Explaining state crime rates

Steve Sailer has an interesting discussion, so I'd like to re-post a similar analysis I once did. I edited the original post to match the selection below (I've learned a little more about statistics and was able to make it more accurate).

Being a complete nerd, I've been fascinated lately by my discovery of Gnumeric, a free spreadsheet program that does multivariate statistics.

I've been messing around with crime data. Using the FBI's list of violent crime rates per 100,000 by state as the dependent variable, I factored in race, diversity, income and toughness on crime:

*Race (four variables). Proportions of a given state that are black, Hispanic, Asian and Native American. If a state is 10 percent Hispanic, for example, that figure is .10.

*Diversity. The ELF for that state, or the odds that two randomly chosen people will be of different ethnicities. Calculated by squaring each race's proportion and adding them together, then subtracting that number from 1. A higher ELF means more diversity.

*Income. The average, adjusted for cost of living.

*Toughness on crime. Average time served in months for a violent offense.

The raw correlations are about what you'd expect -- states with higher proportions of whites, low diversity, high incomes and tough sentences tend to have low violent crime rates. But once you throw them all into one regression equation, the results are interesting.

A disclaimer: I'm pretty good at math but don't have much formal training in statistics. Numbers are on the spreadsheet. (One thing Gnumeric does not do well is export files that Google Spreadsheets can use, so I had to do it in html. E-mail me, and I'll send a .xls file as an attachment.)

Here are the coefficients:

% Black: 1902
% Hispanic: 1213
% Native American: 1941
% Asian/Pacific Islander: 175*
Diversity: -308*
Income: .000021*
Tough on crime: -1.14*

The starred results are statistically insignificant, with P values > .05 and/or t-stats between -1.96 and 1.96.

Two very weird things. One, income has no significant effect.

Second, once one takes into account where diversity is coming from, diversity in and of itself doesn't explain the crime rate either (if anything, higher diversity means lower crime). I think this says something for immigration -- the problem isn't that we're taking in people who are different than we are; the problem is that we're drawing from specific populations that commit crimes (or have kids who commit crimes).

This squares with an international analysis I did once failing to find a correlation between diversity and murder rate. (I also found, oddly enough, that diversity and incarceration rate were related -- having different people side by side makes them distrust each other and likely to throw each other in prison, but it doesn't make them kill each other, assuming my results are right and mean something.)