[WS:] My problem with applying statistical methods to this problem is that homicides are rare events whereas statistical methods depend, ultimately, on the law of large numbers. Small samples are notoriously unreliable, because small changes can produce large differences in sample parameters.
Pooling broader areas together may increase sample sizes, but also brings more variables to be considered, which makes the exercise counterproductive.
Another - and far more serious in my opinion problem is the so called fallacy of composition http://en.wikipedia.org/wiki/Fallacy_of_composition which is inherent in applying statistical methods to this problem. That is, homicides are individual - and rare - acts committed by individuals wheres statistical methods by their nature pertain to groups (samples or populations). The fact that we can describe the properties of the sample (e.g. its ethnic composition or mean age to use a simple example) does not mean that these properties can be applied to explain individual acts in that sample. For example, if we find that a sample (or population) of people convicted of homicide has, say, 60% blacks, 90% males and average age of 21 does not mean that being black,male and 21 years of age explain in any meaningful way why a particular individual in question killed someone. For one thing, a great majority of black males age 21 do not kill anyone, so knowledge of these facts does not get us very far in explaining this person's behavior. Ditto for gun ownership. A great majority of gun owners do not kill anyone, so claiming that gun ownership causes homicide - on the assumption that there is a statistical correlation between the two variables - is basically the fallacy of composition.
I can imagine an experimental design akin to Milgram's obedience studies http://en.wikipedia.org/wiki/Milgram_experiment in which subjects are provoked to act violently, and the experimental variable is the presence or absence of a gun (fake of course!) and the dependent variable is the outcome of the altercation i.e. whether the subject is more likely to act violently toward the provocateur if he/she has access to a gun. I can imagine that such experimentation could shed some valid light on the subject, albeit I sincerely doubt that it would be allowed by the Human Subject Committee, whose approval is required by federal laws.
In any case, I do not have the time to examine the research in this area, but I would be glad if someone more familiar with this topic did so and reported it back to the class.
-- Wojtek
"An anarchist is a neoliberal without money."