A gun in the home increases personal safety.


Keeping a gun in the home carries a murder risk 2.7 times greater than not keeping one, according to a study by Arthur Kellermann. The National Rifle Association has fiercely attacked this study, but it remains valid despite its criticisms. The study found that people are 21 times more likely to be killed by someone they know than a stranger breaking into the house. Half of the murders were over arguments or romantic triangles. The study also found that the increased murder rate in gun-owning households was entirely due to an increase in gun homicides only, not any other murder method. It further found that gun-owning households saw an increased murder risk by family or intimate acquaintances, not by strangers or non-intimate acquaintances. The most straightforward explanation is that the presence of a gun increases the possibility that a normal family fight or drinking binge will become deadly. No other explanation fits the above facts.


Most people keep guns in their homes for self-protection. The image of an unknown criminal breaking into your house is an important one for gun advocates, because it justifies keeping a gun in the home. But to gun control advocates, a gun in the home means that a family fight or a drinking binge is more likely to turn deadly. Which view is more accurate?

In an attempt to answer this question, a team led by Dr. Arthur Kellermann of Emory University conducted a survey of 388 homes that had experienced homicides. (1) They found that 76.7 percent of the victims were killed by a spouse, family member or someone they knew, and there was no forced entry into the home 84.3 percent of the time. Strangers comprised only 3.6 percent of the killers. However, the killer was never identified in 17.4 percent of the cases.

Of the 420 homicides they originally investigated, 96.4 percent were illegal. Only 3.6 percent were ruled legally excusable homicide (that is, self-defense).

After eliminating the impact of other variables like illegal drugs and domestic violence, the researchers found that the risk of getting killed was 2.7 times greater in homes with a gun than without them. No protective benefit of possessing a firearm was ever found, not even for a single one of the 14 subgroups studied.

Needless to say, the National Rifle Association and other pro-gun advocates have fiercely attacked this survey. Kellermann's work has been branded "junk science," "unpublishable," "biased," "seriously flawed," "fraudulent" and "grand malpractice." The NRA also criticized the Centers for Disease Control for continuing to fund such anti-gun research, and the Republican Congress pressured the CDC to shut it down completely. Thus, the reaction of Republicans and the NRA to this controversial study was not to call for more studies to clarify the issue, but to censor all further scientific research.

Pro-gun advocates respond that they are not promoting censorship, only objecting to wasting tax dollars on blatantly biased, deeply flawed research. Pro-gunners feel that the sound bites generated by this study will become part of a popular mythology against guns that will be hard to correct. But this objection is based on a faulty view of the research method. The best way to correct bad science is to subject it to expert criticism: namely, peer review. Kellermann's study was, and it passed. Pro-gunners might then wish to criticize the peer review process. For example, they might accuse Kellermann's peer reviewers of sharing his bias (although there are protocols in peer review to avoid this). The principled response, then, would be to examine and reform the peer review process. For example, Republicans in Congress might have called for pro-gun criminologists like Gary Kleck to be included in all future peer review of CDC-funded studies. (Not likely, given Kleck's aversion to the process.) Another principled response would be for the NRA -- one of the richest organizations in America -- to start funding its own research by way of rebuttal. But to shut down all further research is both censorship and anti-science.

It is apparent from the attacks on Kellermann's study that most of his critics have not even read it. Simply reading the original article in The New England Journal of Medicine (October 7, 1993) would answer 95 percent of their objections. The study was well-designed and is entirely valid. Like any scientific study, it has its limitations. It does not prove that guns cause a higher murder rate in the home, only that the two are associated for some reason. And there are more variables that need to be explored. But the Kellermann study is a legitimate addition to the small but growing scientific literature on the benefits and costs of domestic firearms.

The rest of this essay will be divided into two parts: a detailed description of the Kellermann survey, and a rebuttal of its criticisms.

The survey

Kellermann chose to conduct this survey using the "case-control method" (or CCM). This method examines the differences between two groups: one that possesses a certain trait, and another that does not. For instance, a researcher may compare a "case group" that has lung cancer to a "control group" that is free of the disease. After asking them questions about their behavior and environment, he may learn that the case group generally smokes, but the control group does not. Conclusion: smoking is correlated to lung cancer. In this instance, the arrow of causality is easy to determine, because it is unlikely that lung cancer causes people to start smoking. But sometimes the arrow of causality is more difficult to determine, as in the case of gun ownership and murder.

Kellermann's team identified 388 victims ("case subjects") who were killed in private homes. Surviving members of the household ("proxies") formed the case group which answered the survey. The researchers also gave an identical survey to a control group of 388 other people, who were matched to the victims by age, race, sex and neighborhood.

The homicides which were studied came from three metropolitan areas. The first two were Shelby County, Tennessee (which includes Memphis), and King County, Washington (which includes Seattle), both from August 1987 to August 1992. The third was Cuyahoga County, Ohio (which includes Cleveland), from January 1990 to August 1992. King County is predominately white and enjoys a relatively high standard of living. Cuyahoga County is 25 percent African-American, as is 44 percent of Shelby County. The poverty levels of these counties were 5, 11 and 15 percent, respectively. (The national poverty rate in 1992 was 15 percent.)

The team originally identified 444 cases of homicide in the home, about a fourth of the total number of homicides for those counties. This number was reduced to 420 for the study for various reasons, then to 405 because a control couldn't be found, and then to 388 because a proxy couldn't be interviewed. The high response rate of case proxies (92.6 percent) and matching controls (80.6 percent) is typically considered to have minimized nonresponse bias.

The survey asked 31 questions about the subjects' environment and behavior. The results are listed below. The first two columns reflect the percentage of those who answered yes to the question. The third column reflects the crude odds that a murder would be more likely for those who answered yes. For example, for the first question, murder was 2.4 times more likely in a household where any member drank alcohol. An odds ratio of 1.0 represents no extra risk. Keep in mind that the crude odds are confounded by other variables, and by themselves do not tell the whole story. Another analytical step is needed to arrive closer to the truth.

                              Case      Control    Crude odds
Behavioral factors            Subjects  Subjects   Ratio
Any household member drank     73.3%     55.9%      2.4
    alcoholic beverages
Case subject or control drank  62.8      41.9       2.6
    alcoholic beverages
Drinking caused problems       24.8       5.7       7.0
    in the household
Any household member had        9.0       0.8      10.7
    trouble at work because
    of drinking
Case subject or control had     5.5       0.3      20.0
    at work because of
Any household member           11.4       2.3       9.8
    hospitalized because of
Case subject or control         7.6       0.5       14.0
    hospitalized because of
Any household member used      31.3       6.0        9.0
    illicit drugs
Case subject or control        20.3       4.2        6.8
    used illicit drugs
Any physical fights in the     25.3       3.4        8.9
    home during drinking
Any household member hit or    31.8       5.7        7.9
    hurt in a fight in the
Any family member required     17.3       2.1       10.2
    medical attention because
    of a fight in the home
Any adult household member     29.9       18.8       2.1
    involved in a physical
    fight outside the home
Any household member arrested  52.7       23.4       4.2
Case subject or control        36.0       15.7       3.5

Environmental Factors
Home Rented                    70.4       47.6       5.9
Public Housing                 11.1        9.8       1.5
Case subject or control        26.8       11.9       3.4
    lived alone
Deadbolt locks 68.8 75.3 0.8 Window bars 19.2 20.9 0.8 Metal Security Door 25.4 26.8 0.9 Burglar alarm 7.1 11.1 0.6 Controlled security access 13.9 9.8 2.3 to residence
Dog or dogs in home 24.2 22.4 1.1 Gun or guns in home 45.4 35.8 1.6 Handgun 35.7 23.3 1.9 Shotgun 13.6 16.8 0.7 Rifle 12.2 13.9 0.8 Any gun kept unlocked 29.6 17.8 2.1 Any gun kept loaded 26.7 12.5 2.7 Guns kept primarily for 32.6 22.2 1.7 self-defense

The above chart is an example of "univariate analysis," or a straight comparison between the two groups. But this analysis is incomplete. There are many variables that simultaneously contribute to the odds of a person being murdered: drug use, domestic violence, criminal history, level of protection, etc. A person who answers yes to the question "Does anyone in the house use illicit drugs?" might be nine times more likely to be murdered, but that doesn't eliminate all the other variables that also contribute to the total murder risk. To isolate the risk attributed to drug use alone, researchers need to perform "multivariate analysis," which zeroes out all these other factors. That way, we can learn how drug use in and of itself raises the murder risk.

Kellermann's team found only six variables that were strong enough to be included in the final model. They found that the following variables were associated with the following increased murder risks:

                                Murder risk,
Variable                        Odds adjusted ratio
Illicit drug use                5.7 times
Being a renter                  4.4
Household member hit or
  hurt in a fight in the home   4.4
Living alone                    3.7
Guns in the house               2.7
Household member arrested       2.5

If there were a protective benefit to having a gun in the home, this survey would have found it. After all, if the survey could detect an increased murder risk from the presence of a gun in the home, there's no reason it couldn't from the absence of one as well. But the team found no protective benefits of a gun in the home whatsoever, for any of the subgroups studied.

Of all the methods of murder, guns were responsible for 49.8 percent of the victims killed at home. In homes that kept a gun, the overall murder risk was 2.7 times greater, but for gun homicides it was 4.8, while for non-gun homicides it was 1.2. Notice that 1.2 is not significantly different from 1, so there was no increased risk for non-gun homicides. In other words, people who kept a gun in the home were at higher risk for gun homicides only, not any other type of homicide. This is an important point, because it strongly suggests that gun availability tends to turn ordinary family arguments into something fatal, rather than the murder victims knew they were at risk and armed themselves with a gun.

Alcohol was not included in the multivariate analysis, despite its strong association in the univariate analysis, because alcohol was also related to all the other variables in the final model. Including alcohol in the final model did not substantially alter the results. Furthermore, the odds-adjusted ratio of alcohol was not significantly greater than 1.

The researchers also conducted a stratified analysis of their final model, which found that the link between guns and homicide existed in all 14 subgroups studied. This included women as well as men, whites as well as blacks, and the old as well as the young. Most tellingly, they found the strongest association between guns and homicide among family members and intimate acquaintances (7.8 times more likely). Guns were much less associated to homicides by acquaintances, unidentified intruders, or strangers (1.8 times). Again, this supports the interpretation that guns allow family fights to turn deadly. Here is a complete list of the murder risk by subgroup:

                          Murder risk,
Subgroup                  Adjusted odds ratio
   Female                 3.6 times
   Male                   2.3
   White                  2.7
   Black                  2.9
   15-40                  3.4
   Over 40                2.3
Suspect related to or
intimate with victim:
   Yes                    7.8
   No                     1.8
Evidence of forced entry
   Yes                    2.5
   No                     2.8
Victim resisted assailant
   Yes                    3.0
   No                     3.1
Method of homicide
   Firearm                4.8
   Other                  1.2

Also revealing are the circumstances surrounding the 420 homicides:

Characteristic                   Percent of victims
   Inside residence                88.8%
   Within immediate property       11.2
Sex of victim
   Female                          36.9
   Male                            63.1
Race or ethnic group of victim
   White                           33.3
   Black                           61.9
   Native American, Eskimo, Aleut   1.0
   Asian or Pacific Islander        1.7
   Other                            2.1
Age of victim (years)
   15-24                           13.8
   25-40                           40.7
   41-60                           25.2
   Over 61                         20.2
   Altercation or quarrel          44.0
   Romantic triangle                6.9
   Murder-suicide                   4.5
   Felony-related                  21.9
   Drug-dealing                     7.6
   Homicide only                   13.3
   Other                            1.7
Relationship of offender to victim
Spouse 16.7 Intimate acquaintance 13.8 First-degree relative 9.5 Other relative 2.9 Roommate 2.9 Friend or acquaintance 31.0 Police officer 1.0 Stranger 3.6 Unknown (unidentified suspect) 17.4 Other 1.4 Method of homicide Handgun 42.9 Rifle 2.4 Shotgun 3.6 Unknown firearm 1.0 Knife or sharp instrument 26.4 Blunt instrument 11.7 Strangulation or suffocation 6.4 Burns, smoke, scalding 2.4 Other 3.3 Victim resisted assailant Yes 43.8 No 33.3 Not noted 22.9 Evidence of forced entry Yes 14.0 No 84.3 Not noted 1.7 Legally excusable homicide Yes 3.6 No 96.4

Several points about this chart are noteworthy. The first is that at least 76.7 percent of the murderers were relatives, friends or acquaintances of the victim. In fact, the victim's murderer was 21 times more likely to be a relative or acquaintance than a stranger. Even in the 14 percent of the cases involving forced entry, the vast majority of the intruders were known to the victim. The threat of forced entry is the most commonly cited reason for possessing a domestic firearm, but the researchers found no protective benefit for this subgroup either.

The researchers write: "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."

Of the 388 homicides surveyed, 21 victims died while unsuccessfully trying to defend themselves with a gun. Only 15 of the deaths were ruled justifiable homicide or legal self-defense, and four of these were by the police.

The authors did present their study with several limitations. First, they acknowledged that they limited their study of homicides to those which occurred in the home, their goal simply being to measure the effectiveness of gun protection in the home. Homicides at other locations (such as bars, work or the streets) were not counted. Therefore, the dynamics of homicide in these locations might be quite different.

Second, they acknowledged that their research was conducted in urban settings that lacked a substantial Hispanic population. The dynamics of homicide in that community therefore might be quite different.

Third, they acknowledged that the arrow of causality could point in the opposite direction in some of the cases. For example, a person might acquire a gun in response to a specific threat. If the threat was then carried out, the correlation between the gun and the murder could be partly attributed to the failure of the weapon to provide protection.

Fourth, they acknowledged that a third, unidentified factor might be responsible for both gun possession and murder risk. For example, the victims may have had violent, aggressive personalities or some other psychological disorder that predisposed them to both greater gun possession and murder. The authors note that they included several behavioral markers for aggression and violence in their survey, but they did not conduct a full "psychological autopsy" given the impractical nature of such a task. Still, they note that "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."

So, what are the study's conclusions? The authors write:

"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 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."

It is important to note that Kellermann's findings agree with many other studies. For example, the FBI reports that in 1993, only 1.7 percent of all handgun murders were justifiable homicides. Kellermann's team found that only 3.6 percent of the 420 homicides it studied were justifiable. The FBI found 19.1 percent of all homicides to be felony-related; Kellermann found 21.9 percent of those in the home to be felony-related. In 1994, the FBI found that only 13 percent of all murder victims were killed by strangers. Kellermann found that 3.6 percent of the domestic homicides were strangers and 17.4 percent were never identified. The FBI found that 12 percent of all killers in 1994 were related to the victim; Kellermann found this figure to be 12.4 percent in domestic homicides. (2)

Kellermann's research also confirms numerous studies like the one done by Linda Saltzman, which found that assaults by family members or intimate acquaintances are far more fatal when the weapon is a gun. (3) There are also many cohort and interrupted time-series studies that demonstrate a strong link between gun availability and homicide rates in the community. (4) Kellermann's study has now confirmed this correlation at the individual household level as well.

Criticisms of the study

Pro-gun advocates have raised a number of objections to this survey. The following are actual arguments taken from the Internet and the NRA: (5)

1. "99.8 percent of the protective uses of guns do not involve homicides," says Paul Blackman of the NRA. Defensive gun uses include waving the weapon, firing warning shots, wounding the intruder, etc.

2. Guns do not emit magic rays that control people's minds, or magnetize murderers to the doorstep. 3. People threatened by violence bought guns to defend themselves, hence the correlation between gun ownership and murder. 4. Kellermann's study didn't document whether a firearm used in a particular homicide was the same one kept in the home, or whether it might have been carried in by the murderer. 5. Proxies for the murder victim were not asked if the gun had previously been used for self-defense. 6. "These people were highly susceptible to homicide," says Paul Blackman of the NRA. "We know that because they were killed." 7. Of course if someone gets shot in their home, there's bound to be a gun in the home. And drowning victims are always found near water. 8. The majority of the homicides were not committed by guns, so could not have been committed by Kellermann's scary "guns in the home." 9. The researchers did not include in their analysis those cases where the home-owner shot a non-resident intruder. 10. This study was conducted by medical doctors who were out of their league; this is an issue best left for criminologists. 11. The use of the case-control method allows for spurious associations. 12. A disproportionate number of survey respondents were criminals, hence the correlation between gun ownership and murder. 13. The study was conducted in urban areas, which have high crime. This would promote both gun ownership and death in violent crimes. 14. The victims typically had stunningly different lifestyles from the controls: more drinking, crime, drug use, domestic violence, etc. 15. The survey failed to ask about other variables. 16. The survey failed to determine the strength of the variables (severity of drug use, domestic violence, crime, etc.) 17. The study underestimated the amount of drug use or other domestic problems, which was really the cause of an increased murder risk for gun owners. 18. The number of guns in the control homes were underreported. Conclusion

The Kellermann study is valid, if incomplete -- as any study must necessarily be. More research needs to be done on other possible variables contributing to the murder rate, although Kellermann has apparently identified the most important ones. The results could be refined by determining the severity of some factors, like criminal background. And it would be good to reconfirm the honesty of the respondents' answers. But the study itself is sound, and gun-control advocates can use it with confidence.

Return to Overview

1. Arthur Kellermann et. al., "Gun Ownership as a Risk Factor for Homicide in the Home," The New England Journal of Medicine, October 7, 1993, pp. 1084-1091.

2. Federal Bureau of Investigations, Crime in the United States, annual.

3. Linda Saltzman, et. al., "Weapon Involvement and Injury Outcomes in Family and Intimate Assaults," Journal of the American Medical Association, 1992;267, pp. 3043-7.

4. A.J. Reiss, Jr. and J.A. Roth, eds., Understanding and Preventing Violence: Panel on the Understanding and Control of Violent Behavior (Washington D.C.: National Academy Press, 1993), pp. 42-97; P.J. Cook, "The Effect of Gun Availability on Robbery and Robber Murder: A Cross Section Study of Fifty Cities," Policy Stud Rev Annu 1979;3, pp. 743-81; J.H. Sloan, A.L. Kellermann, D.T. Reay, et. al., "Handgun Regulations, Crime, Assaults, and Homicide: a Tale of Two Cities," New England Journal of Medicine, 1988;319, pp. 1256-62; C. Loftin, et. al., "Effects of Restrictive Licensing of Handguns on Homicide and Suicide in the District of Columbia," New England Journal of Medicine, 1991;325, pp. 1615-20.

5. I am deeply indebted to Tim Lambert of the University of New South Wales for providing many of these objections and rebuttals, which came from his archived postings to the Internet newsgroup talk.politics.guns. Many of the responses here are based on his expert answers.

6. J.D. Wright, P. Rossi, K. Daly, E. Weber-Burdin, "Weapons, crime and violence in America: a literature review and research agenda," (Washington, D.C.: Government Printing Office, 1983), pp. 212-60, 361-411; P.J. Cook, "The effect of gun availability on robbery and robber murder: a cross section study of fifty cities," Policy Stud Rev Annu 1979; 3, pp. 743-81.