Firearms and suicide in US cities
- Correspondence to Dr Matthew Miller, Department of Health Policy and Management, Harvard School of Public Health, Room 305, Kresge Building, 677 Huntington Avenue, Boston, MA 02115, USA;
- Received 16 July 2013
- Revised 9 October 2013
- Accepted 20 October 2013
- Published Online First 3 December 2013
On an average day in the USA more than 100 Americans die by suicide—half use firearms. Suicide rates overall and by firearms are higher, on average, in states where household firearm ownership is more common. In general this means in states where a greater proportion of the population lives in rural areas. The current ecological study focuses on the relation between measures of household firearm prevalence and suicide mortality in urban areas (metropolitan statistical areas and divisions) using survey-based measures of firearm ownership. Suicide rates (1999–2010) for metropolitan statistical areas that are comprised of large US cities come from death certificate records; rates of household firearm ownership come from the 2002 and 2004 Behavioural Risk Factor Surveillance System. Higher rates of firearm ownership are strongly associated with higher rates of overall suicide and firearm suicide, but not with non-firearm suicide. Stratification by gender, age and race did not materially affect the association between firearms and suicide. This study provides evidence consistent with previous case–control work and extends evidence from previous state- and region-level ecological studies that firearms in the home impose suicide risk above and beyond baseline.
More than 100 Americans die by suicide every day in the USA—half use firearms.1 Suicide rates overall and by firearms are higher, on average, in states where household firearm ownership is more common, which, in general, means in states where a greater proportion of the population lives in rural areas. In contrast, rates of suicide by methods other than firearms do not correlate significantly with rates of household firearm ownership.2 ,3 This pattern has been reported in ecological studies at the state level that have adjusted for several potential confounders, including aggregate measures of psychological distress, alcohol and illicit drug use and misuse, poverty, education, unemployment, and the proportion of the state's population living in urban areas.2–,6 The current study examines whether household firearm ownership—as estimated by results from the Behavioural Risk Factor Surveillance System (BRFSS)—is associated with suicide rates across urban areas. As such, this is the first ecological study to use direct measures of firearm ownership to address the possibility that previous findings of a gun–suicide connection based on state-level data may have been confounded by co-varying gun ownership and underlying risk of completed suicide in rural compared with urban areas.
The US Office of Management and Budget (OMB) defines metropolitan statistical areas (MSAs) as locations with at least one urbanised area with a population of at least 50 000, plus adjacent territory that has a high degree of social and economic integration with the core as measured by commuting ties. Large MSAs are subdivided into multiple metropolitan divisions (MDs) when the MSA has a core population of at least 2.5 million. Our analyses are restricted to MSAs and MDs that include a city with at least 200 000 people and for which representative estimates of household firearm ownership are available from the BRFSS (n=44).
County-level suicide mortality data were obtained through the Centers for Disease Control and Prevention (CDC)'s Wide-ranging Online Data for Epidemiologic Research (WONDER) detailed mortality database.1 Suicides were characterised as suicide by firearm (International Classification of Diseases, 10th Revision (ICD-10) E-codes X72–X74) and by non-firearm methods (E-codes X60–X71, X75–X84, Y87.0, and U03). Mortality data over the 12-year period, 1999–2010, were aggregated for each of the 44 MSAs and MDs of interest using the OMB's 2000–2010 specification of the county-level composition of MSAs and MDs.i
County level data on the percentage of individuals living in households with firearms (firearm prevalence) were obtained from the 2002 and 2004 BRFSS; 2004 is the most recent year for which state level estimates are available.7 The BRFSS, the world's largest telephone survey (more than 200 000 adult respondents annually), is an ongoing data collection programme sponsored by the CDC, with all 50 states participating. A detailed description of the survey methods used by BRFSS is available elsewhere.8–,10 Briefly, trained interviewers collect data on a monthly basis by using an independent probability sample of households with telephones among the non-institutionalised US population aged ≥18 years. All BRFSS questionnaires and data are available on the internet (http://www.cdc.gov/brfss). Firearm ownership information was obtained by the interviewer, who began the firearm section of the survey by first informing the respondent that ‘the next three questions are about firearms. We are asking these in a health survey because of our interest in firearm-related injuries. Please include weapons such as pistols, shotguns, and rifles; but not BB guns, starter pistols, or guns that cannot fire. Include those kept in a garage, outdoor storage area, or motor vehicle.’ ‘Are any firearms kept in or around your home?’ was the key question used to assess the presence of firearms in the home. Firearm prevalence estimates exclude respondents who could not, or refused to, answer the BRFSS firearm questions (2.8% of respondents in high-gun and 2.6% in low gun areas). We used the mean of the 2002 and 2004 estimates; when, as was the case for counties in Texas, Ohio, California, Florida and North Carolina, firearm ownership was assessed in either 2002 or 2004, but not both years, firearm prevalence was assigned on the basis of a single year's estimate. Eleven of 44 counties had 1 year of data. On average, estimates for those counties with two measures differed by 4% and had a SD of 10%.
Our primary outcomes of interest were the rate of suicides, the rate of suicides involving firearms, and the rate of suicides involving methods other than firearms (aggregated over the 12-year study period). Standard linear regression analyses were undertaken for the population overall, for different age strata, and by sex. Sensitivity analyses examined the relationship between firearm prevalence and suicide mortality among white people to account for differences in racial composition. Further stratification by race (eg, black people) was not possible because BRFSS estimates for gun ownership by black people for our MSAs and MDs is too imprecise to produce reliable estimates. Results are similar when negative binomial regression, rather than standard linear regression, is used (not shown).
To illustrate our main findings more concretely, we compare suicide deaths in our study period in MSAs and MDs most disparate in their firearm prevalence. The group of high-prevalence and the group of low-prevalence urban areas were assembled so that the number of person-years in the two groupings was approximately equal: 17 MSAs/MDs with the highest firearm prevalence are compared with the five MSAs with the lowest firearm prevalence.
Regression analyses further quantify these relations (table 1). Suicide rates are, on average, 0.16 deaths (per 100 000 population) higher in MSAs where firearm ownership rates are one percentage point higher. The relation between firearm ownership and suicide rate is entirely due to the relation between firearm ownership and firearm suicide, as reflected in a β coefficient associated with firearm suicide (0.16 (95% 0.12 to 0.20)), which is very similar to that for overall suicide (0.14 (95% CI 0.07 to 0.22)), and a β coefficient relating firearm ownership and non-firearm suicide, which is essentially null: −0.02 (95% CI −0.06 to 0.02).
Stratified analyses by sex and age produce patterns similar to those in primary analyses. Sensitivity analyses restricted to white people (ie, white suicide rates in relation to BRFSS estimates of white firearm ownership rates) tell the same story, as do analyses for the population overall (not shown in table 1), with significant β coefficients for firearm suicide and suicide overall of 0.12 and 0.10, respectively, and a non-significant β coefficient of −0.03 for non-firearm suicide.
Table 2 shows that, although the aggregate number of people residing in the high-gun MSAs/MDs (firearm prevalence=34%) and low-gun MSAs/MDs (firearm prevalence=11%) is approximately equal (by design), one and a half times as many adults completed suicide in the high-gun (57 704) compared with the low-gun (39 030) areas. This difference in total suicide deaths over the 12-year period is almost entirely attributable to differences in firearm suicides (32 081 vs 13 545), with little difference in the number of non-firearm suicides (25 623 vs 25 485). A similar pattern is observed for men, women and children.
Consistent with previous empirical work from individual-level11–,21 and state- and region-level ecological studies,2–,6 we find that higher rates of firearm ownership are associated with higher rates of overall suicide and firearm suicide, but not with non-firearm suicide. Our findings hold for the population overall, for men, for women, and across all age groups, are not explained by differential racial composition across MSAs and MDs, and are consistent with previous state-level ecological work, which controlled for rates of major depression, serious suicide thoughts, serious mental illness, alcohol and drug dependence and misuse, urbanisation, poverty and unemployment.3 ,6 ,22
Our findings are also consistent with results from previous cross-sectional studies that have examined suicide rates across US urban areas23–,25 and generally found a significant relationship between gun ownership levels and both firearm suicide and overall suicide. A major problem with these studies is that they did not use survey-based measures of firearm ownership levels, and were forced to use somewhat questionable proxies (eg, one partial measure that all three studies used was the percentage of homicides that were firearm homicides).
Our study should be considered in light of several potential limitations. First, our measure of firearm availability is household firearm ownership. This is a reasonable measure of exposure because most firearm suicides involve firearms from the victim's home,19 but this measure does not provide potentially important information about various characteristics of firearm availability that may affect risk, such as how firearms are stored. Second, firearm prevalence data in primary analyses come from the 2002 and 2004 BRFSS (2004 being the latest year for which such data are available), whereas mortality data are from 1999–2010. The effect of this temporal discrepancy on our results is likely to be small because guns are highly durable and, as has been observed previously, the cross-sectional pattern of household firearm ownership tends to be remarkably constant over time.26 In addition, the main findings are similar when mortality data are restricted to 2002–2004. Third, our study uses aggregate data for our exposures and outcomes of interest and, as is always the case with analyses based on aggregate data, drawing causal inferences about individual risk factors can be problematic.27 Fourth, our study does not control for many factors other than firearm ownership levels that may affect rates of suicide (eg, rates of mental illness, divorce rates, poverty). Although residual confounding by these factors is possible, previous ecological work that controlled for state-level aggregate measures, including psychological distress, alcohol and illicit drug use/misuse, poverty, education and unemployment, found little difference between results from bivariate and multivariate analyses.3 In addition, there is no a priori reason to expect that any such residual confounding would explain why firearm suicide and overall suicide rates (but not non-firearm suicide rates) were selectively elevated in our study's high- versus low-gun areas.
Despite these limitations, our results support the hypothesis, established in previous individual-level work,11 –21 that the availability of lethal means is associated with an increased risk of death by suicide. Our study suggests as well that the population attributable risk related to the presence of household firearms is large, operates across gender, race and age categories, and is not confounded by underlying differences in baseline suicide risk across the rural–urban continuum.
What is already known on this subject
Suicide rates overall and by firearms are higher, on average, in states where household firearm ownership is more common. In general this means in states where a greater proportion of the population lives in rural areas. This pattern has been reported in ecological studies that use survey-based estimates of household firearm ownership at the state level, but not previously at the city level, raising the possibility that state-level analyses are confounded by differences in urbanisation.
What this study adds
This study suggests that the population attributable risk of suicide related to the presence of household firearms is large, operates across gender, race and age categories, and, moreover, is not confounded by underlying differences in baseline suicide risk across the rural–urban continuum.
Contributors MM, DA and DH made substantial contributions to conception and design, and interpretation of data; MW made substantial contributions to acquisition of data and analysis. MM wrote the initial draft, and all authors made substantial contributions to revising it critically for important intellectual content; all authors gave final approval of the version to be published. MM is responsible for the overall content.
Funding This work was supported by the Joyce Foundation, grant number 12-34332.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Data used in this study are publicly available and the data sources are cited in the text.
↵i Executive Office of the President, Office of Management and Budget. OMB Bulletin No. 10-02 Update of Statistical Area Defintions and Guidance on Their Uses. 2009. http://www.whitehouse.gov/sites/default/files/omb/assets/bulletins/b10-02.pdf