[cuchulainn]

I've noticed over the years that many of the studies (*) that "find" that guns increase risk to owners use this method:

1) They compare deaths associated with guns to deaths not associated with guns.

2) But they make conclusions about those who own guns versus those who do not.

Those are separate comparisons. To me, it looks like they make declarations about the differences between apples and oranges by studying the differences between peaches and bananas.

Sure, they're both "fruit comparisons" but...

Is this method just hokum, or is there some sort of valid statistical analysis going on here?

How do you translate from +(gun death) vs -(gun death) to +(gun owned) vs -(gun owned)?

(*) for example Arthur L. Kellermann, MD, MPH and others, Suicide in the Home in Relation to Gun Ownership, The New England Journal of Medicine, Vol. 327, No. 7, August 13, 1992, pp. 467-472 (incidentally, this study counted as "gun associated" non-gun suicides where a gun was in the home, but that's a separate flaw).

[themic]

remember, more accidental gun deaths happen to people who own or handle guns than to those who don't.

seriously, there's a wide range of statistical analysis. one doesn't expect regular news media to aim for chi-square tests or statistical significance; one does expect higher research institutions to go to that level. The Brookings Institute and CATO are good examples.

Check out this finer points of math, both for and against gun ownership, in this report. Caution, the air of science is sometimes a cover for an underlying agenda. Sometimes, it's not. You make the call. Either way, towards the end when other dudes start issuing rebuttal papers, the math gets kindof fun.

http://www.brook.edu/press/books/chapter_1/evaluatinggunpolicy.pdf

[earl_simmons]

I'm no statistician, but the papers I've read have been retrospective case-control studies. The researcher looks at a study population of people who have undergone the outcome of interest (like death). The researcher then looks at the histories of the subjects to determine if a statistically significant association can be found between gun ownership and the outcome. The researcher should ensure that the subjects are equally susceptible to the outcome, and he also must control for potential confounders that could explain any apparent association. I don't think one can determine if any particular study is valid without reading it carefully. Any study can be described in a simplistic manner that makes it appear invalid or frivolous. And no studies I've read in this area have asserted a causal relationship between gun ownership and undesirable outcomes.

[Glock_PhD]

As anyone who has ever done any statistical analysis would know you can make numbers say a lot of things but without the proper controls (comparisons) in place they mean nothing. The other problem and often the largest is that these are more suited to association studies rather than cause and effect. For one easy example there have been studies done just to show the weakness of association test that show certain zodiac signs live longer. These were perfectly done tests yet, have no merit in their final conclusion. You can do this with anything and it is often done with guns. You have to remember cause and effect. The example that was given where all suicides were counted did the guns have an effect? I don't know, were there plants in the house, was this an important factor. It can easily be seen how you can take two unrelated things and associate them and sometimes you even get something that is statisticaly significant.

[Art Eatman]

To illustrate the fallacy of the kellerman "43:1" deal: He assumed that the measure of success for a homeowner's defense against a burglar was the death of the burglar. So, sure, it may well be that for every dead burglar there are 43 dead residents of homes with guns.

So what? When you lump all gun deaths together--murder, accident, suicide--and then compare this with the small percentage of Kellerman's definition of "success" against burglars, you wind up with this meaningless 43:1 fatality ratio.

I've always said that Kellerman takes a rather bloodthirsty approach. If I'm lying when I yell through a door that I have a gun, and the burglar leaves, I call it successful self-defense. Same for display of a gun, a warning shot, or a wounded-but-alive burglar.

Journals of psychology for over a half-century have had articles claiming that if one is serious about suicide, the absence of a particular method doesn't matter. While over half of deaths by firearm are suicide, "magicking" away all firearms would not notably affect the rate.

Art

[earl_simmons]

Kellerman would respond that while death is required for the positive outcome (successful deterence of a criminal), death is also required for the negative outcome (death of the homeowner). Instances in which a homeowner deterred a criminal by displaying a firearm or delivering non-fatal gunshot wounds are not considered occurences of the positive outcome; however, instances in which homeowners suffer nonfatal wounds (i.e., crimes involving the threatened use of a firearm or the use of firearms to administer non-fatal wounds, failed suicide attempts, non-fatal unintentional gunshot wounds) are not considered occurences of the negative outcome. This doesn't necessarily mean the study is not flawed, but it would not appear that there is a methodological problem here.