Saturday, July 29, 2006

Race matters (con't)

Previously I posted on the telling fact that the percentage of a state's population that is black and Hispanic reveals more about that state's rate of violent crime than a host of other variables combined. Mping, who runs the Fat Knowledge blog, threw out some devil's advocate factors to consider. I haven't been able to find anything on police per capita (please let me know if you are aware of available data). Some factors suggested that I did take a look at and their respective correlations with a state's rate of violent crime:

% of population using illicit drugs in the last month: .130 (not significant)
% of pop between ages 18-24: .185 (not significant)
Median age: .173 (inverse) (not significant)
% of households containing one male, no spouse, and no children: .541
% of single-mother households: .699

The first three are all outside reliability at 90% confidence, but the loose correlations they have with violent crime trend in the direction that conventional wisdom asserts they should (more drug use and more 18-24 year-olds means more crime, the older the state's population the lower its criminality).

Both measures of household composition are statistically significant, again trending in the expected direction. When men assume more responsibility, they tend to get into less trouble. This is an argument for policies that make housing prices affordable enough for men to raise families and make a monthly mortgage payment. Our immigration policies appear to do the opposite, however.

The lack of a father figure removes a traditional disciplinary figure from the equation. Single mothers, around 60% of whom are impoverished, must pull double-duty. When a male child reaches early adolescence, he is physically unchallenged by mom. The staggering increase in out-of-wedlock births over the last forty years is yet another tragic legacy of the most socially disastrous decade of the US' existence, the sixties. Propensity to commit violent crime is only one of several sad circumstances for children in this situation. But with an r of .699, it is a very real relationship.

Still, it's not as strong a predictor as the percentage of the population that is either black or Hispanic (r = .800). Antero Kalva tells me that at the county level the percentage of single-mother households trumps the percentage of the population that is black or Hispanic in predicting violent crime, rendering the latter statistically insignificant when single-mothers are included.

Unfortunately I'm not a prospective PhD student working on a dissertation so the thought of entering data for 3,140 counties is daunting. At the state level, controlling for race eliminates the statistical significance of single-mother households. Moreover, the percentage of the population that is black or Hispanic is a more powerful predictor of violent crime (recall r = .800) than the following factors combined: % single-mother households, % single-male only households, % of the population using illicit drugs in the last month, % of the population between 18-24, median age, poverty rate, % of the population with a bachelor's degree, % of the population having completed high school, the gini coefficient, the rate of gun ownership, and the rate of unemployment (combined r = .784).

The correlation between single-mother households and the percent of the population that is black or Hispanic is .829, affirming that the two serve as proxies for one another, as an anonymous commenter suggested (especially for blacks, where the correlation is an even higher .875).

Why would the 'causal' factor be single-mother households at the county level and race at the state level? The sheer number of counties that have almost no black or Hispanic population strikes me as the most plausible explanation. Since Antero presumably did not adjust for county population size, the bulk of the data examined will have been from Bush Country. As there is almost no black or Hispanic population to speak of in most of these counties, other non-racial factors move in to fill the gap for the whites living in them. Well, excluding race, the percentage of single-mother homes is the only factor that retains statistical significance when the others previously listed are included. So in places like, say, all of Kansas west of Topeka, single-mother households are probably the critical factor.

I wonder what the relationship looks like if only counties where the population of blacks and Hispanics comprise at least 5% of the population are included. Antero, if you are running your numbers on SPSS, this should be easy to find out (as I once again tell myself I need to get the full version so I can enjoy more statistical alacrity than Excel allows for). This would address the issue of race without artificially crippling it as an explanation by including data from counties from which it cannot be a factor because it doesn't exist (since the pertinent policy question has to do with the US' shifting racial composition).

An interesting aside: perusing through The Color of Crime, the criminality of Native Americans stuck out (see p9). Including the percentage of Native Americans with blacks and Hispanics and then correlating with violent crime, the r creeps up to .810, a seemingly trivial but actually rather substantial increase when it is considered that only 1% of the US population is Native American.

++Addition++Antero sent me his county level data and, as he said, the percentage of single mothers is a better predictor of violent crime than racial data.

A couple of interesting things: the correlations between crime and a host of factors, including race and single mother households, are much weaker on the county level than on the state level (% single mothers, for example, for the 2,702 counties with complete racial and criminal data sets, correlates with violent crime at .538 compared to .699 statewide; % black and Hispanic correlates with violent crime at .428 at the county level and .800 statewide). Thus, even as the apparently strongest single predictor of violent crime, the percentage of single mother households still only 'explains' less than 30% of a county's violent crime rate. In contrast, knowing the full racial composition of a state explains 70% of that state's violent crime rate.

Perplexing. I would expect the relationships to become progressively weaker, not stronger, as the source of data broadened. Maybe there are dilution effects on a local level that become neutralized on a larger level (like the effect of immigration on wages in local economies compared to its effect on the national economy), although I can't readily conceive what the causation would be. Also, there are many missing data on the county level. For example, New York (the highest population density in the nation), Bronx, Richmond, Cook (Illinois), and Queen counties all do not have violent crime data, and they represent a considerable number of people (twelve million) in rough areas of the country that are not being factored into the county level analysis (but presumably are included in the FBI statistics).

In any case, more reason to lament the breakup of the nuclear family and re-stigmatize illegitimacy.

(Human biodiversity)


Steve Sailer said...

I guess the question is whether there are large black populations anywhere that don't have lots of single mothers? There are small black populations in a few states, most notably Hawaii, where blacks have middle class family structures and low crime rates. Still, most of the blacks in Hawaii are there due, in some fashion, to the U.S. military, which won't take very low IQ applicants or those with a serious criminal record, and typically wants high school graduates.

Antero Kälvä said...

There were approximately 731 counties with >95% white population in 2000. Excluding those or weighting counties by population doesn't make % black & hispanic population significant when % single-mom households is included.

Outside of murder, I haven't found any subset of counties or set of variables that would make the race effect larger than the single-mom effect.

It may be a good idea to include population density or % rural/urban population in the state level analysis. It's a big factor at the county level wrt any type of crime, overall it's about as important as single motherhood.

Petr Bjork said...

This is fascinating. I believe you are correct when you say that race is an independent risk factor for violent crime. Single motherhood is likely also an independent risk factor, although race and single motherhood contain a high level of interaction.

It occurs to me that it is politically repugnant to many people to admit the obvious that there are strong racial differences in violent crime rates. They seem to fear that this would call into question the huge apparatus of the social welfare and affirmative action systems.

I wonder if there is any relationship between population IQ and violent crime rates? This seems to be a question that is studiously ignored in the intellectual community.

crush41 said...

There are only 28,000 blacks in Hawaii. 45,000 of Hawaii's population is in the military. Including family, it appears that just about all of Hawaii's blacks are, as Steve said, involved in the military and therefore not representative. Unfortunately I'm not aware of any sites that show single motherhood rates by race by state.


Will you send me your data? It's quite perplexing that there would exist such variance at the county and state levels. I'm away from home for a few days but I'll include population density when I return to the hovel (data set is there).


It's hard to come by reliable estimates of crime rates by country. For example, it strains my credulity to believe that you are over 26 times more likely to be assaulted in Iceland than you are in Costa Rica.

As for IQ and crime, the percentage of a state that is black or Hispanic and the percentage of families that are headed by single mothers are both inversely correlated with IQ. Using Richard Lynn's correlation results with math and science tests and IQ by country I'm going to formulate more estimation of IQ by state and then see how those estimates correlate with violent crime.

faq said...

That is definietly odd. I wonder if you up it to counties that are 20% or more black and Hispanic. That way the counties will be more representative of national demographics.

Also entertaining autobiography. Was that really for school?

Anonymous said...

Single parenthood is clearly correlated with race, and simply assuming that all the correlation between crime and single motherhood should be subtracted from the correlation with race is not a correct inference. For example, if low-IQ is correlated with not having a job, which is correlated with low earnings, would you subtract the effect of not having a job from the low-IQ/low-earnings correlation?

The bottom line is that race is one of the most important factors in explaining crime, and so solutions need to acknowledge this (eg, don't be afraid of disparate impact in policing). As long as we pretend this is not so to avoid offense, the problem will persist or get worse.

crush41 said...


I agree with the thrust of what you are saying. But if single motherhood is the causal agent, and blacks are only more likely to head single mother households, an environmental solution becomes at least conceivably more tenable than if race is a stronger predictor than any mutable factor (like drug use, educational attainment, etc).


Thank you for sending the excel file. I look forward to exploring it later this week.


Yeah, it was for a mawkish HR class I had to take from a washed up post-modern feminist who was clearly out of place in a classroom full of unscrupulous pragmatics. Glad you enjoyed it!

pdn said...

Re your comments on assaults in Iceland vs Costa Rica. I have lived in Costa Rica for the past three years and am married to a Costa Rican. My guess is that very few assaults are reported to the police.
The police are widely considered to be inefficent and corrupt.The middle and upper classes depend on private security firms to protect their homes. The poor regard the police with indifference or contempt.

Anonymous said...

Why examine the correlation of "variables" when you can just look at statistics of crime perpetrators, which do still contain data on race/ethnicity? After all, we're not like France...yet.

crush41 said...


Yes, my uncle worked for the US embassy in Guatemala for several years and confirms what you say. In fact, as a mid-level administrator, he had two full-time personal security providers. Comparing international crime statistics reported by the governments of various countries is only as transparent and honest as the government reporting them.


The relationship between ethnicity and several other variables do, of course, exist either way, but if there are other factors that serve as stronger predictors than race, that makes the criminality outcomes appear more environmentally caused.

JSBolton said...

County level data allows the variation in law enforcement severity to acquire great influence.
You might have to use murder rates, since 20 counties out in the country might be 20 times more likely to arrest a shoplifter than one urban one.

JSBolton said...

One indication of the differential severity of law enforcement and length of sentences between jurisdictions such as counties, is the comparison between groups' crime rates and their representation in prison populations nationally.
When the group crime rates for the more imprisonable offenses are 9X the majority for blacks, and ~3X, for hispanics; the national prison population totals should be 8% majority, over 65% black, over 20% hispanic, over 3% indian, and 2% asian.
The majority % is actually much larger than 8%, everal times larger; which indicates greater severity in jurisdictions as the white % increases.

crush41 said...


Help a simpleton with your numbers. Since blacks are 9x as likely to commit violent crimes as are whites, and there are 6x as many whites as blacks, shouldn't the black prison population only be about 50% more than the white one, in absolute terms?

Perspicacious suggestion on murder rates, as they do correlate more strongly by race than by the percentage of single mothers at the county level.

Antero Kälvä said...

It's not so strange that the correlations increase going from counties to states. That happens when there's enough variation between the states in the independent variables.

But it is really strange that race has so little effect at the county level when controlling for number of single mothers, and the reverse is true at the state level.