firearms legislation \ possession rate on suicide rates(Kellermann et al., 1992; Anestis and Houtsma, 2018), on pediatric firearms–relatedmortality Goyal et al. (2019), on homicides rate (Duggan, 2001; Kovandzic et al.,2013; Siegel et al., 2013, 2014), and on the death rate of police officers on duty (Lester,1987; Mustard, 2001; Swedler et al., 2015).With approximately 1000 deaths per year, United States (U.S.) holds anotheringlorious record among industrialized countries: the highest rate of homicides com-mitted by police forces (Hemenway et al., 2019). This raises the question of whetherthe high rate of police fatal shootings results from relaxed firearms legislation \ highpossession rate. From a speculative point of view, one can argue that the diffusion of1rearms increases the probability of police officers to face armed people while on duty,thus increasing the probability of being involved in potentially dangerous situationsthat require the use of guns from their part. Furthermore, the increased probabil-ity to face dangerous situations is a factor of stress that may lead law enforcementofficers to overreact or, more generally, to commit mistakes. A positive relation be-tween firearms diffusion and deadly assaults to police officers (Swedler et al., 2015)corroborates the theory of an increase danger to officers for higher levels of firearmsownership.Thanks to the availability of independent datasets that remedy the underreportingof police fatal use of force in official statistics (Conner et al., 2019), this topic has re-cently been investigated. Hemenway et al. (2019) find a positive association betweenfirearms prevalence and fatal police shooting rates. Kivisto et al. (2017) report thatU.S. states with stricter firearms legislation have lower incidence rates of fatal po-lice shootings. Both studies are, however, cross-sectional and thus, despite the use ofseveral controls common in the dedicated literature, may suffer from the omitted vari-able problem. Kivisto et al. (2017), for example, mention, among the limitations oftheir paper, exactly this problem, stating that “it is possible that states with strictergun legislation also have better training for police officers and more stringent hiringpractices, or that states that are already safe are more likely to implement strictergun laws”.The novelty of the present study is to use a panel dataset to investigate the relationbetween firearms legislation \ possession rate and fatal police shootings. This allowsto control for unobserved fixed characteristics at state level that may have biasedprevious analysis, providing, therefore, more robust results.2 Literature Review
As discussed before, firearms legislation and ownership is a strongly investigated topic,particularly in the United States. A number of studies document negative impactsof firearms legislation and prevalence, such as increased suicide and homicide rate,although evidence is not unambiguous. From an extensive literature review, Kleck(2015) concludes that guns diffusion is a positive determinant of crime rate, but thisrelation looses statistical significance in the most methodologically rigorous papers.Branas et al. (2009) find that possessing a gun increases the probability of being shotduring an assault, thus dismantling the opinion of weapons having a protective role.On the other side, Kleck and Patterson (1993), by comparing 170 U.S. cities, findscarce evidence that guns restrictions have some positive role in reducing the rate ofviolence. Similarly, Altheimer (2008) evidences that gun availability has no effect indetermining the number of total individual assaults and robbery, but only increasesthe number of the ones committed with a fire–weapon.Regarding one of the most serious violent crimes, homicide, the results are mixed.Duggan (2001), Siegel et al. (2013) and Siegel et al. (2014) find a significant andpositive relation between guns diffusion and the number of homicides. However,this view is strongly opposed by Kovandzic et al. (2013), which, on the contrary,document a negative relation. They cite a potential deterrent effect of guns as anexplanation for their results. A similar debate has surrounded the permission ofcarrying concealed weapons, with Lott and Mustard (1997) showing a positive role ofsuch law in reducing violent crimes whereas Dezhbakhsh and Rubin (1998) rejectingthis finding and claiming the opposite. Findings regarding firearms diffusion andlegislation are therefore very discordant even when considering aspects such as crimeand homicide rate that are among the most studied.Shifting the attention on police forces, there is a paucity of research regarding the3ssociation between firearms diffusion and legislation and the number of killings of,and by, police officers. When considering the former, the killing of law enforcementofficers, we have again contrasting findings. Lester (1987) and Swedler et al. (2015)find that increasing levels of households gun ownership are a clear factor of risk forpolice officers, whereas Mustard (2001), limiting the attention to the possibility ofcarrying concealed weapons, puts in evidence a potential protective role of this law.Kivisto et al. (2017) and Hemenway et al. (2019) investigate the relation betweenfirearms and killings committed by police officers. The two studies both rely on in-dependent and open source databases to retrieve the number of fatal police shootings:Kivisto et al. (2017) on The Counted, maintained by The Guardian, and Hemenway et al.(2019) on Fatal Force, created by The Washington Post. Furthermore, they both relyon a cross sectional analysis using the fraction of suicides with a fire–weapon on thetotal number of suicides as a proxy for guns diffusion, as previously done in otherpapers, e.g. Kleck (2004) and Azrael et al. (2004). Compared to Hemenway et al.(2019), whose focus is exclusively on guns diffusion as a cause of fatal police shootings,Kivisto et al. (2017) further consider firearms legislation, using the Brady Campaignscorecards as an indicator of law strength. In both papers, firearms ownership isfound to positively affect the number of fatal police shootings. Even after control-ling for firearms prevalence, firearms regulations on gun trafficking and on child andconsumer safety significantly reduces fatal police shootings (Kivisto et al., 2017).Given the contrasting evidence emerged in other topics related to firearms diffusionand legislation and since both the last mentioned papers rely on a cross sectionalanalysis that may be plagued by the omitted variable bias, the extension to a paneldata setting seems a necessary further step. This may help to strengthen the findingsreached so far or to contest their validity as the result of a biased analysis.4
Data and Methods
Following Kivisto et al. (2017) and Hemenway et al. (2019), our units of observa-tion are the 50 U.S. states, with District of Columbia having being excluded forlack of data in several covariates. The covered time period is from January 1,2012, to December 31, 2018, and all variables are expressed as yearly values, form-ing a dataset with seven time periods. Different databases have been consultedand merged in order to have all the variables of interest and the necessary con-trols: Fatal Encounters, Giffords scorecards, the U.S. Census Bureau data portal, theFederal Bureau of Investigation’s (FBI’s) Crime Data Explorer and the Centers forDisease Control and Prevention’s (CDC’s) Web-based Injury Statistics Query and Reporting System(WISQARS). Following is a description of all variables.
Our dependent variable, the number of deaths caused by police shooting per millioninhabitants (
Pol Shoot ), is retrieved from the Fatal Encounters database. We choosean independent database to alleviate the likely problem of underreporting of suchepisodes in the FBI’s official statistics (Williams et al., 2019). Compared to otheropen source repositories, e.g. The Counted, Fatal Force, Mapping Police Violenceand Gun Violence Archive, the Fatal Encounters database covers the longest timespan – from 2000 to present. Specific cases of police shooting can be retrieved byselecting the category “Deadly use of force” and the subcategory “Gunshot”.Our independent variable of interest is the strength of firearms regulations at statelevel (
Giff Score ). In order to obtain a synthetic measure of the strength of a statelegislation, we rely on the Giffords scorecards, available for the period 2010-2018, withthe exclusion of the year 2011, hence the need to drop the year 2010. The overallscore is an aggregation of seven component scores, namely: background checks and5ccess to firearms (
BCAF ), other regulations of sales and transfers (
ORST ), classes ofweapons and magazines \ ammunitions ( CWAM ), consumers and child safety (
CCS ),gun owner accountability (
GOA ), firearms in public places (
FPP ) and a residualclass (
OTH ). Disaggregation of the overall score allows us to test the role of eachcomponent in explaining fatal police shootings, following Kivisto et al. (2017). Thisis helpful in identifying the areas where intervention should be prioritized to reducepolice shooting episodes. Since the scoring system has been slightly modified severaltimes during the study period, we have implemented a harmonization procedure,retaining only the sub-indicators that remained unaltered over time. Kivisto et al.(2017), using data from the same source, eliminate the weighting system in favor ofa “1 law = 1 point” scale. They argue that a weighting system necessarily entails adegree of arbitrariness. However, we think that the equal weighting implied by the“1 law = 1 point” scale is analogously arbitrary. We, therefore, prefer to rely onthe weights assigned by professional lawyers, thus leaving the Giffords scorecard scaleunaltered.Stricter legislation on firearms may reduce the quantity of fire–weapons owned bycitizens, but may also promote safer use, e.g. by denying dangerous subjects accessto guns or by increasing the safety of circulating weapons. Besides examining if lawsto promote safe gun use are effective in reducing fatal police shootings, we can testif the effect of firearms legislations operate via the former channel by looking at therelation between fatal police shootings and gun diffusion. Lacking state–level data ongun ownership for our study period, we rely on a commonly adopted proxy (
Suicide )– the percentage of suicides committed with a fire–weapon over the total of suicides(Kleck, 2004; Azrael et al., 2004; Kivisto et al., 2017; Hemenway et al., 2019). Thesedata are retrieved from the Web-based Injury Statistics Query and Reporting System(WISQARS). 6able 1: Descriptive Statistics
Variable N. Obs. Mean Std. Dev. Min. Max.
Pol Shoot
350 3.51 2.11 0 10.85
Giff Score
350 31.98 24.42 4 105.50
Crime
350 3627.32 1388.63 1026.24 8849.56
Suicide
350 51.54 12.35 13.20 74.30
PC Income
350 29712.58 4828.93 20119 52500.00
Urban
350 73.59 14.44 38.70 95
Poverty
350 9.99 2.80 4 19.20
White
350 76.95 12.67 24.30 95.10
Low Edu
350 11.19 2.97 6.10 18.60
Unemp.
350 3.96 1.18 1.80 7.90
Young
350 23.15 1.31 19.80 27.50
Giff Score disentangled
BCAF
350 7.19 5.51 0 22
ORST
350 4.12 5.89 0 24
CWAM
350 1.82 3.86 0 14
CCS
350 2.25 2.02 0 9
GOA
350 2.76 5.00 0 17.50
FPP
350 9.16 4.36 0 19
OTH
350 4.67 2.18 0 10Regarding control variables, we retrieved data on the number of violent crimes(per million inhabitants,
Crime ) from the FBI’s Crime Data Explorer, where a crimeis defined as any of the four offenses – murder and non-negligent manslaughter, rape,robbery, and aggravated assault. All the other controls are retrieved from the U.S.Census Bureau. Note that all values are projections on the 2010 census data. Thesecontrols include per–capita income in 2010 inflation–adjusted dollars (
PC Income )and the percentage of people living in urban areas (
Urban ). It must be noted that,7or this last variable, only figures for the year 2010 were available, thus it is treatedas a time–invariant covariate. Other socio–economic characteristics, such as povertyrate (
Poverty ), unemployment rate (
Unemp. ), and the percentage of adults withan education lower than high school diploma (
Low Edu ), are also included. Thepercentage of young population, aged 18–34, (
Young ) and the percentage of whiteCaucasians (
White ) over the whole population are controlled for in the analysis. Thelast variable is added since several studies find a racial bias in police shootings (Ross,2015; Nix et al., 2017; Mesic et al., 2018). Table 1 reports some key statistics of allthe mentioned variables.
The statistical analysis is divided into two main parts. In the first part, we focus on therole of the legislative strength as a whole, thus considering the overall score providedby Giffords for each U.S. state. In the second part, we analyze the component Giffordsscores separately. This analysis should provide more specific policy indications withregard to the legislative field where intervention may be more productive in reducingfatal police shootings.The effects of any changer in legislation may take time to be observed. In allour analysis, therefore, both the overall and the component Giffords scores enterin their first lags. The inclusion of lags more distant in time (two or three years) isprecluded by the limited number of time periods at our disposal. We have actually runregressions with the contemporaneous level of the Giffords scores, but none of themhas resulted in significant coefficients (results are available from the authors uponrequest). The possibility to include both the lag and the current level is preventedby their high correlation ( ρ = 0 .
99) that most likely causes a problem of collinearity.Figure 1 shows the Spearman’s rank correlation coefficients of all variables, except8he component Giffords score. Note
Giff Score L denotes the lagged Giffords score.From Figure 1 it is possible to observe that several covariates have a relatively lowcorrelation coefficient. The exceptions are the violent crime rate ( ρ = 0 .
49) and theper–capita income ( ρ = − . Poverty , White , Low Edu , Unemp. and
Young ) is equal to zero (p–value = 0.6098 for model (1) and p–value= 0.8595 for model (2)), we omit these variables in any subsequent analysis. Despiteits lack of significance,
Urban is kept in all models because of its high correlationwith the lagged Giffords score ( ρ = 0 .
62) and because it is a significant control inprevious studies (Hemenway et al., 2019). Models (3) and (4) in Table 2 report theresults with the parsimonious set of explanatory variables. From Table A1 in theAppendix, it is possible to observe non significant p–values for the RESET and forthe Mundlak test. Compared to models (1) and (2), the coefficients of models (3)and (4) are negligibly different.A conditional fixed effect (FE) and a random effect (RE) Poisson regressions(models (1) and (2) in Table 2) are our main models of interest. The functional formspecification is checked through a RESET test, by adding the squared residuals ofthe Poisson regressions (FE and RE) and checking their significance (Ramsey, 1974).Results are reported in Table A1 in the Appendix together with the coefficients of theyear dummies (year 2013 as base) included in all models. The p–value of the squaredresiduals is far above the 10% significance level, thus dismissing possible concernsabout misspecification. The choice to report both the FE and RE estimates is due tothe fact that the Mundlak test – also reported in Table A1 – has a significance levelvery close to the 5% level (Mundlak, 1978). Although this test, chosen comparedto the more common Hausman test given the presence of year dummies and a timeinvariant covariate (
Urban ), suggests the use of the random effect model, the close-ness of the p–value to the threshold level and the concerns about the distributionalassumptions of the Poisson RE model lead us to report the fixed effect estimates aswell. However, it can be observed that the estimated coefficients are very similar inboth specifications. 10able 2: Regressions Results: Total Giffords Score (Lagged) (1)Poisson FE (2)Poisson RE (3)Poisson FE (4)Poisson RE
Pol Shoot Pol Shoot Pol Shoot Pol ShootGiff Score L
Crime
Suicide
PC Income
Urban
Poverty
White
Low Edu
Unemp.
Young t-statistics in parentheses. P-value: + p < < < < As a robustness check, we also report the results of linear models, both FE and RE,after that the dependent variable has been transformed with a Yeo–Johnson powertransform in order to render its distribution more normal–like (Yeo and Johnson,11000). The results of the linear models are shown in Table 3. Although the transfor-mation of the dependent variable prevents the computation of meaningful marginaleffects, the sign and significance of the coefficients serve to confirm the results of, orto signal a possible problem in, the Poisson regressions. The distribution of the de-pendent variable before and after the Yeo–Johnson transformation is shown in FigureA1 in the Appendix.For our second purpose, we regress the rate of fatal police shootings on the com-ponent Giffords scores rather than the overall score. Here we present only the parsi-monious models. Four models have been run, two Poisson – FE and RE – and twoanalogous linear regressions. Results are reported in Table 4, models (5), (6), (7) and(8), and auxiliary information can be found in Table A1 in the Appendix. From thislast table, it is possible to see that the p–values of the RESET and of the Mundlaktest are all above conventional significance levels. As for the previous models, thedifference in the significance of the coefficients of the Poisson and of the linear regres-sions are very modest, so as the difference in the coefficients between the FE and REmodels. 12able 3: Regressions Results: Linear Models (1b)Linear FE (2b)Linear RE (3b)Linear FE (4b)Linear RE
Pol Shoot Pol Shoot Pol Shoot Pol ShootGiff Score L -0.00879* -0.00924** -0.00925* -0.00803*(-2.32) (-2.86) (-2.34) (-2.52)
Crime
Suicide -0.00515 0.00193 -0.00562 0.00237(-0.55) (0.36) (-0.59) (0.45)
PC Income -1.83-E005* -1.95-E005* -1.44-E005+ -155-E005*(-2.07) (-2.22) (-1.75) (-2.12)
Urban
Poverty -0.0248 -0.0259(-0.80) (-0.87)
White -0.0159 -0.00411(-0.41) (-1.03)
Low Edu
Unemp.
Young -0.0476 -0.0138(-0.92) (-0.47)
Const. t-statistics in parentheses. P-value: + p < < < < (5)Poisson FE (6)Poisson RE (7)Linear FE (8)Linear RE Pol Shoot Pol Shoot Pol Shoot Pol ShootBCAF L
ORST L
CWAM L
CCS L
GOA L
FPP L
Crime
Suicide
PC Income
Urban
Const. t-statistics in parentheses. P-value: + p < < < < Results and Discussion
In the presentation of results, we will focus solely on the Poisson regressions, giventhe difficult interpretation of the coefficients of the linear models discussed earlier.Note that all the reported coefficients relative to the Poisson models are incidence rateratios (IRR). From Table 2, it is possible to observe that the Giffords score (lagged)is always statistically significant, although only at 10% level in the RE models, whileat 5% in the FE models. An increase of one point in the overall Giffords score causesan approximately 1% reduction in the number of fatal police shootings per million ofinhabitants if considering the FE model. The percentage reduction is slightly lowerfor the RE model: 0.8% when including all controls – model (3) – and 0.7% whenexcluding the subset of jointly non-significant covariates – model (4).An important point to notice is the lack of significance, in all models of Table 2,of the proxy for firearms diffusion:
Suicide . The strong correlation with the lag ofthe Giffords score ( ρ = − .
83) may suggest that the effect of this last masks the oneof the former. However, if running the same regressions without the inclusion of theGiffords score, the p–value of
Suicide remains above 0.1 in all models: 0.152, 0.106,0.140 and 0.118 for, respectively, models (1), (2), (3) and (4) without
Giff Score L (full results available from the authors upon request).When considering Table 4 and the disentangled categories composing the Giffordsscore, only the coefficient related to the lag of one category, gun owner accountability,is significant (at 5% level). This happens to be true both in the FE and in the REPoisson regressions, so as in the linear models. In particular, an increase of one pointin the strength of the gun owner accountability category is associated with a decreasein per–million inhabitants fatal police shootings of 3.7% (FE model) or of 3.8% (REmodel).We can further notice that in all models, both in Table 2 and 4, the sign of the15oefficients of the main control variables is as expected. In particular, the number ofviolent crimes positively impacts the number of fatal police shooting episodes whereasper–capita income has the opposite effect. A last word is dedicated to the significanceof the violent crime rate that is very high (0.1%) in all RE Poisson models but absentin the FE models. This is possibly due to the persistent nature in time of thisphenomenon that, in the FE model, gets captured by the fixed effect component.
The present study shows that increasing levels of firearms regulation are significantlyassociated with a lower number of fatal police shooting cases. In particular, a pointincrease in the overall Giffords score leads to a decrease of in fatal police shootingsof 0.7%–1%, depending from the model. When considering separately the variouscategories composing the Giffords score, one point increase in the strength of gunowner accountability leads to a decrease of approximately 3.7%–3.8% in the numberof people killed by police officers. The diffusion of fire–weapons, instead, has nostatistically significant role in determining the considered outcome. This finding hasbeen achieved through the use of a panel dataset, thus controlling for unobservedheterogeneity through the use of FE Poisson models.It is interesting to compare our results with previous findings. Regarding theeffect of firearms diffusion, our results clearly contradicts the previous findings ofKivisto et al. (2017) and of Hemenway et al. (2019). In fact, we do not find a sta-tistically significant effect of guns diffusion in determining the number of fatal policeshootings.Considering the strength of firearms regulations, our analysis basically confirmsthe findings of Kivisto et al. (2017). However, this is true only for the overall score.When evaluating each category separately, significant differences emerge. First of16ll, it must be noted that the results are not easily comparable, given the differentscoring system used in Kivisto et al. (2017), namely the Brady scorecards, and in thepresent analysis. However, a comparison is not impossible. In Kivisto et al. (2017),two categories remained significant after all controls were added, namely promotingsafe storage via child and consumer safety laws and curbing gun trafficking. Theformer corresponds to the category consumers and child safety (
CSS ) in the Giffordsscorecards and the latter is included in other regulation of sales and transfers (
ORST ),both not significant in our models. The gun owner accountability, the category foundsignificant in the present analysis, is instead composed by three elements: licensingof gun owners and purchasers, having the highest weights, followed by registration offirearms and reporting lost or stolen firearms. This is an important difference, withpotentially relevant implications for policy–makers. Furthermore, it is reasonable tothink that the first sub-category, namely the need of gun owners to have a license,has a great discriminant power in determining the final identity of gun owners. Thisfurther suggests that the qualitative side of gun diffusion (who owns a gun) is moreimportant in limiting the number of police shootings than the quantitative side (howmany guns are owned).
Conclusions
The present analysis has shown that police shooting episodes are significantly reducedby stricter levels of firearms law. While this finding partially confirms what emergedin previous studies, we also find that the diffusion of fire–weapons is inconsequen-tial in determining the number of police shootings, thus contradicting the precedentevidence.The policy recommendations that can be derived from the present paper are quitestraightforward. Improving the strength of firearms regulations seems an effective way17or reducing the number of people killed by law enforcement officers. Actually, thepolicy prescriptions can be even more specific. In fact, from the analysis it emergesthat the most effective intervention for reducing fatal police shooting episodes wouldbe to strengthen the rules of gun owner accountability, namely licensing of gun ownersand purchasers, registration of firearms and reporting lost or stolen firearms. Thesepolicy prescriptions are different from the ones provided by previous studies.Another important lesson, and a departure from previous findings, is the lack ofstatistical significance of the diffusion of guns in causing fatal police shootings. Thissuggests that the cause may be more qualitative (who owns the guns) rather thanquantitative (how many guns). The fact that the only significant legislative categoryemerged from this study is the gun owner accountability further strengthens thishypothesis.There are several possible ways in which the analysis could be expanded in orderto have more precise and specific prescriptions. One possibility would be to furtherdisaggregate each category of the Giffords score into its subcategories. We have notpursued this road due to the limited number of observations at our disposal. Anotherinteresting extension would be to consider the episodes of police shootings directedtowards unarmed citizens (Hemenway et al., 2019). The lack of this information inthe Fatal Encounters database has prevented us to conduct this analysis.18 eferences
Altheimer, I. (2008). Do Guns Matter-A Multi-level Cross-National Examination of Gun Availability on Assault and Robbery Victimization.
W. Criminology Rev. , 9:9.Anestis, M. D. and Houtsma, C. (2018). The association between gun ownership and statewide overall suicide rates.
Suicide and Life-Threatening Behavior , 48(2):204–217.Azrael, D., Cook, P. J., and Miller, M. (2004).State and local prevalence of firearms ownership measurement, structure, and trends.
Journal of Quantitative Criminology , 20(1):43–62.Bangalore, S. and Messerli, F. H. (2013). Gun ownership and firearm-related deaths.
The American Journal of Medicine , 126(10):873–876.Branas, C. C., Richmond, T. S., Culhane, D. P., Ten Have, T. R., and Wiebe, D. J.(2009). Investigating the link between gun possession and gun assault.
AmericanJournal of Public Health , 99(11):2034–2040.Conner, A., Azrael, D., Lyons, V. H., Barber, C., and Miller, M. (2019).Validating the National Violent Death Reporting System as a source of data on fatal shootings of civilians by law enforcement officers.
American Journal of Public Health , 109(4):578–584.Dezhbakhsh, H. and Rubin, P. H. (1998). Lives saved or lives lost? The effects of concealed-handgun laws on crime.
The American Economic Review , 88(2):468–474.Duggan, M. (2001). More guns, more crime.
Journal of Political Economy ,109(5):1086–1114.Goyal, M. K., Badolato, G. M., Patel, S. J., Iqbal, S. F., Parikh, K., and Mc-Carter, R. (2019). State gun laws and pediatric firearm-related mortality.
Pedi-atrics , 144(2):e20183283. 19emenway, D. (2017).
Private guns, public health . University of Michigan Press.Hemenway, D., Azrael, D., Conner, A., and Miller, M. (2019).Variation in rates of fatal police shootings across US states: the role of firearm availability.
Journal of Urban Health , 96(1):63–73.Kellermann, A. L., Rivara, F. P., Somes, G., Reay, D. T., Francisco, J.,Banton, J. G., Prodzinski, J., Fligner, C., and Hackman, B. B. (1992).Suicide in the home in relation to gun ownership.
New England Journal ofMedicine , 327(7):467–472.Kivisto, A. J., Ray, B., and Phalen, P. L. (2017).Firearm legislation and fatal police shootings in the United States.
American Journal of Public Health , 107(7):1068–1075.https://doi.org/10.2105/AJPH.2017.303770.Kleck, G. (2004). Measures of gun ownership levels for macro-level crime and violence research.
Journal of Research in Crime and Delinquency , 41(1):3–36.Kleck, G. (2015). The impact of gun ownership rates on crime rates: A methodological review of the evidence.
Journal of Criminal Justice , 43(1):40–48.Kleck, G. and Patterson, E. B. (1993). The impact of gun control and gun ownership levels on violence rates.
Journal of Quantitative Criminology , 9(3):249–287.Kovandzic, T., Schaffer, M. E., and Kleck, G. (2013).Estimating the causal effect of gun prevalence on homicide rates: A local average treatment effect approach.
Journal of Quantitative Criminology , 29(4):477–541.Krug, E. G., Powell, K. E., and Dahlberg, L. L. (1998).Firearm-related deaths in the United States and 35 other high-and upper-middle-income countries.
International Journal of Epidemiology , 27(2):214–221.20ester, D. (1987). The police as victims: The role of guns in the murder of police.
Psychological Reports , 60(2):366.Lott, Jr, J. R. and Mustard, D. B. (1997).Crime, deterrence, and right-to-carry concealed handguns.
The Journal ofLegal Studies , 26(1):1–68.Mesic, A., Franklin, L., Cansever, A., Potter, F.,Sharma, A., Knopov, A., and Siegel, M. (2018).The relationship between structural racism and black-white disparities in fatal police shootings at the state level.
Journal of the National Medical Association , 110(2):106–116.Mundlak, Y. (1978). On the pooling of time series and cross section data.
Economet-rica: journal of the Econometric Society , 46(1):69–85.Mustard, D. B. (2001). The impact of gun laws on police deaths.
The Journal of Lawand Economics , 44(S2):635–657.Nix, J., Campbell, B. A., Byers, E. H., and Alpert, G. P. (2017).A bird’s eye view of civilians killed by police in 2015: Further evidence of implicit bias.
Criminology & Public Policy , 16(1):309–340.Pallin, R., Spitzer, S. A., Ranney, M. L., Betz, M. E., and Wintemute, G. J.(2019). Preventing firearm-related death and injury.
Annals of Internal Medicine ,170(11):ITC81–ITC96.Ramsey, J. B. (1974). Classical model selection through specification error tests.
Frontiers in Econometrics , 1:13–47.Ross, C. T. (2015). A multi-level Bayesian analysis of racial bias in police shootings at the county-level in the United States, 2011–2014.
PloS One , 10(11):e0141854. 21iegel, M., Negussie, Y., Vanture, S., Plesku-nas, J., Ross, C. S., and King III, C. (2014).The relationship between gun ownership and stranger and nonstranger firearm homicide rates in the United States, 1981–2010.
American Journal of Public Health , 104(10):1912–1919.Siegel, M., Ross, C. S., and King III, C. (2013).The relationship between gun ownership and firearm homicide rates in the United States, 1981–2010.
American Journal of Public Health , 103(11):2098–2105.Swedler, D. I., Simmons, M. M., Dominici, F., and Hemenway, D. (2015).Firearm prevalence and homicides of law enforcement officers in the United States.
American Journal of Public Health , 105(10):2042–2048.Williams, H. E., Bowman, S. W., and Jung, J. T. (2019).The limitations of government databases for analyzing fatal officer-involved shootings in the United States.
Criminal Justice Policy Review , 30(2):201–222.Yeo, I.-K. and Johnson, R. A. (2000). A new family of power transformations to improve normality or symmetry.
Biometrika , 87(4):954–959. 22 ppendix
Figure A1: Distribution of the Dependent Variable (
Pol shoot ) before and after theYeo-Johnson Power Transformation ( λ = 0 . (a) Before(b) After Ancillary Information for Table 2 (1) (2) (3) (4) ln( α ) RESET Test Mundlak Test N. Obs.
300 300 300 300
Ancillary Information for Table 4 (5) (6) (7) (8) ln( α ) RESET Test Mundlak Test N. Obs.
300 300 300 3002013 as base year.1) The RESET Test reports the p–value of the squared residuals.2) The Mundlak Test reports the p-value of the joint significance test of the timeaverage of all regressors.