Gun Ownership as a Risk Factor for Homicide in the Home

Saturday, 9th of January 2016

This is the 1993 CDC article, cited 178 times in the medical literature, which led to the 1996 congressional ban on CDC research concerning guns and public health. That ban is still in force.

Gun Ownership as a Risk Factor for Homicide in the Home

Arthur L. Kellermann, Frederick P. Rivara, Norman B. Rushforth, Joyce G. Banton, Donald T. Reay, Jerry T. Francisco, Ana B. Locci, Janice Prodzinski, Bela B. Hackman, and Grant Somes

N Engl J Med 1993; 329:1084-1091October 7, 1993DOI: 10.1056/NEJM199310073291506

Also at http://www.nejm.org/doi/pdf/10.1056/NEJM199310073291506

Homicide claims the lives of approximately 24,000 Americans each year, making it the 11th leading cause of death among all age groups, the 2nd leading cause of death among all people 15 to 24 years old, and the leading cause of death among male African Americans 15 to 34 years old1. Homicide rates declined in the United States during the early 1980s but rebounded thereafter2. One category of homicide that is particularly threatening to our sense of safety is homicide in the home.

Unfortunately, the influence of individual and household characteristics on the risk of homicide in the home is poorly understood. Illicit-drug use, alcoholism, and domestic violence are widely believed to increase the risk of homicide, but the relative importance of these factors is unknown. Frequently cited options to improve home security include the installation of electronic security systems, burglar bars, and reinforced security doors. The effectiveness of these protective measures is unclear, however.

Many people also keep firearms (particularly handguns) in the home for personal protection. One recent survey determined that handgun owners are twice as likely as owners of long guns to report “protection from crime” as their single most important reason for keeping a gun in the home3. It is possible, however, that the risks of keeping a firearm in the home may outweigh the potential benefits4.

To clarify these issues, we conducted a population-based case-control study to determine the strength of the association between a variety of potential risk factors and the incidence of homicide in the home.

METHODS

Identification of Cases

Shelby County, Tennessee; King County, Washington; and Cuyahoga County, Ohio, are the most populous counties in their respective states. The population of King County is predominantly white and enjoys a relatively high standard of living. In contrast, 44 percent of the population of Shelby County and 25 percent of the population of Cuyahoga County are African American. Fifteen percent of the households in Shelby County and 11 percent in Cuyahoga County live below the poverty level, as compared with 5 percent in King County5-7.

All homicides involving residents of King County or Shelby County that occurred between August 23, 1987, and August 23, 1992, and all homicides involving residents of Cuyahoga County that occurred between January 1, 1990, and August 23, 1992, were reviewed to identify those that took place in the home of the victim. Any death ruled a homicide was included, regardless of the method used. Assault-related injuries that were not immediately fatal were included if death followed within three months. Cases of homicide involving children 12 years of age or younger were excluded at the request of the medical examiners.

Selection of Case Subjects and Recruitment of Case Proxies

A home was defined as any house, apartment, or dwelling occupied by a victim (i.e., a case subject) as that persons principal residence. Homicides occurring in adjacent structures (e.g., a garage) or the surrounding yard were also included. Murder-suicides and multiple homicides were considered a single event. In the case of a murder-suicide, the homicide victim was included if he or she was older than the suicide victim; in multiple homicides, the oldest victim was included.

Reports made at the scene were collected to ensure that study criteria were met. In King County, the medical examiners staff conducted all investigations of the homicide scene. In Shelby County and Cuyahoga County, police detectives conducted these investigations. In addition to recording the details of the incident for law-enforcement purposes, investigators obtained the names of persons close to the victim who might provide us with an interview at a later date, thereby serving as proxies for the victim. These lists were supplemented with names obtained from newspaper accounts, obituaries, and calls to funeral homes.

Approximately three weeks after a victims death, each proxy was sent a signed letter outlining the nature of the project. A $10 incentive was offered, and a follow-up telephone call was made a few days later to arrange a time and place for an interview. At the time of this meeting, informed consent was obtained.

Selection and Recruitment of Controls

After each interview with a case proxy, we sought a control subject matched to the case subject according to sex, race, age range (15 to 24 years, 25 to 40 years, 41 to 60 years, and 61 years or older), and neighborhood of residence. To minimize selection bias, the controls were identified by a previously validated procedure for the random selection of a matching household in the neighborhood8-10. After marking off a one-block avoidance zone around the home of the case subject, the interviewer started a neighborhood census at a randomly assigned point along a predetermined route radiating out from the case subjects residence. Households where no one was home were approached twice more, at different times of day and on different days of the week. If contact could not be established after three tries, no further efforts were made. After each neighborhood census was completed, an adult (a person 18 years old or older) in the first household with a member who met the matching criteria was offered a $10 incentive and asked to provide an interview. Whenever possible, attempts were made to interview a proxy for the actual matching control subject. When no interview was granted, the next matching household on the route was approached. If a closer match on the route was found on the second or third visit to the neighborhood, an adult respondent in the closer household was interviewed and any earlier, more distant interviews were discarded. Overall, census data were obtained from 70 percent of the households approached to identify each match. Eighty-four percent of the interviews were obtained from the closest matching household, 13 percent from the second, 3 percent from the third, and <1 percent from the fourth.

Interviews

Case and control interviews were identical in format, order, and content. Each was brief, highly structured, and arranged so that more sensitive questions were not broached until later in the interview. Items drawn from the Short Michigan Alcoholism Screening Test,11 the Hollingshead-Wilson two-factor index of social position,12 and a 1978 poll of gun ownership by Decision Making Information13 were included. Particularly sensitive questions were preceded by “permissive” statements, such as the following: “Many people have quarrels or fights. Has anyone in this household ever been hit or hurt in a fight in the home”?

Statistical Analysis

Data from reports prepared by medical examiners and police were used to describe the study population. Interview data were used for risk assessment, because these were collected in an analogous manner from the case proxies and matching control households. Since members of a household might acquire firearms or remove them from the home in response to a homicide in the neighborhood, answers were adjusted to reflect the state of affairs on the date of the homicide. Mantel-Haenszel chi-square analysis for matched pairs was used to calculate the crude odds ratio associated with each variable. Multivariate analyses used conditional logistic regression, the appropriate technique for a matched-pairs design14.

Potentially confounding variables were identified and controlled for by a two-step process. First, models containing closely related variables (such as those describing the use of alcohol in the home) were constructed to identify the variable or variables in each set that were most predictive of whether the household in question was a case or a control household. Next, a model that incorporated the variables selected in this initial step was constructed to select those that remained significant after we controlled for the effects of the remaining variables in the model. An additional model was constructed to look for interaction effects among the significant variables. Since no interaction terms significantly altered the adjusted odds ratios, the final model included six variables and was based on complete data from 316 matched pairs. After this analysis, an alternative modeling procedure was used to retain potentially confounding variables if they were even marginally significant (P<0.20). Although this approach added two variables, it did not significantly alter the adjusted odds ratios of the six included in our final model.

After completing this initial series of calculations, we examined the relation between homicide in the home and gun ownership, using various strata of the full study sample. To limit bias resulting from potentially faulty reporting, one analysis was limited to pairs with a case interview obtained from a proxy who lived in the home of the victim. To determine whether gun ownership was associated with an increased risk of homicide by firearms as compared with homicide by other means, cases were stratified according to method. To discern whether guns in the home decrease the risk of an intruder-related homicide or increase the risk of being killed by a family member, additional analyses stratified according to circumstance and the relationship between the victim and the offender were also conducted. After these were completed, a comparable series of stratified analyses was performed to assess more clearly the relation between homicide and previous violence in the home.

RESULTS

Study Population

There were 1860 homicides in the three counties during the study period. Four hundred forty-four (23.9 percent) took place in the home of the victim. After we excluded the younger victim in 19 double deaths, 2 homicides that were not reported to project staff, and 3 late changes to a death certificate, 420 cases (94.6 percent) were available for study.

Reports on the Scene

Most of the homicides occurred inside the victims home (Table 1TABLE 1Characteristics of 420 Homicides Committed in the Homes of the Victims.). Eleven percent occurred outside the home but within the immediate property lines. Two hundred sixty-five victims (63.1 percent) were men; 36.9 percent were women. A majority of the homicides (50.9 percent) occurred in the context of a quarrel or a romantic triangle. An additional 4.5 percent of the victims were killed by a family member or an intimate acquaintance as part of a murder-suicide. Thirty-two homicides (7.6 percent) were related to drug dealing, and 92 homicides (21.9 percent) occurred during the commission of another felony, such as a robbery, rape, or burglary. No motive other than homicide could be established in 56 cases (13.3 percent).

The great majority of the victims (76.7 percent) were killed by a relative or someone known to them. Homicides by a stranger accounted for only 15 cases (3.6 percent). The identity of the offender could not be established in 73 cases (17.4 percent). The remaining cases involved other offenders or police acting in the line of duty.

Two hundred nine victims (49.8 percent) died from gunshot wounds. A knife or some other sharp instrument was used to kill 111 victims (26.4 percent). The remaining victims were either bludgeoned (11.7 percent), strangled (6.4 percent), or killed by other means (5.7 percent).

Evidence of forced entry was noted in 59 cases (14.0 percent). Eighteen of these involved an unidentified intruder; six involved strangers. Two involved the police. The rest involved a spouse, family member, or some other person known to the victim.

Attempted resistance was reported in 184 cases (43.8 percent). In 21 of these (5.0 percent) the victim unsuccessfully attempted to use a gun in self-defense. In 56.2 percent of the cases no specific signs of resistance were noted. Fifteen victims (3.6 percent) were killed under legally excusable circumstances. Four were shot by police acting in the line of duty. The rest were killed by another member of the household or a private citizen acting in self-defense.

Comparability of Case Subjects and Controls

Potential proxy respondents were identified for 405 of the 420 case subjects (96.4 percent). Interviews were obtained from 93 percent of those approached in Shelby County, 99 percent in Cuyahoga County, and 98 percent in King County. The households of those who agreed to be interviewed did not differ from the households of those who refused with respect to the age, sex, or race of the victim or the method of homicide (firearm vs. other).

Interviews with a matching control were obtained for 99.7 percent of the case interviews, yielding 388 matched pairs. Three hundred fifty-seven pairs were matched for all three variables, 27 for two variables, and 4 for a single variable (sex). The demographic characteristics of the victims and controls were similar, except that the case subjects were more likely to have rented their homes (70.4 percent vs. 47.3 percent) and to have lived alone (26.8 percent vs. 11.9 percent) (Table 2TABLE 2Demographic Characteristics of 388 Pairs of Case Subjects and Controls.). Although efforts were made to conduct every interview in person, proxy respondents for the case subjects were much more likely than the controls to request a telephone interview (40.2 percent vs. 12.6 percent). Despite efforts to interview a proxy respondent for each control, only 48.2 percent of the control interviews were obtained in this manner.

Univariate Analysis

Alcohol was more commonly consumed by one or more members of the households of case subjects than by members of the households of controls (Table 3TABLE 3Univariate Analysis of Hypothesized Risk on Protection Factors Derived from Data on 388 Matched Pairs of Case Subjects and Controls.). Alcohol was also more commonly consumed by the case subjects themselves than by their matched controls. Case subjects were reported to have manifested behavioral correlates of alcoholism (such as trouble at work due to drinking) much more often than matched controls. Illicit-drug use (by the case subject or another household member) was also reported more commonly by case households than control households.

Previous episodes of violence were reported more frequently by members of case households. When asked if anyone in the household had ever been hit or hurt in a fight in the home, 31.8 percent of the proxies for the case subjects answered affirmatively, as compared with only 5.7 percent of controls. Physical fights in the home while household members were drinking and fighting severe enough to cause injuries were reported much more commonly by case proxies than controls. One or more members of the case households were also more likely to have been arrested or to have been involved in a physical fight outside the home than members of control households.

Similar percentages of case and control households reported using deadbolt locks, window bars, or metal security doors. The case subjects were slightly less likely than the controls to have lived in a home with a burglar alarm, but they were slightly more likely to have controlled security access. Almost identical percentages of case and control households reported owning a dog.

One or more guns were reportedly kept in 45.4 percent of the homes of the case subjects, as compared with 35.8 percent of the homes of the control subjects (crude odds ratio, 1.6; 95 percent confidence interval, 1.2 to 2.2). Shotguns and rifles were kept by similar percentages of households, but the case households were significantly more likely to have a handgun (35.7 percent vs. 23.3 percent; crude odds ratio, 1.9; 95 percent confidence interval, 1.4 to 2.7). Case households were also more likely than control households to contain a gun that was kept loaded or unlocked (Table 3).

Multivariate Analysis

Six variables were retained in our final conditional logistic-regression model: home rented, case subject or control lived alone, any household member ever hit or hurt in a fight in the home, any household member ever arrested, any household member used illicit drugs, and one or more guns kept in the home (Table 4TABLE 4Variables Included in the Final Conditional Logistic-Regression Model Derived from Data on 316 Matched Pairs of Case Subjects and Controls.). Each of these variables was strongly and independently associated with an increased risk of homicide in the home. No home-security measures retained significance in the final model. After matching for four characteristics and controlling for the effects of five more, we found that the presence of one or more firearms in the home was strongly associated with an increased risk of homicide in the home (adjusted odds ratio, 2.7; 95 percent confidence interval, 1.6 to 4.4).

Stratified analyses with our final regression model revealed that the link between guns and homicide in the home was present among women as well as men, blacks as well as whites, and younger as well as older people (Table 5TABLE 5Homicide in the Home in Relation to Gun Ownership, According to Subgroup.). Restricting the analysis to pairs with data from case proxies who lived in the home of the victim demonstrated an even stronger association than that noted for the group overall. Gun ownership was most strongly associated with homicide at the hands of a family member or intimate acquaintance (adjusted odds ratio, 7.8; 95 percent confidence interval, 2.6 to 23.2). Guns were not significantly linked to an increased risk of homicide by acquaintances, unidentified intruders, or strangers. We found no evidence of a protective benefit from gun ownership in any subgroup, including one restricted to cases of homicide that followed forced entry into the home and another restricted to cases in which resistance was attempted. Not surprisingly, the link between gun ownership and homicide was due entirely to a strong association between gun ownership and homicide by firearms. Homicide by other means was not significantly linked to the presence or absence of a gun in the home.

Living in a household where someone had previously been hit or hurt in a fight in the home was also strongly and independently associated with homicide, even after we controlled for the effects of gun ownership and the other four variables in our final model (adjusted odds ratio, 4.4; 95 percent confidence interval, 2.2 to 8.8) (Table 4). Previous family violence was linked to an increased risk of homicide among men as well as women, blacks as well as whites, and younger as well as older people (Table 6TABLE 6Homicide in the Home in Relation to Prior Domestic Violence, According to Subgroup.). Virtually all of this increased risk was due to a marked association between prior domestic violence and homicide at the hands of a family member or intimate acquaintance (adjusted odds ratio, 20.4; 95 percent confidence interval, 3.9 to 104.6).

DISCUSSION

Although firearms are often kept in homes for personal protection, this study shows that the practice is counterproductive. Our data indicate that keeping a gun in the home is independently associated with an increase in the risk of homicide in the home. The use of illicit drugs and a history of physical fights in the home are also important risk factors. Efforts to increase home security have largely focused on preventing unwanted entry, but the greatest threat to the lives of household members appears to come from within.

We restricted our study to homicides that occurred in the home of the victim, because these events can be most plausibly linked to specific individual and household characteristics. If, for example, the ready availability of a gun increases the risk of homicide, this effect should be most noticeable in the immediate environment where the gun is kept. Although our case definition excluded the rare instances in which a nonresident intruder was killed by a homeowner, our methodology was capable of demonstrating significant protective effects of gun ownership as readily as any evidence of increased risk.

Previous studies of risk factors for homicide have employed correlational analysis15 or retrospective-cohort16 or time-series17 designs to link rates of homicide to specific risk factors. However, hazards suggested by ecologic analysis may not hold at the level of individual households or people18. In contrast to these approaches, the case-control method studies individual risk factors in relation to a specific outcome of interest. Case-control research is particularly useful when the list of candidate risk factors is large and the rate of adverse outcomes is relatively low. Under these circumstances, it is usually the analytic method of choice19.

Although case-control studies offer many advantages over ecologic studies, they are prone to several sources of bias. To minimize selection bias, we included all cases of homicide in the home and rigorously followed an explicit procedure for randomly selecting neighborhood control subjects. High response rates among case proxies (92.6 percent) and matching controls (80.6 percent) minimized nonresponse bias. Case respondents did not differ significantly from nonrespondents with regard to the age, sex, and race of the victim and the type of weapon involved. Although double homicides and murder-suicides were considered single events to avoid overrepresenting their effects, the number of cases excluded for this reason was small.

Other threats to the validity of the study were less easy to control. A respondents recollection of events can be powerfully affected by a tragedy as extreme as a homicide in the home. To diminish the effect of recall bias, we delayed our contact with the case proxies to allow for an initial period of grief. We also used a simple, forced-choice questionnaire to ascertain information in a comparable manner from case proxies and controls. We tried to obtain data on victims and controls as similarly as possible by interviewing proxy respondents for the controls whenever possible. Although we were able to do so only 48 percent of the time, the responses we obtained from this subgroup were consistent with those obtained from the study population overall.

Potential misreporting of sensitive information was a serious concern, since we had no way to verify each respondents statements independently. If case proxies or controls selectively withheld sensitive information about illicit-drug use, alcoholism, or violence in the home, inaccurate estimates of risk could result. We attempted to minimize this problem by reassuring our respondents of the confidentiality of their responses. We also placed “permissive” statements before each potentially intrusive question to encourage honest replies. Very few respondents refused to answer our questions, although all were assured that they were free to do so.

The rate of domestic violence reported by our control respondents was somewhat less than that noted in a large telephone survey20. This may be due to regional or temporal differences in rates of battering, variations in the way we phrased our questions (e.g., screening as compared with an exploratory line of inquiry), or the increased anonymity afforded by telephone interviews as compared with our face-to-face encounters.

Underreporting of gun ownership by control respondents could bias our estimate of risk upward. We do not believe, however, that misreporting of gun ownership was a problem. In two of our three study communities, a pilot study of homes listed as the addresses of owners of registered handguns confirmed that respondents answers to questions about gun ownership were generally valid21. Furthermore, the rate of gun ownership reported by control respondents in each study community was comparable to estimates derived from previous social surveys22 and Cooks gun-prevalence index15.

Four limitations warrant comment. First, our study was restricted to homicides occurring in the home of the victim. The dynamics of homicides occurring in other locations (such as bars, retail establishments, or the street) may be quite different. Second, our research was conducted in three urban counties that lack a substantial percentage of Hispanic citizens. Our results may therefore not be generalizable to more rural communities or to Hispanic households. Third, it is possible that reverse causation accounted for some of the association we observed between gun ownership and homicide -- i.e., in a limited number of cases, people may have acquired a gun in response to a specific threat. If the source of that threat subsequently caused the homicide, the link between guns in the home and homicide may be due at least in part to the failure of these weapons to provide adequate protection from the assailants. Finally, we cannot exclude the possibility that the association we observed is due to a third, unidentified factor. If, for example, people who keep guns in their homes are more psychologically prone to violence than people who do not, this could explain the link between gun ownership and homicide in the home. Although we examined several behavioral markers of violence and aggression and included two in our final logistic-regression model, “psychological confounding” of this sort is difficult to control for. “Psychological autopsies” have been used to control for psychological differences between adolescent victims of suicide and inpatient controls with psychiatric disorders,23,24 but we did not believe this approach was practical for a study of homicide victims and neighborhood controls. At any rate, a link between gun ownership and any psychological tendency toward violence or victimization would have to be extremely strong to account for an adjusted odds ratio of 2.7.

Given the univariate association we observed between alcohol and violence, it may seem odd that no alcohol-related variables were included in our final multivariate model. Although consumption of alcoholic beverages and the behavioral correlates of alcoholism were strongly associated with homicide, they were also related to other variables included in our final model. Forcing the variable “case subject or control drinks” into our model did not substantially alter the adjusted odds ratios for the other variables. Furthermore, the adjusted odds ratio for this variable was not significantly greater than 1.

Large amounts of money are spent each year on home-security systems, locks, and other measures intended to improve home security. Unfortunately, our results suggest that these efforts have little effect on the risk of homicide in the home. This finding should come as no surprise, since most homicides in the home involve disputes between family members, intimate acquaintances, friends, or others who have ready access to the home. It is important to realize, however, that these data offer no insight into the effectiveness of home-security measures against other household crimes such as burglary, robbery, or sexual assault. In a 1983 poll, Seattle homeowners feared “having someone break into your home while you are gone” most and “having someone break into your home while you are at home” 4th on a list of 16 crimes25. Although homicide is the most serious of crimes, it occurs far less frequently than other types of household crime2. Measures that make a home more difficult to enter are probably more effective against these crimes.

Despite the widely held belief that guns are effective for protection, our results suggest that they actually pose a substantial threat to members of the household. People who keep guns in their homes appear to be at greater risk of homicide in the home than people who do not. Most of this risk is due to a substantially greater risk of homicide at the hands of a family member or intimate acquaintance. We did not find evidence of a protective effect of keeping a gun in the home, even in the small subgroup of cases that involved forced entry.

Saltzman and colleagues recently found that assaults by family members or other intimate acquaintances with a gun are far more likely to end in death than those that involve knives or other weapons26. A gun kept in the home is far more likely to be involved in the death of a member of the household than it is to be used to kill in self-defense4. Cohort and interrupted time-series studies have demonstrated a strong link between the availability of guns and community rates of homicide2,15-17. Our study confirms this association at the level of individual households.

Previous case-control research has demonstrated a strong association between the ownership of firearms and suicide in the home10,23,24. Also, unintentional shooting deaths can occur when children play with loaded guns they have found at home27. In the light of these observations and our present findings, people should be strongly discouraged from keeping guns in their homes.

The observed association between battering and homicide is also important. In contrast to the money spent on firearms and home security, little has been done to improve societys capacity to respond to the problem of domestic violence28,29. In the absence of effective intervention, battering tends to increase in frequency and severity over time28-30. Our data strongly suggest that the risk of homicide is markedly increased in homes where a person has previously been hit or hurt in a family fight. At the very least, this observation should prompt physicians, social workers, law-enforcement officers, and the courts to work harder to identify and protect victims of battering and other forms of family violence. Early identification and effective intervention may prevent a later homicide31,32.

Supported by grants (CCR 402424 and CCR 403519) from the Centers for Disease Control and Prevention.

We are indebted to the men and women of the following law-enforcement agencies and offices for their support of this project: in Shelby County, Tennessee, the Memphis Police Department, Shelby County Sheriffs Department, Bartlett Police Department, Collierville Police Department, Germantown Police Department, Millington Police Department, and Shelby County Medical Examiners Office; in Cuyahoga County, Ohio, the Cleveland Police Department and Cuyahoga County Coroners Office; and in King County, Washington, the Seattle Police Department, Bellevue Police Department, King County Sheriffs Department, and King County Medical Examiners Office. Without their assistance, this work would not have been possible. We are also indebted to Noel Weiss and William Applegate for their comments and suggestions, to Vivian C. Driscoll and Steven Walker for their help with data collection, and to LaGenna Betts for her assistance in the preparation of the manuscript.

SOURCE INFORMATION

From the Departments of Internal Medicine (A.L.K., J.G.B., B.B.H.), Preventive Medicine (A.L.K.), Biostatistics and Epidemiology (A.L.K., G.S.), and Pathology (J.T.F), University of Tennessee, Memphis; the Departments of Pediatrics (F.P.R.), Epidemiology (F.P.R.), and Pathology (D.T.R), University of Washington, Seattle; Harborview Injury Prevention and Research Center, Seattle (F.P.R., J.P.); and the Departments of Biology (N.B.R., A.B.L.) and Epidemiology and Biostatistics (N.B.R.) and the Center for Adolescent Health (N.B.R.), Case Western Reserve University, Cleveland.

Address reprint requests to Dr. Kellermann at the Emory Center for Injury Prevention, School of Public Health, Emory University, 1599 Clifton Rd., Atlanta, GA 30329.

TOOLS