His conclusion is that on a county-by-county level in the US, race is the strongest predictor of homicide rates, with rates of single-motherhood being next strongest. Poverty is a reasonable predictor, but weaker than these and several others.

I read this closely a couple years ago, but only skimmed it now to remind myself of the conclusions. I'd be interested to hear peoples' thoughts on whether his statistical techniques are appropriately applied. I recall it was was convincing to me at the time.