To the best of our knowledge, this is the first study using data from within the past decade to simultaneously model the effect of multiple state firearm laws on homicide and suicide rates at the state level using a multi-year panel design. Using a difference-in-differences analysis, we found that laws requiring universal background checks and those prohibiting firearm possession by people with a conviction for a violent misdemeanor were associated with significant reductions in the overall homicide rate, while “shall issue” laws were associated with a significant increase in the homicide rate. There was no significant association between homicide and the other laws studied, and we did not find consistent relationships between any of the laws and overall suicide rates.
This study has several strengths. First, it is one of the first studies to clearly define each law with attention to the detailed provisions of the law, including its scope, exceptions, and exemptions. One reason for some of the conflicting results of previous studies (Online Supplemental Tables S1, S2) may be the inconsistent definition of state statutes.
Second, using a difference-in-differences approach helps to address the major threat to validity in this type of research: states with lower homicide rates to begin with may be more likely to enact stronger gun laws. By including state and year fixed effects, we are using a “within-estimator” that assesses differences within states over time.5, 6 Studies that do not include state fixed effects are also assessing differences across states at a given time (“between effects”), which may reflect different propensities of states with lower or higher homicide rates to enact laws, rather than law effects. Thus, the difference-in-differences approach is less subject to the possibility of “reverse causation” (i.e., it is the level of the homicide rates that are affecting the law enactment, not the other way around). The inclusion of state fixed effects has the added advantage of controlling for any differences between states in time-invariant factors.
Third, including a large panel of time-varying state factors as independent variables helps address the problem of omitted variable bias. Nevertheless, it is still possible that states which were experiencing large declines in homicide were more likely to enact a particular law; even the within-estimator may not be sufficient to rule out the possibility of reverse causation.
Our finding of a negative association between universal background checks (including permit requirements) and homicide rates is consistent with several other studies.3, 16,17,18,19,20 Our finding of a negative association between violent misdemeanor laws and homicide rates is consistent with one other recent study, which reported a 24% reduction in intimate partner homicide in states with these laws.21 However, caution should be exercised when interpreting this finding because only two states implemented violent misdemeanor laws during the study period. While historically the literature on the impact of concealed carry–permitting laws has been inconsistent and several studies have found an association between “shall issue” laws and reduced murder rates,7, 22,23,24,25,26,27,28,29 the three most recent studies to examine these laws found a positive association with homicide rates.3, 30, 31
Our finding that there was no association between stand your ground laws and homicide rates conflicts with the findings of two previous studies on these laws.32, 33 However, both of these studies examined only the decade of 2000–2010. When we restrict our analysis to that decade, we obtain similar results.
A second important finding of this study is that changes in household gun ownership were not found to be significantly associated with homicide or suicide rates, a result that differs from several previous studies.34, 35 The discrepancy in these results could possibly be due to our inclusion of state fixed effects. It is possible that although there is a strong cross-sectional relationship between the prevalence of firearm ownership and homicide and suicide rates, small changes in firearm ownership that are observed over time are not sufficient enough to result in measurable differences in overall population homicide or suicide rates. Even if we had survey-based measures of household gun ownership, the margin of error is probably greater than the actual change in gun ownership levels from year to year. There is too much noise in our measure of gun ownership and too little variability in true levels of household gun ownership to determine if changes in gun ownership are related to differences in homicide or suicide rates. Few of the previous studies included state fixed effects. Because of the conflict with the existing literature, further study is required before any definitive conclusion is drawn.
It is important to note that the absence of an observed association of a law and overall homicide or suicide rates does not necessarily mean that these laws are ineffective. It may also be that the laws are not broad enough to affect overall population death rates or that the laws are not being adequately enforced.
Several other limitations deserve mention. First, the firearm ownership proxy has been validated with cross-sectional data, but not with longitudinal data.36 It is not clear whether this proxy is able to accurately measure changes in household gun ownership over time.
Second, while we controlled for a range of state-level factors associated with homicide death rates, there may be unidentified omitted variables. For example, in the early 1990s, firearm homicide rates were very high in many cities, seemingly related to the crack cocaine epidemic.37, 38 Nevertheless, when we restrict the analysis to the period 2000–2016, our results remain essentially unchanged, although the precision of the estimates decreases.
Third, we accounted only for the presence or absence of firearm law provisions, not for the implementation and enforcement of these laws. Fourth, trying to incorporate the most important explanatory variables in a large regression almost invariably leads to some multicollinearity. For example, when we use all the other independent variables to explain variations in the gun ownership proxy, the adjusted R2 is 0.69.
Finally, we do not disaggregate homicide rates by the age or other characteristics of either the offender or victim, which could mask the effect of laws intended to affect a particular subpopulation. For example, age restrictions on gun possession would only be expected to affect youth suicide rates, not adult rates.
In conclusion, this study provides evidence that universal background checks and laws prohibiting gun ownership by people with a history of a violent misdemeanor are associated with lower overall homicide rates, while laws that provide no discretion to law enforcement officials in approving concealed carry permits are associated with higher homicide rates. Further research on the impact of state firearm laws is necessary to assess causality and should rely upon detailed definitions of each law.