Richard Florida and gun deaths January 31, 2011Posted by mwahlstrom in Brottsstatistik / Crime statistics.
In the wake of the recent tragedy in Arizona, Richad Florida and a colleague of his, Charlotta Mellander, enter the discussion on explanations for geographic variation of gun-related deaths in the US. Florida, who has become (in)famous for his conception of the “creative class” as the key to local economic success, now strays into the field of criminology. With not much success. To be fair to Richard and Charlotta, the blog post is clearly not meant as a scientific contribution, and I should add that I sympathise with Florida’s supporting stricter gun control laws, as well as his apparent belief in the detrimental effects of some forms of aggressive rhetoric from the local political establishment. However, when you are a distinguished acacemic you can be expected to do your analyses properly if you want to draw the kinds of conclusions that Florida does.
Florida lists a number of state-level variables and their correlation (or lack of correlation) with “gun deaths”. He and Mellander purports to find that gun related deaths are positively correlated with poverty, high density of working-class jobs, weapons in high-schools and (quite amusingly) the proportion of votes fo McCain. Apart from being negatively related to proportion of “creative class” jobs (surprise, surprise!), gun related deaths are found to be negatively related to e.g., “ states that ban assault weapons (-.45), require trigger locks (-.42), and mandate safe storage requirements for guns (-.48)”. They also find negative correlation with proportions of immigrants and general standards of living. Finally he makes a point of there being neither significant correlation with mental illness and stress levels at the state level, nor with the “proportion of neurotic personalities”. Inequality and unemployment also seem to have little effect at this level.
Firstly, Florida correctly notes that “correlation does not imply causation, but simply points to associations between variables.” This is true, but Florida appears, quite understandably, to want to say something about causation as well. To do this he would need a convincing theory that supports the use of his variables in the first place, and there is very little theorietical justification of this kind.
Secondly, the data is presented as a series of bivariate correlations. This is surely easiest to understand, but were the variables all theoretically motivated, they should be incorporated in a multivariate model. Some of the relationships that appear to be significant could quite possibly be accounted for in terms of other variables, and those seeming to be insignificant may turn out to be significant when the other variables are held constant.
Thirdly, some variables are arguably measured on the wrong level. Surely, it is not plausible that it is the share of the population with mental illness that affects gun deaths; if anything it is mental illness on the part of the person causing the death (her own or someone else’s), i.e. a property on an individual level. If individual level factors are to be taken into account in the analysis, then a multilevel analysis would be more appropriate.
In sum, even though Florida’s conclusions may indeed be correct, they are hardly supported by the data analysis. A review of the literature on the topic would probably have been much more valuable, and more respectful of those who do real research in this general area.