May 2001


circe wrote:

In homes with guns the homicide of a household member is three times more likely to occur than in homes without guns. New England Journal 1993;329.1084-1091.

David Friedman writes:

You may also have noticed that the death rate in hospitals is much higher than in hotels. So if only we abolished the hospitals … .

But the study controlled for a variety of factors such as age, sex, neighbourhood, criminal record, drug use etc etc. If take your hospitals/hotels analogy and control for severity of disease/trauma we would find that hospitals are safer. For example, if you just looked at people suffering severe heart attacks, we would find them more likely to survive if they were in a hospital, rather than a hotel.

In other words, a statistic like that assumes that the causation runs “have a gun in the home, therefore more likely to be killed.” But an alternative, perhaps more plausible, causation is “Be in a situation where you are at risk of being killed, therefore have a gun in the home.”

This alternative is less plausible when you look at the stratified analysis in Kellermann’s study. Gun owners were not at any greater risk of non-gun homicide, only of with-gun homicide. It does not seem plausible that people would only obtain guns to defend against threats from gun-armed enemies and not also arm themselves against an enemy that carries a knife.

Similarly, gun owning households were only at greater risk of homicide from intimates (spouse, lover, family) i.e. those people likely to have access to the household’s gun, and not at greater risk of homicide from outsiders. It does not seem plausible that people would only arm themselves against threats from inside the household and not from outside.

Mary Woods wrote:

Kellermann also used a control case study, limiting his cases to the following criteria; “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.” ( NEMJ vol.329, no.15, pg. 1084). Right there, that raised a red flag in my mind. If the case studies included deaths occurring three months after an assault, one has to question the validity of those cases. The questions that come to mind were; what were the cause of death for those who were assaulted? Were the deaths related to the initial assault or were they from other causes? These question were never answered in this article or any other that Dr. Kellermann published in the New England Journal of Medicine or other medical journals that he has written for.

The answer to the question is in the quotation from the paper. If the death was not related to the initial assault, it would not have been ruled a homicide.

David Friedman writes:

I have no strong opinion on the Kellerman study at this point, not having made a sufficient effort to untangle the arguments by both sides.

That doesn’t seem to stop most people who post about it.

Incidently you can get the study (in pdf) and all of the data here

Most of the criticism of Kellermann’s study available on the web is of truly dreadful quality.

The best available is by Gary Kleck: link

Steve Kangas has written a reasonable summary of the contrary position: link

But I think you are misreading the language you quote. There are two sentences, each of which describes a sufficient condition for including a death. The first is that it was ruled a homicide. That covers all homicides, however long it takes the victim to die; the words are “any death ruled a homicide.” The second is that assault related injuries are included if death followed within three months. So those are cases that were not ruled homicides.

You are mistaken. The first sentence gives the general rule, the second clarifies the definition of homicide used. (That is, the death had to occur within three months for it to be counted as a homicide.) If you remain unconvinced, I suggest looking at the context of those two sentences. The study is clearly about homicides, not “homicides plus some other deaths”