State Intimate Partner Violence–Related Firearm Laws and Intimate Partner Homicide Rates in the United States, 1991 to 2015

Background: To prevent intimate partner homicide (IPH), some states have adopted laws restricting ﬁrearm possession by intimate partner violence (IPV) offenders. “Possession” laws prohibit the possession of ﬁrearms by these offenders. “Relinquishment” laws prohibit ﬁrearm possession and also explicitly require offenders to surrender their ﬁrearms. Few studies have assessed the effect of these policies. To study the association between state IPV-related ﬁrearm laws and IPH rates over a 25-year period (1991 to Design: IPH rates CI, to 15.5% IPH rates (CI, to relinquishment of a 6.6% reduction in IPH in implementation of the laws. Causal interpretation is limited by the ob-servational and ecological nature of the analysis. Conclusion: Our ﬁndings suggest that state laws restricting ﬁrearm possession by persons deemed to be at risk for perpetrating intimate partner abuse may save lives. Laws requiring at-risk persons to surrender ﬁrearms already in their possession were associated with lower IPH rates.

E very year, more than 1800 persons in the United States are killed by their intimate partners, and approximately 50% of these homicides are committed with firearms (1). Approximately 85% of victims of intimate partner homicide (IPH) are women, and IPH accounts for nearly 50% of all homicides involving women in the United States each year (2,3). Several studies (4 -7) have shown that in situations of intimate partner violence (IPV)-which may include physical violence, sexual violence, stalking, and/or psychological aggressionabusers' access to firearms increases the risk for IPH as much as 5-fold (5). Because of the association between firearm access and IPH, regulating the possession of firearms by IPV offenders is one approach to reducing IPH (3, 6 -8).
Federal legislation enacted in the United States in 1968 prohibited firearm possession by persons convicted of an IPV-related felony (9). This legislation was augmented in 1996 to extend the prohibition to those convicted of an IPV-related misdemeanor (10). In addition, the 1994 Violence Against Women Act barred firearm possession by persons subject to permanent IPV-related restraining orders (11). However, there is limited federal capacity or willingness to enforce these restrictions. Therefore, several states have enacted their own statutes to mirror these federal laws and explicitly authorize state officials to enforce the restrictions curtailing IPV perpetrators' possession of firearms (we refer to these as "possession" laws).
However, a substantial loophole in the federal statute limits the ability of states to enforce these laws, even if they have codified the federal statute into their own laws. Although the federal statute prohibits certain IPV offenders from possessing firearms, it does not explicitly require them to surrender guns already in their possession (12,13). In other words, in some states, a person may technically be prohibited from possessing a firearm, but it is up to the person to go to a police station to relinquish the weapon. Without statutory authorization, law enforcement cannot confiscate the firearms. This loophole has been termed the "relinquishment gap" by the Law Center to Prevent Gun Violence (14). As the Law Center stated in its recent report on this loophole, "One of the most glaring gaps in the nation's gun laws-even in states with the strongest gun laws in the country, like California-is the lack of an effective firearm relinquishment policy. Few state legislatures have taken any meaningful steps to actually enforce their criminal gun restrictions by ensuring that armed offenders give up their firearms after they are convicted of serious crimes" (14). Recently, however, some states have taken steps to enforce their gun restrictions by going beyond federal law.
One way that some states have done this is by enacting legislation that explicitly requires persons prohibited from possessing firearms due to an IPV-related misdemeanor or restraining order to surrender firearms already in their possession (we refer to these as "relinquishment" or "surrender" laws) (14). These laws typically enhance enforcement by putting offenders on official notice that surrender of their firearms is required and by specifying a time by which the transfer must take place. For example, California law requires that when a court issues an IPV-related restraining order, it orders respondents to surrender all firearms in their possession within 24 hours by transferring them to a law enforcement official or a federally licensed gun dealer (15). As of 2016, 11 states explicitly required persons convicted of an IPV-related misdemeanor to surrender their firearms after conviction, and 15 states explicitly required persons subject to an IPV-related restraining order to surrender their firearms for as long as the order is in effect (14). A second approach that some states have used is requiring law enforcement officials to remove firearms from the scene of an IPV incident. A third approach involves extending the prohibition of firearm possession to persons convicted of stalking.
We are aware of only 3 studies that have evaluated the effect of state firearm policies on IPH rates (16 -19). These studies were limited because they used data from 2003 or earlier and had little power to detect an effect of IPV-related firearm policies because so few states had enacted such policies. Most important, at the time of these studies, few states had enacted laws that required IPV offenders to surrender firearms already in their possession, although 16 states had adopted a relinquishment law by 2013. The number of state IPVrelated firearm laws has increased sharply since 2003, partly in response to the reauthorization of the Violence Against Women Act in 2005. A new provision made Violence Against Women Act grants conditional on state courts' informing convicted IPV offenders of the federal and any state restrictions on firearm ownership. This study builds on and extends previous studies in 2 key ways. First, our data include homicide rates and state laws over a much longer and more recent time frame (1991 to 2015). Second, we drew data from the most comprehensive coding of state IPV-related laws to date, paying particular attention to relinquishment provisions that might be critical to the larger policy goal of restricting firearm possession by potentially dangerous persons.

Design Overview
We conducted a panel study to examine the association between state IPV-related firearm laws and total and firearm-related IPH rates between 1991 and 2015. The panel consisted of state-level predictor and outcome variables for each of the 50 states during each of the 25 years. The outcome variable was the statespecific IPH rate in a given year. Four categories of laws were considered: 1) prohibition of firearm possession by persons convicted of an IPV-related misdemeanor, with or without a relinquishment requirement; 2) prohibition of firearm possession by persons subject to an IPV-related restraining order, with or without a relinquishment requirement; 3) laws authorizing removal of firearms from the scene of a domestic violence incident; and 4) prohibition of firearm possession by persons convicted of stalking ( Table 1). We also examined whether laws in the first 2 categories included a mechanism to ensure that persons surrender firearms already in their possession. Because we used secondary data sources without personal identifiers, the Institutional Review Board of Boston University Medical Center deemed this not to be human subjects research.

Outcome Variables
Annual Firearm, Nonfirearm, and Total IPH Rates, by State. The only national data source that records homicide victim-offender relationships is the Supplementary The SHR victim-offender relationship categories include spouses, common-law spouses, former spouses, and dating partners, but former dating partners are not specifically categorized (24). This prevented us from assessing the effect of IPV-related firearm laws on abuse of noncohabiting dating partners, which is underreported in the SHR because these cases are often not classified as IPH (24).
Missing Data on Victim-Offender Relationship. The SHR is limited by missing data on the victim-offender relationship in approximately one third of homicides. Fox and Swatt developed a multiple imputation approach for these missing data that is generally viewed as the strategy of choice (25). The imputation procedure attempts to ascertain the likely victim-offender relationship using known variables about the case and the observed association of those variables with the victim-offender relationship in cases where this relationship is clear. Fox provided us with multiply imputed files covering 1990 to 2015 (26). In our prior work, we showed that regression results obtained using the imputed data are similar to those obtained using only cases in which the victim-offender relationship is known (27). Nevertheless, we conducted analyses using both the imputed and nonimputed data to ensure that the imputation process did not alter the findings. The correlation between imputed and nonimputed IPH rates was 0.93.

Main Predictor Variable
Using searches conducted with the WestlawNext and HeinOnline legal resources, with laws then collected from state legislature Web sites, Everytown for Gun Safety developed a database of state IPV-related firearm laws over time (28). Using this database, we coded 4 categories of laws and their operative provisions (a total of 6 variables) as present or absent for each state during each of the 25 years from 1991 through 2015 (Appendix Table 1, available at Annals .org). Data on these provisions for all 50 states for 2015 are shown in Appendix Tables 2 and 3 (available at Annals.org). State laws on IPV-related felonies were not included in our analyses because most states have procedures in place for the surrender of firearms by persons convicted of felonies.

Control Variables
We accounted for secular trends in homicide rates by including year fixed effects. To account for dynamic effects and to address the potential problem of omitted-variables bias, we included the lagged IPH rate (that is, the rate in the previous year). We also controlled for state-level household firearm ownership, us-ing a proxy (a variable that serves in place of an unmeasurable variable) that we developed in earlier research (29). This proxy is necessary because no survey assessed household firearm ownership at the state level throughout the study period (29). We also considered the following factors, retaining in the model only those that were significantly related to the outcome: population distribution by age, sex, and race/ethnicity; region (East, South, West, and Midwest); degree of urbanization; rates of education, poverty, unemployment, divorce, and self-reported depression; levels of household income and income inequality; population density; per capita gross domestic product, personal disposable income, and alcohol consumption; rates of nonhomicide violent crime (aggravated assault, robbery, and forcible rape), stranger homicide (homicide committed by a person unknown to the victim), and property crime (burglary, larceny-theft, and motor vehicle theft); incarceration rate; and per capita number of law enforcement officers (Appendix Table 4, available at Annals.org). We lagged the state laws by 1 year; for example, we used laws in 1991 to predict homicide rates starting in 1992. Thus, the law data used in our analyses covered 1990 through 2014.

Statistical Analysis
Because homicide rates are skewed and overdispersed rather than normally distributed, we modeled this outcome using a negative binomial model following the approach in our previous studies (27, 30 -33).
To account for clustering of observations among states, we used a generalized estimating equations approach (34 -38). We used SEs that are robust to the presence of serial autocorrelation and heteroscedasticity (39). Model fit was assessed using the quasi-information criterion, indicated by Pan as the criterion of choice for generalized estimating equations models (40). This test indicated that an exchangeable working correlation matrix produced the best fit for the data.
To develop a parsimonious model and to avoid overfitting the model, we conducted a stepwise variable selection procedure. Both a forward and a backward selection procedure resulted in the inclusion of 3 covariates: proportion of the population that was African American, violent crime rate, and divorce rate. Five variables were automatically included in all models: year fixed effects, region fixed effects, stranger homicide rate, household firearm ownership, and lagged IPH rate.

Role of the Funding Source
This research was funded by a grant from the Evidence for Action: Investigator-Initiated Research to Build a Culture of Health program of the Robert Wood Johnson Foundation. The funder had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; and preparation, review, or approval of the manuscript.

ORIGINAL RESEARCH
State Intimate Partner Violence-Related Firearm Laws and Homicide Rates

RESULTS
The number of states with IPV-related firearm laws increased substantially between 1990 and 2015, but very few states enacted laws requiring IPV offenders to surrender firearms they already had ( Figure 1). For example, by 2015, 26 states prohibited firearm possession by persons convicted of an IPV-related misdemeanor, but only 11 of those states also explicitly required relinquishment of weapons. Similarly, 24 states prohibited firearm possession by persons subject to an IPV-related restraining order, but only 15 of those states also explicitly required relinquishment of firearms. By 2015, 2 states (California and Illinois) had enacted all of the laws, but 16 states had enacted none. During the study period, 3 states (Washington, Alabama, and Arkansas) repealed a law that was already in place.
Nationally, the total IPH rate decreased from 1.19 per 100 000 persons in 1991 to 0.60 per 100 000 persons in 2015, and the firearm-related IPH rate decreased from 0.68 to 0.36 per 100 000 persons. There was a nearly 5-fold range in average IPH rates across states in 2015, with a low of 0.36 per 100 000 persons in Minnesota and a high of 1.67 per 100 000 persons in Louisiana. The mean IPH rate across all states decreased from 1.18 per 100 000 persons in 1991 to 0.67 per 100 000 persons in 2015.
We first examined models that included each of the laws one at a time ( Table 1). Laws that prohibited firearm possession by persons subject to an IPV-related restraining order and required them to surrender firearms they already had were associated with 10.8% lower total IPH rates (95% CI, Ϫ16.8% to Ϫ4.4%) and 15.0% lower firearm-related IPH rates (CI, Ϫ23.3% to Ϫ5.9%) compared with the absence of both laws. Laws prohibiting firearm possession by persons subject to IPV-related restraining orders that did not also require offenders to surrender firearms they already had were not significantly associated with total or firearm-related IPH rates. The 3 other categories of laws were not significantly associated with total or firearm-related IPH rates.
We next developed a final model that included only the presence or absence of an IPV-related restraining order firearm relinquishment law but did not include restraining order firearm possession laws (states with these laws were in the reference group). These laws were significantly associated with 9.7% lower total IPH rates and 14.0% lower firearm-related IPH rates but were not significantly associated with non-firearmrelated IPH rates ( Table 2).
To ensure that imputation of the victim-offender relationship in some cases did not affect the results, we repeated the analysis using only nonimputed data (that is, including only confirmed IPHs). The results were essentially unchanged: restraining order relinquishment laws were associated with 8.4% lower IPH rates (CI, Ϫ16.0% to Ϫ0.04%) (Appendix Table 5, available at Annals.org).
We then estimated a model in which the relationship between restraining order relinquishment laws and IPH rates was allowed to vary by state. In this model, these laws were associated with significantly lower IPH rates in 9 states, nonsignificantly lower rates in 3 states, and nonsignificantly higher rates in 2 states (Figure 2; Appendix Table 6, available at Annals.org).  In 10 of 14 states that enacted these laws, the percentage change in IPH rates from before to after the law was greater than that for other states without such a law during the same period (Appendix Table 7, available at Annals.org). The regression model estimated lower IPH rates associated with implementation of the law for all of these states.
To check the validity of our findings, we investigated whether laws requiring persons subject to IPVrelated restraining orders to surrender their firearms were associated with other crime-related outcomes not expected to be affected by these laws. We found no significant relationship between these laws and stranger homicide rates, violent crime rates, or prop-

ORIGINAL RESEARCH
State Intimate Partner Violence-Related Firearm Laws and Homicide Rates erty crime rates. We also confirmed that the relationship between restraining order relinquishment laws and IPH rates remained significant when we included region-specific time trends and state fixed effects in the model. We also reran the analysis using the traditional proxy for household gun ownership (the percentage of suicides committed with a firearm) used in other studies (41), and the results were unchanged. Finally, the results remained unchanged when we controlled for state laws requiring universal background checks, permits to purchase handguns, or waiting periods for the purchase of handguns. We saw a clear trend of gradually decreasing IPH rates in states without restraining order relinquishment laws, but the rate of decrease slowed after 2005 and the average IPH rate actually increased from 2013 to 2015 (Figure 3). In contrast, IPH rates in states with restraining order relinquishment laws continued to decrease at their previous rate after 2005 and dropped slightly from 2013 to 2015. Differences in IPH rates in 2015 between states with and without restraining order relinquishment laws are shown in the Appendix Figure  (available at Annals.org).
If one assumes a causal relationship, our final model ( Table 2) suggested that there were 75 fewer IPHs in 2015 among states with restraining order relinquishment laws than would have been expected without these laws. The model also suggested that if all 50 states had such laws in place, there would have been an additional 120 fewer IPH deaths across the nation in 2015 than would have been expected without these laws.

DISCUSSION
We examined the association of state IPV-related firearm laws with IPH rates using data subsequent to 2003, a period in which many states enacted such laws. We found that state laws that both prohibited the possession of firearms by persons subject to an IPV-related restraining order and required these persons to surrender their firearms were associated with firearm-related IPH rates that were 14.0% lower than in states without these laws. Laws that prohibited the possession of firearms by persons subject to a restraining order but did not require them to surrender firearms already in their possession were not significantly associated with IPH rates.
A basic implication of these findings is that laws that identify firearm owners who are at high risk for using their weapons against their partners and require the relinquishment of those weapons may save lives. Women who obtain restraining orders are at particularly high risk for partner violence, given that fear of violence (often created by threatened or actual violence) typically motivates the desire for a protective order. These findings seem to demonstrate the value of identifying high-risk situations based on known episodes of past violence and removing firearms from such situations to prevent future violence.
Although our study did not find a statistically significant association between laws prohibiting IPV misdemeanants or convicted stalkers from owning guns and rates of IPH, current data do not allow us to assess the extent to which implementation might mediate their effect. Such laws may be effective only if law enforcement  has the authority or mandate to seize firearms from offenders. At this time, there are not enough states with such provisions to assess their effect.
Other state-level variables that were related to IPH in our models were residence in the South, the prevalence of household firearm ownership, the stranger homicide rate, the lagged IPH rate, the proportion of the population that was African American, the violent crime rate, and the divorce rate. Data from a national survey conducted by the Centers for Disease Control and Prevention show that physical (excluding sexual) IPV is approximately 35% higher among African American women than white women (42). Thus, it may be that the significant coefficient for this variable reflects a higher rate of IPH among African Americans.
The chief potential threat to the validity of our findings is that states that have enacted laws requiring subjects of IPV-related restraining orders to surrender their firearms may differ from those that have not in ways that were not measured. Another important limitation of this research is that even if laws are written similarly, their enforcement may vary by county, city, or town within a given state. There may also be differences in how the judicial system in each state adjudicates IPV cases and in how state law handles protective orders in general. Our findings may also reflect the effect of laws other than IPV-specific ones. Finally, to avoid the ecological fallacy, caution must be used in drawing inferences from this study with regard to the relationship between both the main exposure variable (state laws) and the covariates and IPH risk at the individual level.
Despite these limitations, the results of this study suggest that laws prohibiting firearm possession by persons subject to IPV-related restraining orders may be associated with lower rates of firearm-related IPH, but only if the law includes an explicit requirement that these persons relinquish their firearms. They are especially grateful to Everytown for Gun Safety (particularly Ted Alcorn, Courtney Zale, Elizabeth Avore, Jonas Oransky, Sarah Tofte, Cecily Wallman-Stokes, and Billy Rosen), Legal Sciences LLC, and the LawAtlas Project, which produced the database from which the 32 provisions in this study's database were coded. * The law being tested prohibits persons who are subject to IPV-related restraining orders from possessing firearms and requires them to surrender firearms they already have. The models included year fixed effects, region, household gun ownership, stranger homicide rate, lagged IPH rate, proportion of the population that is African American, violent crime rate, and divorce rate. The reference group is states with no law requiring surrender of firearms by persons subject to an IPV-related restraining order. Results for Minnesota are not reported because there was only 1 y of data after implementation of the law, so the estimate is highly unstable. † Statistically significant (P < 0.05). ‡ All variables are standardized such that the percentage shown is the percentage difference in IPH rates associated with a 1-SD increase in the listed factor. The year in which the law was implemented is shown. IPH = intimate partner homicide; IPV = intimate partner violence.