Kellermann AL. and Reay DT. "Protection or Peril? An Analysis of Firearms-Related Deaths in the Home." N Engl J. Med 1986. 314: 1557-60.
methodological and conceptual errors:
- prejudicially truncated data
- non-sequitur logic
- correct methodology described, but not used, by the authors
- repeated the harshly criticized methodology of Rushforth from a decade earlier
- deceptively understated the protective benefits of guns
To suggest that science has proven that defending oneself or one's family with a gun is dangerous, gun prohibitionists often claim: "a gun owner is 43 times more likely to kill a family member than an intruder." This is Kellermann and Reay's flawed risk-benefit ratio for gun ownership,  heavily criticized for its deceptive approach and its non-sequitur logic.    Clouding the public debate, this fallacy is one of the most misused slogans of the anti-self-defense lobby.
The true measure of the protective benefits of guns are the lives saved, the injuries prevented, the medical costs saved, and the property protected -- not the burglar or rapist body count. Since only 0.1% to 0.2% of defensive gun usage involves the death of the criminal,  any study, such as this, that counts criminal deaths as the only measure of the protective benefits of guns will expectedly underestimate the benefits of firearms by a factor of 500 to 1,000.
Interestingly, the authors themselves described, but did not use, the correct methodology. They acknowledged that a true risk-benefit consideration of guns in the home should (but did not in their "calculations") include "cases in which burglars or intruders are wounded or frightened away by the use or display of a firearm [and] cases in which would-be intruders may have purposely avoided a house known to be armed...." 
Kellermann and Reay had repeated the harshly criticized folly of Rushforth  from a decade earlier. In 1976 Bruce-Biggs criticized Rushforth noting that the protective benefits of guns are the lives saved and the property protected, not the burglar body count.  Kellermann and Reay would have done well to heed that simple caveat. Objective analysis, even by their own standards, shows the "more likely to kill a family member than intruder" comparison to be deceptively appealing, though only a specious contrivance.
Caveats about earlier estimates of 1 million protective uses of guns each year  have led Kleck to perform the largest scale, national, and methodologically sound study of the protective uses of guns suggesting between 800,000 and 2.4 million protective uses of guns each year  -- not quite as "intangible" as Kassirer claimed  -- as many as 75 lives protected by a gun for every life lost to a gun, as many as 5 lives protected per minute. Guns not only repel crime, guns deter crime as is shown by repeated National Institute of Justice surveys of criminals.  These are the benefits of guns overlooked by scientists whose politics overshadow their objectivity.
At his presentation to the October 17, 1993 Handgun Epidemic Lowering Program conference, Kellermann emotionally admitted his anti-gun bias, a bias evident in the pattern of Kellermann's "research."
Kellermann AL, Rivara FP, Rushforth NB et al. "Gun ownership as a risk factor for homicide in the home." N Engl J Med. 1993; 329(15): 1084-91.
methodological and conceptual errors:
- used only one logistic regression model to describe multiple socially distinct populations
- psychosocially, economically, and ethnically unrepresentative study populations
- study populations, compared to general population, over-represented serious social dysfunction and financial instability, factors that would expectedly increase risks of homicide
- unrepresentative nature of dysfunctional study populations prevents generalizing results to population at large
- when properly used, an "odds ratio" only estimates relative risk of study and control populations -- misleading because the ratio gives no estimate of actual or baseline risk
- one week after publication of this article, during his presentation to a gun prohibition advocacy group, H.E.L.P. Conference (Chicago, October 18, 1993), the lead author emotionally admitted his anti-gun bias, and similar to Kellermann AL. and Reay DT. "Protection or Peril? An Analysis of Firearms-Related Deaths in the Home." N Engl J. Med 1986. 314: 1557-60.:
- ignored criticisms of 1986 methodology, so, for the second time, repeated the harshly criticized methodology of Rushforth from 1976
- non-sequitur logic
- In 1986, correct methodology described, but never used, by the lead author
- failed to consider the protective benefits of guns
Kellermann and his co-authors have persisted in their discredited methodology. In a 1993 New England Journal of Medicine article,  Kellermann et al. once again attempted to prove that guns in the home are a significant risk.
Both the case studies and control groups in this study were socially and demographically unrepresentative of the areas studied or of the nation as a whole. The groups had exceptionally high incidence of social dysfunction and instability. For example, 52.7% of case subjects had a history of a household member being arrested, 24.8% had alcohol-related problems, 31.3% had a household history of illicit drug abuse, 31.8% had a household member hit or hurt in a family fight, 17.3% had a family member hurt so severely in a family fight that medical attention was required. Both the case studies and control groups in this study had very high incidence of financial instability. For example, both case subject and control heads of household had a median Hollingshead socioeconomic score of 4 (on a scale of 1 to 5 with 1 being the highest level of socioeconomic status). These are factors that would expectedly be associated with higher rates of violence, including homicide. The subjects and controls did not even reflect the racial profile of the studied counties; 62% of the subjects were Black compared with 25% of the overall population of the three studied counties.
The unrepresentative nature of the case and control groups undercut the authors' attempts to generalize from this study to the nation at large. The results cannot even be generalized to the counties studied because both the case and control groups did not even represent the ethnic or socioeconomic diversity of the counties studied. With so many complex variables, the authors should have used multiple logistic regression models, but, with their a priori bias, used only one logistic regression model.
Interestingly, according to the authors' own data, guns were next to last in importance of the "risk factors" studied. Alcohol, living alone, family violence, and renting one's home held more risk than guns according to the authors' calculations, yet the most important risks were barely mentioned in the publicity or the authors' discussion. [See Graph 8: -- "Kellermann's Homicide Odds Ratios"] It appears that the authors were more concerned about generating a headline-grabbing "factoid," exaggerating gun risk, than about accurately or honestly assessing the risks of the dysfunctional populations studied.