Sunday, July 23, 2006

Race matters

At least when it comes to crime. Indeed, it matters a great deal more than a host of other social statistics that putatively lead to criminality. In the abstract to the second edition of The Color of Crime, the American Renaissance wrote that "The single best indicator of violent crime levels in an area is the percentage of the population that is black and Hispanic." Then, using single variable regression, they compared correlations by state including DC (see pages 11 and 12).

Wanting to replicate for verification and introduce a few other factors that conventional wisdom often asserts as having a causal relationship to crime, I took the latest numbers available (the Kaiser Foundation has a fabulous, user-friendly site for state-by-state comparisons for several factors) and looked at the relationship of violent crime (murder, non-negligent manslaughter, forcible rapes, robberies, and aggravated assaults) by state and the following factors (with correlations; "inverse" indicates an increase in the factor reduces violent crime; statistically significant unless otherwise stated):

% of population in poverty: .399
% of population with a bahelor's degree: .289 (inverse)
State gini coefficient: .627
% of population owning at least one gun: .060 (inverse, not statistically significant)
% of population with a high school diploma: .388 (inverse)
% of population unemployed: .309
% of population black or Hispanic: .800

Very little variance with what Amren came up with for the measures they used. But most telling are the results of a multivariate regression to determine the combined effects of the first six factors listed (all but the racial categorization). Surely entering all of these factors together will be more predictive than simply knowing what racial box residents check on their Census forms! Actually, they're not. Including all of them yields an r of .689. Thus, knowing how many people in a state are black and Hispanic reveals more about the amount of violent crime than does knowing the number of people in poverty, holding a bachelor's degree, having graduated from high school, owning a firearm, who are unemployed, and the level of income inequality combined.

When the percentage of the population that is black or Hispanic is included and each factor is controlled for by the other variables listed, in addition to the percentage of the population that is black or Hispanic (p-value of .00000109), only the percentage of the population that has completed high school retains statistically significance (p-value of .025), but interestingly becomes positively related to the incidence of violent crime. Racial composition remains just as strong a predictor of crime as it does when the other factors aren't included. Not surprinsingly, the r value (correlation) only increases to .836 with the addition of the other six factors.

As people of European ancestry continue to hemorrhage away their numerical dominance, the US will become an increasingly violent place (or one that has an ever-growing prison population, the seemingly most effective way to cut down on crime being to throw people in jail). The detrimental effects of a browner population will be partially offset by an increase in Asians who, despite suffering higher poverty rates than whites and being a numerically less powerful ethnic special interest group than blacks or Hispanics, are far less likely to commit violent crimes than any other racial group (only a quarter as likely as whites). Every one percent increase in population represented by Asians reduces the number of violent crimes per 100,000 people by 25, a per capita effect stronger than that of blacks, whites, or Hispanics.

What should realists do with this information? Reject the sea of crap about the differences in geographical locations that fails to take into account each location's racial and ethnic composition. Vociferously argue for an end to unfettered, largely Hispanic, illegal immigration and the instituting of a merit immigration system that reduces the number of legal immigrants to a couple hundred thousand annually while insuring that those accepted will increase the quality of life for the current US citizenry.

Paraenthetically, genuine liberals concerned with the perpetually increasing wealth disparities in the US can argue against open borders by pointing to the deleterious effect it has on the wages of the least skilled natives and the strong relationship between housing unaffordability and the proportion of a location's foreign-born population. To insulate themselves from the charges of Hispanophobia and other ad hominen that scare whites into self-immolation, they should point out the moderately strong relationship between crime and wealth inequality referenced above.

(Human biodiversity)


JSBolton said...

That's remarkable that the environmental variables add so little to the prediction made by race.
In order for civilization not to be overwhelmed, we do indeed need to change the sign on immigration's effect from negative to positive.
It is important that scholars and officials become loyal to the continuity of civilization, rather than indulge their will to power, fancying themselves 'change agents'; but with every desired change instigated by them, to be anti-civilizational.

Buffy said...

I think your analysis is probably sound. But since the government agencies that control immigration preferentially hire members of these higher crime minorities to be employees, and have been doing that since at least 1970, it's not likely you're going to change the way immigration policies are implemented.

People contemplating a place to locate their families, or to retire, should consider your analysis. Government officials are happy to sacrifice you or your families to political correctness, but if people have the facts in their hands they can decide for themselves.

mping said...

Interesting analysis. If you don't mind, I will take the other side of the argument just to see how solid your analysis is.

I noticed that you don't look at the number of police per capita which I would think would make a big difference (and differ from state to state). I would also think that most violent crime is committed by young males. Does the number of young males (16-30) differ between states? I would also think if the young males are unmarried or unemployed this would make them more likely to commit these crimes (but probably tough to find that data differentiated by state). I would also guess that violent crimes are more likely to be committed by men who did not have a father figure in their life. Does the rate of single motherhood differ between states? Or what about drug abuse? I would think that those that commit violent crime are more likely to abuse drugs. Does this differ between states?

It has been a while since I looked at multivariate analysis. Can you tell me, if you forget to include an important variable in the equations does it make the whole analysis invalid? I guess what I am wondering is if multivariate analysis is only as good as the one variable that you forget to include.

I also wonder, with the Hispanic immigrants, if they stayed in Mexico would they be just as likely to commit the crimes? If so, is Mexico seeing a drop in crime as we are taking their young male trouble makers off of their hands? Or is there more to it, where coming to America changes their social structure makes them more likely to commit crimes. Does the lack of family and friends and being a minority in the community that can't speak English make the difference?

If I remember correctly, New York City has a larger that average percentage of Hispanics and Blacks but also have a lower than average crime rate. Can you confirm or deny this?

If it is lower, how do you explain this? Might it be that having a good police department is more important than the racial makeup of the population? Of course you would probably make the argument that crime would be that much lower in NYC if it were all white and I wouldn't have anything to come back with you at.

Antero Kälvä said...

If you look at county-level crime numbers and control for some relevant factors, you'll find that the race effect is quite small. % black population and % foreign born population have some effect on violent crime, % hispanic doesn't have any effect at all.

Some predictors of crime at the county level that have a greater effect than race: number of single mom households, number of single male households, number of households with income >75k/year, log of population density, rural population, % white population below poverty level

crush41 said...


I'll look for stats on several of the factors you raise. Obviously some are going to be close proxies for race, like single-mother households. But SES is a rough proxy for race as well, but it is a considerably weaker predictor than race itself.

Mexico's murders-per-capita number is three times higher than that of the US, according to the nationmaster site. But rapes in the US are nearly three times as high as those in Mexico. I'm not sure how reliable international statistics are on things like crime.

As for NYC, you'll have to ask Giuliani. It is one of the safest urban areas in the country.


Can you point me to what you're referencing? I'd like to see the full list of variables. Why would the effect at the state level vary so widely from the county level?

Antero Kälvä said...

I got the county-level data from this website by the Census bureau.

I did a regression with the variables you listed, excluding the gini coefficient and gun ownership which are not available, and adding % of single mom households.
Results for violent crime:
- % black & hispanic, poverty and unemployment are not significant
- % population with hs diploma is significant, but effect is small compared to single moms
- % population with bachelor's degree is significant, but effect is positive (and small)
- R2 = 0.29

Without % single mom households:
- % black & hispanic is significant and gets the highest coefficient
- poverty is significant
- high school diploma not significant
- bachelor's degree is significant, but again has a positive coeff
- R2 is only 0.2 -- single motherhood really has predictive power, it's not just a proxy for poverty and black/hispanic population

Anonymous said...

Single motherhood is no doubt strongly correlated to crime--it is also strongly correlated to blackness and slightly less so to hispanicness.

So does black ethnicity "cause" high rates of single parentage, or does high rate of single parentage "cause" black ethnicity? Pretty ridiculous when you look at it that way. Don't put the cart before the horse.

Looking at the county level without specifying which counties you're looking at doesn't make any sense. Rural vs. urban counties? County populations?
Census dates: 2000 vs 1990 etc.---the demographic change between the dates has been significant.

Antero, nobody can evaluate what you're claiming unless you come clean with your data.

Antero Kälvä said...


The regressions I discussed above include % black and hispanic population as an independent variable in the same equation with % single mom households. When the effect of black & hispanic population is insignificant in the same regression equation with single motherhood, it pretty strongly suggests that race in itself isn't the ultimate cause. But there's still the question of black/hispanic culture causing single motherhood as you point out.

The results above are based on 2726 counties. The data is from the year 2000, except for poverty and unemployment numbers which are from 99. There are 3139 counties in the data set I collected from the USA Counties site, apparently two less than the total number. I'll mail the file to anyone who wants it.

I'm thinking of putting some of these county data analyses together in the form of a blog. It's a pretty interesting set of data.

Anonymous said...

You're to be congratulated for your efforts, antero. Thanks for detailing your data source more completely.

I wonder if county statistics differ significantly from city or state statistics?

I noticed at the census website you linked that state data often contained data that was not contained in the county data, so I wonder how much data is lost by depending upon that particular source.

Crush41 has touched on a significant, though controversial, aspect of society. To give it the respect it's due, it might be best to run the data from several sources.

crush41 said...


You need a name!


Thanks for the link. I've run the numbers at the state level (the site allows that quite easily) and they apparently differ from what you've found at the county level. I'll post on it shortly and offer some of my speculation as to why that may be.

Anonymous said...

Really amazing! Useful information. All the best.