The study from which these claims arise was performed by Arthur Kellermann and two other doctors and reported in volume 329 (1993) of the new england journal of medicine (nejm).  They compared households where a homicide occurred from August 1987 through August 1992 with households where no homicide occurred in three counties chosen for their locations convenient to the three researchers.

For most of the homicides that occurred in homes, the researchers interviewed people (called "proxies") who knew them, in order to learn some things about the victims.  The researchers also identified a group of control subjects matched to the victims for age group, sex and race, and also identified some people who knew these control subjects, with the purpose of asking these people (proxies for the control subjects) about the control subjects just as they asked proxies about the homicide victims.  However, it turned out that the proxies were used for less than half of the control subjects, and the control subjects themselves were asked the questions in the other cases.

After correcting (validly?) for several factors (other than gun ownership) that they found to correlate with the presence of homicide--alcohol use, illegal drug use, past domestic violence, and criminal records--the researchers determined that households where "homicide at the hands of a family member or intimate acquaintance" occurred were almost three times likelier to have kept a gun in the home than control households where such a homicide did not occur.  They had similar findings about unqualified homicide in the home and about having loaded handguns and unlocked guns.

The researchers concluded in their report, "Although firearms are often kept in the home 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."

So, what was wrong with the study?  As with Kellermann's other studies, there really wasn't anything wrong with the study itself except that it didn't actually prove much of anything.  After all, they did collect some data.  The real deficiencies are in what gun control advocates try to make of the study data and results.

FIRST, the kind of study the researchers did can prove that one thing is related to another but cannot prove that one of them causes the other.  Assuming that the researchers' study was valid and correctly done, all that it proves is that a home in which someone is killed is x times more likely to have had a gun in it than homes that are similar in whatever ways the researchers controlled for.

Does the result mean that the presence of the gun caused the killing, or does it mean that people who are at high risk of being killed for other reasons get a gun for protection?  Or, does it just mean that gun possession is related to any number of other things that themselves relate to homicide?  The study method used cannot make these determinations.

Households in which people are murdered strongly tend to be very much different than "average" households in numerous ways.  In this study, for example, 71 percent of the victims had high rates of past criminal activities.  Because of the huge differences between homicide households and typical US households, it is practically impossible for a researcher to control for all of them even when the researcher is highly knowledgable of crime, criminals, guns, and gun use.

Characteristics matching is the way this study was supposed to account for these "confounding" factors, but no life-style factors were considered in the matching.  The number of victims who lived alone was significantly higher than the number of "matching" control subjects.  The controls were all selected to live at least a block away from every victim, so the controls tended to be outside the neighborhoods where crime was highest (so they would naturally tend less to have guns for defense).

Medical researchers are not accustomed to such complexity and don't have the necessary knowledge of all the interfering (confounding) factors.  So, the results of the Kellermann study are really applicable just to households in which people are murdered in the three counties of the Kellermann study, since this is what the researchers started with, rather than all households throughout the U.S. or all U.S. households in which there are guns.  The homicide rates in the three counties in which the study was done were at the time approximately 50% greater than the overall U.S. rate, so the counties were hardly typical of the U.S.

SECOND, the study did not use a very large sample.  The researchers did not use all the home homicide cases they had records of.  There were 444 homicides meeting the "home" criterion.  24 were excluded for what the researchers called "various reasons" leaving 94.6%.  But, then, 7% were dropped because of failure to interview the proxy, and an additional 1% due to failure to find a control, leaving 388 matched pairs.  But, the researchers did not get complete data for all the victims of these 388 pairs, so there were only 316 matched pairs used in the final analysis.  This was only 71.2% of the "home homicides." This is far from "all" of the home homicides as claimed by the researchers.

THIRD, the researchers' data does not show that the guns used in the homicides were generally guns that were ordinarily kept in the homes.  70.9% of the victims were killed by people whose relationship to the victim indicates that the killer did not live in the victim's household, and thus presumably used a gun not kept in the victim's household.  Criminologists who were allowed access to the researchers' data only after many years had passed have been able to determine that the firearm in the home for this study was actually the homicide weapon in no more than 4 percent of the cases.  So, a "family gun" is not more likely to kill you or someone you know than to kill in self defense.  And the home having a gun did not cause the killings; the home probably had the gun because the killing was already likely.

FOURTH, and MOST IMPORTANTLY, the Kellermann study method is not valid because the keystone data element is doomed to be incorrect information.  In the study, the researchers asked some people who knew the victims some questions about guns in the homes of the victims—like whether there was a gun or guns in the home and whether or not the gun was normally kept loaded or unlocked.  The people asked these questions would assume that the answers were already available to the researchers since police had already established much of the facts about the homicides.  So there would be essentially no point to the people lying about the gun matters.  This info about guns in the victims' homes would therefore be very accurate.

The same can't be said of the answers by the control subjects or the proxies for control subjects.  Many of the subjects and proxies for subjects would deny gun ownership, especially since the Kellermann study was being performed for a government agency (the Centers for Disease Control and Prevention).

Of the homicide victim cases, 45.4% of the proxies said that someone in the household owned a gun.  Of the 388 control subjects, 139 of them or their proxies admitted to household possession of a gun for 35.8%, which is obviously less than for the homicide victims.  This difference is the reason the researchers came up with their "2.7 times" or "3 times" figures.

But easily 20% of people asked about gun ownership deny ownership even when they do own a gun.  Even more will when the question is being asked by or for their government.  A 20% false reply rate would require a 25% augmentation of the number who actually admitted possession.  Adding 25% to the number of control subjects and proxies that admitted gun ownership gives a number that is practically the same percentage as that percentage of proxies who admitted that the homicide victims' households had guns.  This would indicate that households experiencing homicide are no more likely than average households to have guns, completely invalidating all the researchers' claims.

Honest researchers who weren't blinded by their emotions would have known that something was wrong with their study when they found that renting (vice owning) a residence resulted in a higher risk of homicide than gun ownership gave, and that having "controlled security access" systems (such as electonic or walls and guards) resulted in a greater risk of homicide than gun ownership did.  People in high risk of homicide are generally in high crime risk situations, so they get security in the form of electonic systems, gated communities, guns, etc.  These things are the results of high homicide risk, not the causes of it.  People who can't afford to buy a home typically can't afford much in the way of security systems, but a handgun is a relatively economical alternative, so renters would tend to be as likely as owners to possess a gun for home protection.

Finally, even though the study wasn't about protective benefits of gun possession, Kellermann can never pass up an opportunity to take a shot at the idea.  So, besides the reference in the abstract, he noted, "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."

This shows either dishonesty or stupidity.  In the first place, the subgroups were only small parts of an already small sample, so there was no basis for even suggesting something about protective benefits.  More importantly, however, how could anyone expect to find such evidence in a sample that was selected to start with on the basis of absense or failure of protection?  They were only examining cases in which someone was killed; so, naturally, they would find no protective benefit amongst the cases they looked at.

Even if some of the victims possessed guns for protection, nobody claims that just having a gun provides protection.  A person who has a gun for defense may very well not be able to get to it (and maybe unlock it and load it) before being attacked or before appearance of an assailant already armed with a gun.

It is foolhardy to try to use a gun for defense against an assailant armed with a gun unless you can get your gun ready and in hand before being confronted by the assailant, or unless the assailant gets sufficiently distracted at some point or you have reason to think the assailant is going to try to kill you immediately regardless of your own actions.  In other words, a person using a gun for protection must constantly evaluate the relative risks of resisting and not resisting.  Most gun owners know this, which is why they have a very good record when it comes to self defense.

See critiques by H. Taylor Buckner [ALT] and Henry Schaffer, PhD [ALT] of the study.