La Griffe Du Lion performed an analysis similar to the one I suggested in the earlier post on the real explanation for the LA/NY murder rate mismatch. He plotted non-white percentage vs. violent crime rate, which roughly shakes out to the analysis I suggested - with two caveats. First is that an increase in the Asian percentage would decrease the violent crime rate, and second is that an increase in Hispanic admixture would increase the violent crime rate, but not as rapidly as black admixture. Thus "non-white" is an imperfect surrogate for a complete breakdown by the major racial groups (W/H/B/A). As a like-minded mathematical/statistical type, Griffe is aware of such things:

I omitted Hawaii because its nonwhite population is not characteristic of the US. Hawaii is largely Asian of various sorts and has a correspondingly low violent crime rate. The line is least squares best fit. Corr Coeff = 0.84. The outlier is DC. Though not a state, I included DC in the fit.

The inclusion of Asians in the non-white percentage won't make that much difference in most other states (except perhaps in California) because Asians are only 3% or so of the US population. Again, though, it would be preferable (but more work) to do a W/H/B/A decomposition of the data. Ok - without further ado - here's the plot: You might miss DC - it's almost off the charts in the upper right corner, with a horrifying 2500 victimizations (!) per 100000 population per capita. This is much higher than a simple linear extrapolation would predict...meaning that the linear fit is likely reasonable for low non-white percentages only. While there aren't many data points in "intermediate" regions (with, say, 60/40 blacks/whites) [1], one can infer what the middle regions would look like from a study of the extremes (e.g. DC and Idaho) and a bit of mathematics. I suggest you check it out... Source: I'm pretty sure Griffe used the FBI Uniform Crime Reports and US Census. Will update this when I get that info from him. Now - for those of you who aren't familiar with such things - a correlation coefficient of .84 is HUGE, especially in social science. And if you look at the scatterplot, you'll see that a linear fit is well justified by the data, at least for low admixture rates (see here for some famous data sets devised by Tukey et. al. to show the problems involved when a linear fit is not justified). Point being: the explanation for violent crime differentials is far simpler than most pundits (bloggers or not) realize or admit. Demography is statistical destiny, and a high percentage of blacks and Hispanics means a high rate of violent victimization. [1] Just a hypothesis - such places (at least at the neighborhood level if not the state level) tend to rapidly swing towards 100% black once whites start to flee.