In the United States, more people kill themselves with firearms than with all other methods combined. 1 In 1998, there were approximately 30,000 suicide deaths, of which 57% were caused by firearms. The number of suicides (30,558) exceeded the number of homicides (17,894), and the number of firearm suicides (17,420) exceeded the number of firearm homicides (12,078). Thus, one cannot really discuss suicide in the United States without talking about guns.
A central question regarding the relation between firearms and suicide is whether the ready availability of firearms increases the suicide rate, rather than merely increasing the proportion of suicides from guns. Case-control studies provide one line of evidence suggesting that the presence of a gun in the home is a risk factor for suicide in the home 2–7 and that the purchase of a handgun from a licensed dealer is associated with becoming a suicide victim. 8 To date, however, case-control studies have been limited geographically, have focused predominantly on youth suicide, and have included relatively small numbers of suicides. For example, of the existing seven case-control studies, five focus exclusively on youth suicide in Western Pennsylvania, which in aggregate represent only about 200 cases (some included in more than one study). 2–6 The two studies that have examined suicide among all age groups are larger but are nonetheless geographically limited, comprising a single HMO in one 8 and two urban areas in the other. 7 In addition, the HMO study used purchase of a handgun from a licensed dealer, rather than the prevalence of firearm ownership, as the primary independent variable. The other study excluded from analysis suicides that occurred outside of the home (which could lead to overestimating the association between suicide and firearms). The single cohort study to examine the relation between firearms and suicide also found an increased risk of suicide among handgun purchasers but was confined to suicides in California between 1991–1996. 9
Nationally representative cross-sectional studies provide a complementary approach to evaluating whether the ready availability of firearms influences the rate of suicide. These studies also address some of the limitations of previous case-control studies, such as recall bias, the exclusion of suicides outside the home, and the generalizability of results. Measures of household gun ownership based on national surveys (or derived from proxies based on mortality statistics) are unlikely to suffer from recall bias. The use of complete census data captures suicides inside and outside the home and, on the national scale, supports the generalizability of findings. Unfortunately, nationally representative cross-sectional studies have been hampered by the lack of direct measures of household firearm ownership at areas smaller than the nine census regions, and by uncertainty regarding the relative merits of the various proxies used to estimate firearm availability. 10,11 Although some cross-sectional studies have found a link between various proxies of firearm prevalence and rates of suicide, 12–14 some have not, 15,16 and others have produced mixed findings depending on the age of the victim, 17 the analytic model, or the firearm proxy used. 18,19
The present study extends previous work by using recent national data, looking across both regions and states, and disaggregating victims by age. We use the two best measures of levels of household firearm ownership currently available—direct survey-based measures for regional analyses 20 and a rigorously validated proxy of household gun ownership for both regional analyses and for analyses across all 50 states. 21 In addition, we adjust for confounders seldom accounted for in nationally representative studies, including rates of poverty, urbanization, unemployment, alcohol consumption, education, and divorce.
We used publicly available data for the nine regions and 50 states in the United States over a 10-year period (1988–1997) to examine the association between levels of household firearm ownership and rates of suicide, firearm suicide, and non-firearm suicide for seven age groupings (5–14, 15–24, 25–34, 35–44, 45–54, 55–64, 65+) and for both males and females. State- and year-specific population figures and data for the number of suicides, firearm suicides, and non-firearm suicides come from the National Center for Health Statistics Mortality Files (NCHS). 1 Deaths from firearms of undetermined intention constitute less than 3% of all firearm deaths and are excluded from analyses. For ages 4 and under suicide is not classified as a cause of death; accordingly this age group is omitted from analyses. Region-specific population and mortality data were derived by aggregating the corresponding state-based data.
At the regional level, we used published survey-based estimates from the General Social Surveys (GSS) to measure the percentage of households with guns and the percentage with handguns. 20 The GSS, conducted by the National Opinion Research Center most years from 1972 to 1993 and biennially since 1994, is the gold standard for national surveys of gun ownership. In its current form, the GSS is conducted in person with a national area-probability sample of 3000 noninstitutionalized adults. The sample is chosen to be representative of each of the nine Census regions and of the nation as a whole, but not of individual states. At the state level, published data on reported household gun ownership are available from the Behavioral Risk Factor Survey Surveillance (BRFSS) for only a nonrandom sample of 21 states and then usually for a single year in the 1990s. 22 To analyze all 50 states, we used a proxy for household firearm ownership: firearm suicides as a fraction of all suicides, referred to in the text as FS/S.
FS/S, which measures the proportion of suicides by firearms (rather than the rate of suicides by firearm), has been validated against many survey-based measures of household gun ownership. A recent study determined that FS/S was the best proxy for household firearm ownership levels of the half dozen or more proxies used in the literature. 21 FS/S is highly correlated with the percentage of households reporting firearm ownership in studies across 16 developed nations (r = 0.91), 23 across the nine U.S. Census regions (r = 0.93), 24 across 21 U.S. states (r = 0.90), 22 across 170 U.S. cities (r = 0.86), 25 and across 14 areas within a single state (r = 0.87). 21
By way of comparison, we juxtaposed regression results using FS/S calculated for each region with those using the direct survey-based estimates of household firearm ownership levels (GSS). We report overall results using the (nonrandom) 21 states for which survey-based household firearm ownership levels are available (assuming that the survey estimates, generally available for a single year in the 1990s, provide unbiased estimates of firearm availability for the 10-year study period). Results for these 21 states using the survey measure are compared with those using the corresponding proxy measure, FS/S. Compared with the 29 states for which survey measures of firearm ownership are unavailable, the 21 states for which survey measures are available have similar mean rates of poverty (14%vs 13%), divorce (5%vs 5%), per capita annual alcohol consumption (2.3 vs 2.4 gallons), high school graduates (79%vs 80%), suicide rates per 100,000 (13 vs 13), firearm suicide rates (8 vs 8), and non-firearm suicide rates (5 vs 5). The 21 states have a greater percentage of the population living in metropolitan areas (71%vs 63%) and are, on average, slightly more populous (mean population 6 million compared with 4.5 million).
The assumption that the relative availability of guns in a region will be reflected by the fraction of all suicides that are firearm suicides is distinct from assumptions about the net effect that gun availability will have on the overall suicide rate. The proxy FS/S is based on the assumption that firearms are likely to be more readily available in states where guns make up a larger fraction of all suicides than in states where guns make up a smaller fraction of all suicides—independent of the rate of suicides in a state. A state's FS/S reflects the distribution of firearm compared with non-firearm means of suicide. FS/S does not reflect the rate of suicides in a state and does appreciably bias our testing of the (null) hypothesis that there is no relation between household firearm levels and overall suicide rates.
We tested the assumption that the proportion of suicides committed with a firearm (FS/S) is not, by construction, correlated with the overall suicide rate (suicides/population). We also assessed the extent (if any) to which FS/S might bias our findings under conditions that reflect the suicide rate, population, and distribution of suicides in the United States. We did so by conducting Monte Carlo-type simulations in which we generated samples of 10,000 observations 1000 times, and then observed the mean correlations between the overall suicide rate and the fraction of suicides attributable to firearms (FS/S). Positive random numbers were generated for the number of non-firearm suicides (NFS), firearm suicides (FS), and total population. Numbers were generated so that FS plus NFS equaled the total number of suicides (S), and so that S was less than the population. The correlation between overall suicide rates (S/population) and FS/S was zero. This zero correlation between FS/S and the suicide rate held both under simulation conditions in which the population and suicide rate were constrained to reflect those within the United States over the 10-year study period, and also under conditions in which the suicide rate and population parameters were allowed to take on values 100 times greater than and 100 times less than those in the United States. For the firearm suicide rate, under conditions that reflect those in the United States with respect to population and suicide rate (but allowing FS/S to vary between 0 and 1), correlations between FS/S and the firearm suicide rate were less than 0.03; an inverse correlation of approximately the same magnitude was seen for FS/S and the non-firearm suicide rate. After imposing the additional distributional constraint that firearm suicides are twice as likely as non-firearm suicides, the correlation between FS/S and the suicide rate remained less than 0.008; for FS/S and the firearm suicide rate, the correlation remained less than 0.05.
Results are presented using the average household firearm ownership level for each region or state over the 10-year study period. Because the stock of firearms in the United States is so high (over 200 million guns) and because firearms have such a long useable life, changes in the stock of firearms for a large population are likely to be quite small from year to year. Not surprisingly, similar results were obtained whether we used (1) the average household firearm ownership level for each region or state over the 10-year study period, (2) yearly GSS or FS/S values for each geographic area (thereby allowing within-area variation over time in firearm ownership levels to influence analyses), or (3) 5-year rolling average of GSS or FS/S.
To make our measures of firearm ownership easier to compare with one another, we standardized all measures to have a mean of zero and a standard deviation of one. The dependent variable in our analyses is the number of suicides per population per region (state) per year for each age group. Distributions of death rates were skewed and variances were greater than the means. Consequently, negative binomial models were used rather than Poisson.
Primary analyses use incidence rate ratios (IRR), obtained by exponentiating beta coefficients in the negative binomial regressions, to express the association between firearm ownership levels and the rates of suicide, gun suicide, and non-gun suicide. Because the standard deviation of each firearm ownership measure is equal to one, the reported IRRs represent the percentage change in suicide rates for each standard deviation change in firearm ownership levels. Across the nine census regions, FS/S has a range of 3.1 standard deviations, and GSS handgun and overall firearm ownership estimates have a range of 3.4 standard deviations; across the 50 U.S. states, FS/S has a range of 4.1 standard deviations. For the 21 states that provide survey estimates of household firearm ownership, the range of the standardized survey measure is 3.2 standard deviations; FS/S has a range of 3.5 standard deviations across these 21 states. Weighting by population does not substantially alter our results. Data presented are not weighted.
The suicide rate for a region or a state in a given year is not independent from the rate in that area in other years. To account for this dependence, standard errors in regressions were corrected by clustering observations by region or state.
State-based analyses control for six state-level characteristics that have been associated with suicide in the literature: the percent of the population living in poverty, the percent living in metropolitan areas, the divorce rate, the unemployment rate, the percent of the population finishing high school, and the per capita alcohol consumption. 26–32 Data for alcohol consumption pertain to gallons of beer, wine, or spirits divided by the population over age 14, and come from the National Institute of Alcohol and Abuse. Data for other control variables come from the Statistical Abstract of the United States. 33 Data were not available for the divorce rate in California and Louisiana other than for 1990; unemployment data were not available for any geographic area for 1988 and 1993; and education data were not available for 1988 and 1992. Complete data were available otherwise. When data were not available, values for missing observations were interpolated from surrounding years, using predicted values generated by regressing available state-level data against time. Analyses with and without observations for which imputations were made did not differ substantively. Regional analyses do not include control variables because of the small number of observations.
To illustrate the size of the differences between states with high and low household firearm ownership levels, we show the actual numbers of suicide, gun suicide, and non-gun suicide deaths for the six states with the highest FS/S compared with the four states with the lowest FS/S. These 10 states were chosen on the basis of their extreme firearm ownership levels, not on the basis of their suicide rates. The number of states in each group was selected so that the population in the two groups over the 10-year period would be approximately equal. The six high-gun states were Louisiana, Alabama, Mississippi, Wyoming, West Virginia, and Arkansas (160 million person-years). The four low-gun states were Hawaii, Massachusetts, Rhode Island, and New Jersey (158 million person-years).
At the regional level, we found a positive association between levels of household gun ownership and overall suicide rates for the entire population, in virtually each age group, and for both men and women (Table 1). For the U.S. population as a whole, a one standard deviation increase in household handgun ownership rates (GSS Handgun) was associated with suicide rates that were, on average, 14% higher in the regions with higher handgun ownership (IRR = 1.14) (Table 1). Because handgun ownership rates increase 3.4-fold from the region with the lowest (New England) to the highest (East South Atlantic) gun levels, our adjusted model suggests that the overall suicide rate in the East South Atlantic is 1.6 times higher than in New England (14% excess compounded 3.4-fold). These results are accounted for by substantially elevated firearm suicide rates in regions with high levels of firearm ownership. There was a small, but never offsetting, decrease in the rate of non-firearm suicides in regions with higher firearm ownership, principally evident in older age groups. Results were almost identical whether household firearm ownership levels were estimated from our survey-based measure (GSS) or from the FS/S proxy.
We also found a positive relation at the state level between rates of household gun ownership and the suicide rate. This was true for the entire population, for each age group, and for both men and women (Tables 2 and 3). Crude results were nearly identical to adjusted results (not shown). As in regional analyses, state-level results are accounted for principally by the substantially elevated firearm suicide rates in states with high levels of firearm ownership. Among most age groups we found a modest, but only partially offsetting, decrease in the rate of non-firearm suicide in states with higher firearm ownership levels.
Survey (GSS)- and proxy- (FS/S)- based estimates of household firearm ownership rates were highly correlated among the 21 states for which survey measures are available, with a correlation coefficient of 0.90 (Table 4). For these 21 states, results obtained with the available survey estimates and with our proxy (FS/S) were similar. When survey measures of household firearm ownership were used, the adjusted IRRs for overall suicide, firearm suicide, and non-firearm suicide were, respectively, 1.13 (95% CI = 1.02–1.24), 1.32 (95% CI = 1.16–1.51), and 0.97 (95% CI = 0.85–1.10). When our proxy, FS/S, was used, the corresponding IRRs were 1.19 (95% CI = 1.05–1.36), 1.52 (95% CI = 1.32–1.78), and 0.90 (95% CI = 0.76–1.06). Similar correspondence was seen in unadjusted comparisons. When the FS/S proxy was used for all 50 states, the corresponding adjusted IRRs were 1.13 (95% CI = 1.06–1.19), 1.43 (95% CI = 1.34–1.53), and 0.87 (95% CI = 0.82–0.93) (Table 2). For the 29 states for which survey estimates of household firearm ownership are not available, the adjusted corresponding IRRs using FS/S were 1.08 (95% CI = 1.04–1.12), 1.36 (95% CI = 1.29–1.43), and 0.85 (95% CI = 0.82–0.89).
The six states with the highest levels of household firearm ownership had approximately the same number of person-years of population over the 10-year period as the four states with the lowest levels of household firearm ownership: 158 million compared with 160 million (Table 3). In the high-gun states, 20,368 individuals were suicide victims, compared with 12,632 in the low-gun states. For every age group, people living in the high-gun states were more likely to become suicide victims. These results were driven by higher rates of firearm suicide. People living in high-gun states were 1.6 times more likely to take their own lives (3.8 times more likely to kill themselves with a gun but only 0.6 times as likely to kill themselves with other means).
For the U.S. population as a whole, a one standard deviation increase in our firearm proxy (FS/S) is associated with suicide rates that were, on average, 13% higher in the states with the higher FS/S (adjusted IRR = 1.13) (Table 2). Because FS/S increases 4.1-fold from the states with the lowest to the highest gun levels, our adjusted model suggests a 1.7-fold difference (13% excess compounded 4.1-fold). The corresponding comparison of the actual number of suicide victims in the four states with the lowest and six states with the highest gun levels finds that the suicide rate was 1.6 times higher in the high-gun states (Table 3).
If suicides are often impulsive acts, 34–37 then the easier it is to obtain lethal means (such as firearms) the more suicides there might be. On the other hand, if the choice of firearm has less to do with method availability than with strength of intent, then suicide rates will be unrelated to the means available. Individuals determined to take their own lives will work to get a gun where guns are less available, or will substitute other equally lethal means. Our findings support the former. Both firearm-related and total suicide rates were associated with state and regional gun levels (despite an inverse association with non-firearm suicide rates). This suggests that, although substitution of deadly means may take place where guns are less available, substitution appears to be incomplete. Consistent with previous results, we found that the association between firearm availability and suicide was largest for the youngest and oldest age groups. 38
Our findings are robust. Regression analyses were not driven by either the most populous states or the states with the most extreme levels of gun ownership. Overall, consistent results were produced if the states analyzed were the 50 states, the 40 most populous states, the 40 least populous states, or if analyses excluded the 10 states most extreme in FS/S (ie, the five states with the highest FS/S and the five states with the lowest FS/S, not shown). When we added the small number of firearm deaths of undetermined cause as suicides, our findings were unchanged (not shown). The close correspondence between FS/S and our survey-based measure of household firearm ownership (GSS) is readily apparent in regional analyses (Table 1) and in analyses for the 21 states where survey-based measures were available.
One reason that FS/S may be a good proxy for household firearm ownership is that guns used for suicide are likely to be household guns. 4,7 Case-control studies have suggested that guns used in suicides are typically household guns, but nationally representative surveillance data are not available to confirm this observation. However, neither survey estimates of household firearm ownership nor FS/S is an ideal measure of firearm availability. For example, telephone surveys typically underrepresent poor people (eg, households without telephones) and it appears that women living in two-adult households with guns often do not have accurate information about whether there is a gun in the home. 39,40 Individuals who have obtained firearms illegally may not report them on surveys. In addition, our measures do not provide direct information about the ease with which high-risk individuals can obtain firearms at flea markets, at gun shows, or on the streets. Nevertheless, our findings suggest that our measures of household firearm ownership are correlated with accessibility to firearms used in suicides.
Our study has limitations resulting from the use of group-level rather than individual-level data. Of particular concern is the possibility that people who commit suicide may not share the characteristics of the populations from which they are drawn. For example, they may not be poor even if they live in disproportionately poor states. They may not own firearms (or live in households with firearms) even if they live in states with disproportionately high firearm levels. However, because a firearm must be the instrument of death for individuals who commit suicide with a firearm, we know that individuals who actually commit suicide with a gun had access to that gun.
Another limitation of our study is that we include only six potential confounders (alcohol, poverty, urbanization, divorce, unemployment, and education), and then only in the state-level analysis. Suicide rates may be affected by many other factors, such as parenting, family violence, the media, a community's social capital, and social fragmentation. It is not clear whether trying to account for these or other area-wide characteristics would increase or reduce the magnitude of the association between rates of household firearm ownership and suicide. Furthermore, addition of other variables would bring a potential cost of introducing several collinear measures at the state level.
Despite these limitations, our study provides robust, nationally representative evidence that household firearm prevalence is related to the rate of suicide. This was true for suicide by firearms and for overall suicide, across U.S. regions and states, for every age group, and for both men and women. Although our study is unable to determine causation, our results are consistent with the notion that instrumentality as well as intent affects the rate of suicide. People in areas of the United States with more firearms are more likely to take their own lives, especially by means of a gun.
1. National Center for Injury Prevention and Control. Centers for Disease Control and Prevention. Injury Mortality Statistics
. Available at: http://www.wonder.cdc.gov/mortsql.shtml
. Accessed 1 April 2002.
2. Brent DA, Perper JA, Goldstein CE, et al
. Risk factors for adolescent suicide
. A comparison of adolescent suicide
victims with suicidal inpatients. Arch Gen Psychiatry 1988; 45: 581–588.
3. Brent DA, Perper JA, Allman CJ, Moritz GM, Wartella ME, Zelenak JP. The presence and accessibility of firearms
in the homes of adolescent suicides. A case-control study [comments]. JAMA 1991; 266: 2989–2995.
4. Brent DA, Perper JA, Moritz G, Baugher M, Schweers J, Roth C. Firearms
and adolescent suicide
. A community case-control study. Am J Dis Child 1993; 147: 1066–1071.
5. Brent DA, Perper J, Moritz G, Baugher M, Allman C. Suicide
in adolescents with no apparent psychopathology. J Am Acad Child Adolesc Psychiatry 1993; 32: 494–500.
6. Brent DA, Perper JA, Moritz G, Baugher M, Schweers J, Roth C. Suicide
in affectively ill adolescents: a case-control study. J Affect Disord 1994; 31: 193–202.
7. Kellermann AL, Rivara FP, Somes G, et al
in the home in relation to gun ownership. N Engl J Med 1992; 327: 467–472.
8. Cummings P, Koepsell TD, Grossman DC, Savarino J, Thompson RS. The association between the purchase of a handgun and homicide or suicide
. Am J Public Health 1997; 87: 974–978.
9. Wintemute GJ, Parham CA, Beaumont JJ, Wright M, Drake C. Mortality among recent purchasers of handguns. N Engl J Med 1999; 341: 1583–1589.
10. Miller M, Hemenway D. The relationship between firearms
: a review of the literature. Aggression and Violent Behavior 1999; 4: 59–75.
11. Brent DA. Firearms
(with discussion). Ann N Y Acad Sci 2001; 932: 225–240.
12. Hellensten J. Motivation and Opportunity. An Ecological Investigation of U.S. Urban Suicides 1970–1990 (dissertation). University of California, 1995.
13. Markush RE, Bartolucci AA. Firearms
in the United States. Am J Public Health 1984; 74: 123–127.
14. Lester D. Firearm availability and the incidence of suicide
and homicide. Acta Psychiatr Belg 1988; 88: 387–393.
15. Lester D. Availability of guns
and the likelihood of suicide
. Sociol Soc Res 1987; 71: 287–288.
16. Lester D. Gun control, gun ownership, and suicide
Life Threat Behav 1988; 18: 176–180.
17. Sloan JH, Rivara FP, Reay DT, Ferris JA, Kellermann AL. Firearm regulations and rates of suicide
. A comparison of two metropolitan areas. N Engl J Med 1990; 322: 369–373.
18. Kleck G, Patterson E. The impact of gun control and gun ownership levels on violence rates. J Quant Criminol 1993; 9: 249–287.
19. Kleck G. Point Blank: Guns
and Violence in America. New York: Aldine de Gruyter, 1991.
20. Davis J, Smith T. General Social Surveys, 1972–1998 [machine-readable data file]. Storrs: The Roper Center for Public Opinion Research, University of Connecticut, 1998.
21. Azrael D, Cook P, Miller M. State and Local Prevalence of Firearms
Ownership. Measurement, Structure, and Trends. National Bureau of Economics Research, working paper 8570. Available at: http://www.nber.org/papers/w8570
. Accessed 2 May 2002.
22. Powell KE, Jacklin BC, Nelson DE, Bland S. State estimates of household exposure to firearms
, loaded firearms
, and handguns, 1991 through 1995. Am J Public Health 1998; 88: 969–972.
23. Killias M. International correlations between gun ownership and rates of homicide and suicide
. CMAJ 1993; 148: 1721–1725.
24. Miller M, Azrael D, Hemenway D. Firearm availability and unintentional firearm deaths, suicides and homicides among 5–14 year olds. J Trauma 2002; 52: 267–274.
25. Kleck G. Targeting Guns
and Their Control. New York: Aldine de Gruyter, 1997.
26. Norstrom T. The impact of alcohol, divorce, and unemployment on suicide
: a multilevel analysis. Soc Forces 1995; 74: 293–314.
27. Kposowa AJ. Marital status and suicide
in the National Longitudinal Mortality Study. J Epidemiol Community Health 2000; 54: 254–261.
28. Kposowa AJ. Unemployment and suicide
: a cohort analysis of social factors predicting suicide
in the U.S. National Longitudinal Mortality Study. Psychol Med 2001; 31: 127–138.
29. Kposowa AJ. Suicide
mortality in the United States: differentials by industrial and occupational groups. Am J Ind Med 1999; 36: 645–652.
30. Moscicki EK. Identification of suicide
risk factors using epidemiologic studies. Psychiatr Clin North Am 1997; 20: 499–517.
31. Baker S, O'Neill B, Ginsburg M, Li G. The Injury Fact Book. 2nd ed. New York: Oxford University Press, 1992.
32. Young TJ. Poverty, suicide
, and homicide among Native Americans. Psychol Rep 1990; 67: 1153–1154.
33. U.S. Bureau of the Census. Statistical Abstracts of the United States: 1998. Washington DC: U.S. Bureau of the Census, 1998.
34. Chapdelaine A, Samson E, Kimberley MD, Viau L. Firearm-related injuries in Canada: issues for prevention. CMAJ 1991; 145: 1217–1223.
35. Peterson LG, Peterson M, O'Shanick GJ, Swann A. Self-inflicted gunshot wounds: lethality of method versus intent. Am J Psychiatry 1985; 142: 228–231.
36. Baker SP. Without guns
, do people kill people? [editorial]. Am J Public Health 1985; 75: 587–588.
37. Adelson L. The gun and the sanctity of human life; or the bullet as pathogen. Pharos 1980; 43: 15–25.
38. Birckmayer J, Hemenway D. Suicide
and firearm prevalence: are youth disproportionately affected? Suicide
Life Threat Behav 2001; 31: 303–310.
39. Ludwig J, Cook P, Smith T. The gender gap in reporting household gun ownership. Am J Public Health 1998; 88: 1715–1718.
40. Azrael D, Miller M, Hemenway D. Are household firearms
stored safely? It depends on whom you ask. Pediatrics 2000; 106: E31.