Our analyses provide evidence of an association between the per capita rate of licensed firearm dealers in a county and its rate of firearm homicide. In particular, we found that having a disproportionately high number FFLs was associated with significantly higher rates of firearm homicide in major cities. As such, FFLs may represent a risk factor for gun homicide that is modifiable. To the best of our knowledge, the association between licensed gun dealers and homicide rates has not previously been estimated. We also found more FFLs to be associated with significantly lower firearm homicide rates in other cities and in suburbs. Possible explanations for both findings are discussed below.
Our evaluation strategy followed the assumption that the number of FFLs in a county gives some indication of the prevalence of firearms (i.e., a proxy of gun availability). Although no available data report the number of guns sold per dealer, the basic finding that major cities having the most FFLs per capita also have the highest rates of gun homicide is consistent with what is known about how FFLs and communities relate in terms of gun availability. Data from gun traces (a determination of the chain of ownership, usually conducted in connection with a criminal investigation) conducted by the ATF may provide the best insight. Between July 1996 and December 1998, the ATF conducted 1,530 trace investigations to determine whether guns used during crimes were trafficked from an FFL into the illegal gun market and to determine the point of first purchase. FFLs accounted for less than 10 percent of the 1,530 trace investigations but for nearly half (40,000) of all firearms involved in these traces. The average number of firearms trafficked by the FFLs under investigation was 350, which far exceeds the average number of firearms trafficked by other means including gun shows (130 guns), unlicensed gun dealers (75 guns), and straw purchasers (37 guns). The large volume of firearms that can be obtained by FFLs is possibly what underlies this discrepancy and why FFLs may figure prominently as a risk factor. In addition, a study where researchers telephoned FFLs and posed as customers provides additional evidence of how FFLs can facilitate the flow of guns to criminals. Gun dealers were generally willing to sell a handgun even when the buyer indicated an intention to purchase the gun illegally on behalf of someone else.
Other ATF data provide additional support for the possibility that gun homicide is a function of local FFLs. Guns are often found to have been used for criminal purposes not far from the gun dealer where they were first obtained. Approximately 62 percent of crime guns traced by the ATF were first purchased from FFLs in the state where they were recovered by police, one-quarter (25.9%) of crime guns were recovered in the county where they were purchased, and 10.5 percent were recovered in a county adjacent to the county of purchase, almost all of which were in the same state (9.5%). Moreover, almost one-third (32.2%) of traced crime guns are recovered by police within 10 miles of the FFL where they were first purchased, and over one-third (34.3%) are recovered between 11 and 250 miles of the FFL where they were first purchased.[6, 8] Thus, an FFL appears most likely to have an effect in the home or surrounding counties.
A recent study showing that gun dealers in or near major cities are at substantially elevated risk of selling guns used in crime may help to explain the strong positive association found here between FFL prevalence and gun homicide in major cities specifically. Also, we found that the association between FFLs and gun homicides in major cities grew stronger from 1993 to 1999. This finding is consistent with what resulted when in the 1990s the federal government took steps to regulate FFLs more closely. Before this time, the process to obtain a license to sell firearms was appreciably simpler. The Gun Control Act of 1968 required the ATF to issue a license to any applicant who was at least 21 years old, had premises from which they intended to conduct business, and who otherwise was not prohibited by law from purchasing a firearm. At the time, the fee to obtain or annually renew an FFL was $10. The ATF received an average of 33,000 applications for FFLs each year over the decade that followed. Fully 169,052 FFLs were active by 1978. That number increased steadily thereafter and by 1992 reached a national peak of 284,117 FFLs. Not all FFLs were legitimate businesses, however. Any FFL enabled the holder to purchase large numbers of firearms, often at wholesale prices, and to buy from sellers in other states. Many of these dealers made few if any registered sales, suggesting they were not truly engaged in the business of firearms dealing as required by federal law, and a substantial proportion of the extant FFLs were not in compliance with applicable federal, state, and local laws.
With the system becoming increasingly difficult for the ATF to monitor, Congress acted and in 1993 and 1994 increased the FFL application fee 20-fold to $200 and imposed new laws intended to shut down inactive or corrupt FFLs. In the years that followed, the number of FFLs nationally dropped from about 260,000 in 1993 to 80,000 in 1999. Our effect estimates suggest that the association between FFL prevalence and homicide may have been weaker in the early 1990s than in later years because there existed a large number of low volume dealers who contributed less to the supply of firearms. Hence, with FFLs over time becoming a better measure of the exposure under study, the actual magnitude of the association between FFLs and gun homicide may be more accurately portrayed in the last few years of our study.
In contrast to what was observed for major cities, we found a negative association between gun homicide and FFL prevalence in other cities and suburbs. When considered in conjunction with the finding that FFL prevalence and FS/S are correlated strongly in major cities but correlated weakly in other cities and suburbs, this suggests that FFL prevalence is not a good proxy for gun availability in other cities and suburbs and hence the models based on those areas should not be interpreted as providing valid estimates of the relation between gun availability and gun homicide. We can consider these findings in light of our understanding of how the relation between FFLs and gun homicide may vary across counties of different urbanization types. In major city areas with higher crime rates, there will be greater criminal demand for guns and, hence, a larger illegal market for guns. It thus seems more likely that a weapon sold in a major city, as compared to one sold in another county type, will end up in the hands of a criminal user through theft, straw purchase, gun trafficking, or some other kind of transaction in the secondhand market. Also, it is possible that handguns rather than long guns account for a higher share of guns sold in major cities. Further, it would also stand to reason that an "average" gun possessor has a greater chance of using a gun criminally in an area with higher rates of gun violence. Also, gun culture and the roles of guns in peoples' lives vary dramatically across urban-rural continuum. Firearm ownership is more widespread in rural areas than urban areas, so the need to purchase a firearm from an FFL may be less necessary in rural than urban areas. The role of firearms certainly varies by county type in terms of firearm mortality, in that rates of firearm-related mortality in the U.S. are equally high in both the most urban and the most rural counties, with the nuance being that it is the gun homicide rate that is high in the most urban counties and it is the gun suicide rate that is high in the most rural counties. Each of these points highlights the importance of stratifying analyses by county type and identifying variables that measure firearm availability accurately in the county type at the focus of a particular study, an important point that has been made previously. The FS/S comparisons suggest that the FFL variable used here provides an adequate proxy in major city counties alone.
If the FFL variable is not a good proxy for gun availability in counties we have defined as other cities and suburbs, our analyses cannot inform the issue of how gun availability relates to gun homicide in counties of these types. It may be the case that the impact of guns on a community may vary by community type and may be protective in other cities and suburbs. In one study of 170 U.S. cities with a population of at least 100,000, however, rates of homicide and of gun-related assault were found to be positively associated with the prevalence of firearms. Even so, it is possible that the mixing in that study of what we have termed major cities, other cities and suburbs may have prevented the authors from detecting modification of this effect across area type. Another study of counties in Illinois initially found that the rate of firearm ownership was negatively associated with all measures of violent crime, including homicide, but had failed to control for urbanization. Subsequent multivariate analyses with control for urbanization found no significant association between the rate of firearm ownership and homicide. A number of ecologic studies conducted at the state level [10, 14] and individual-level studies [35–37] alike have found firearm availability to be a risk factor for homicide rather than a protective factor, yet other studies have not found clear effects of the relation between gun availability and homicide, e.g.,[34, 38] and a recent National Academy of Sciences report concluded that the body of research on this topic is inconclusive. Our findings highlight the need to account for urbanization in the studies that will follow.
Our analysis had several strengths. First, it was conducted at the county level to account for within-state variability in homicide rates and FFL prevalence. This also allowed us to control for the possibilities that the homicide rate in a county was influenced by FFLs in surrounding counties and by the leniency or permissiveness of neighboring state firearm laws. Second, our analysis examined the link between FFLs and homicide over county urbanization type. Third, the study years coincided with a period when changes to federal firearm licensing regulations produced a change in the composition of the pool of FFLs nationally. As FFLs became fewer the pool became more homogenous, and hence may have provided an exposure variable that became a better measure of gun availability over time. The finding that the association between FFLs and gun homicide in major cities grew stronger over time adds support to our interpretation of the results, as does the finding based on our comparison with FS/S that FFL prevalence appears to be a good proxy for gun availability in counties defined as major cities but not in other county types.
Also, two aspects of the present study are unique and have implications for how firearm homicide is studied and how the incidence of firearm homicide may be reduced. First, because the link between gun homicide and gun availability (as measured by the prevalence of FFLs) was found to vary significantly within states according to the urbanization levels of counties, it appears that studies conducted using broader geographic units of analysis (states, census divisions, etc.) may fail to detect important nuances in the nature of gun availability.
Second, to the best of our knowledge, our study, in focusing on gun dealers as a potential risk factor for homicide, is the first to assess a tangible measure of gun availability that can be modified as part of prevention activities. Law enforcement, city planners, and legal strategists in cities with high gun homicide rates can concretely focus in on excessive or problem gun dealers as opposed to the more nebulous issue of "gun availability." Moreover, local efforts to close down illegal gun commerce have already shown the potential to be effective.[33, 40, 41] As one example, zoning laws, which control the location and operation of stores and individual dealers licensed to sell firearms, have been used to regulate locations of FFLs in several U.S. communities.[11, 12, 42] An attempt to launch a coordinated effort to identify and act on problematic gun dealers will surely face challenges, however. For one, a key component of such efforts will be the policing activities of the ATF which, in having inspected fewer than 10% of FFLs in each year since 1979 and fewer than 5% of FFLs in most of those years, may have resources insufficient for the task.
Our analysis also had limitations. As discussed, we could not account for the actual volume of firearms introduced by each FFL into the community. Although gun sales data are not currently available for a more focused test, the recent National Academy of Sciences report called for better information on FFLs to be collected and made available for research. Additionally, we did not explicitly accommodate spatial autocorrelation in the estimation of the FFL effect estimates. Refitting the models presented in Table 2 and Table 3 including a simultaneous autoregressive (SAR) structure, assuming that correlation declined in proportion to the square of the distance between counties, generally changed coefficients by less than 5%. Also, data for FFLs for years after 1999 were not available. Thus we could not analyze a more recent period. Nevertheless, we do not anticipate that the relation observed here between FFLs and homicide would have changed since the study period and thus this characteristic of the data should not be interpreted as devaluing the findings. Finally, it may be that high rates of homicide may lead to increased demand for firearms and hence additional FFLs, in which case the results of the "FFL as risk factor" hypothesis that our models have been designed to test would be spurious. A stronger analytic approach would be to test whether within-county increases in FFL prevalence were followed by increases in the rate of gun homicide. We considered that approach, but found many instances in which a county experienced no gun homicides in certain years but some homicides in the subsequent year, which prevents an annual change in homicide rate from being calculated. Also, as discussed above, we found evidence that FFL prevalence became a better proxy for firearm availability over time, which led to our preference for the 1999 models and our judgment to refrain from including change models in the current manuscript. An instrumental variable approach could be pursued as well, to attempt to remove from the models the influence of circularity that may exist. We hope this manuscript will inform how such design alternatives may be approached, and acknowledge their need given the cross-sectional nature of the present study.