Tuesday, November 04, 2008

Moron density and violence: I haven't been particularly impressed with IQ's power to explain criminality. The effect always seems to be very modest.

One problem is that criminologists never examine macrolevel relationships. I cannot recall a single published study that assesses the relationship across geographical units. (I'm sure they are out there somewhere--I just haven't seen one).

I used Audacious' estimates of state-level IQs and correlated this with homicide rates averaged over the years 1999-2005. (Homicide is a rare event--criminologists routinely aggregate over several years).

The size of the Pearson correlation? It's -.80!!! That means that 64% of state-level homicide rates is explained by average differences in IQ. Even percent black explains only half the variance.

For example, New Hampshire has a mean IQ of 101, and 0.6 homicides per 100k population per year. By contrast, Louisiana's IQ is 95, and the homicide rate is 9.7.

So why are these results so much stronger than those found in individual-level studies? Well, macrolevel studies typically reduce measurement error, thus strengthening the correlation. In addition, studies typically focus on minor offenses, like delinquency, and it might be that IQ plays a more important role in serious crime.

According to research, most violence is mutual combat, so it won't usually happen unless two idiots cross paths. The higher the density of morons, the more frequent the occasions where conflict escalates into violence. Plus, a community of dolts might generate of culture of aggressiveness which would exacerbate the problem.

Unfortunately, mass immigration from low IQ countries is increasing moron density, so in the long run we can expect a more criminal society.

6 comments:

  1. I think the proper term is "ecological correlation", is it not?

    I'm not just being pedantic: High ecological IQ might have _very_ strong ecological effects visible at the level of ecological correlations.

    ReplyDelete
  2. I think you have to be careful with using State level data. There is no way from the data to know if the murders in New Hampshire are being committed but retards but Louisiana murders are geniuses. Maybe the reason that authors like Herrnstein and Murray, concentrate on low level crimes because they are more common. You would need a lot of subjects to show a relationship with something as rare as murder. H and M do show a clear relationship between petty criminal activity and IQ though. I read a reanalysis of Herrnstein and Murray’s data set that questioned their conclusion in black: arguing that the IQ to crime relationship is reversed in black (e.g., smarter blacks commit more crime). There are some questionable details but it is a pretty interesting hypothesis.

    ReplyDelete
  3. State level data is fine if you are a policy maker responsible for ecological effects like crime rate percapita. Ecological correlations are often the first place epidemiologists look because ecological phenomena are what they are interested in affecting -- and it helps shape the probability distribution of hypotheses.

    Idiots who jump on ecological correlations as "invalid" simply because one cannot narrow down the probability distribution of hypotheses as much as one can with double-blind controlled studies with resolution to the individual are worse than pedants.

    ReplyDelete
  4. Ron,

    Impressive work, as usual. Similarly, incarceration rates and IQ maintain a firm relationship at the state level. The correlation is strong (+.65), but the relationship with homicide is even more robust.

    ReplyDelete
  5. State level data are fine for generating hypotheses. I just want something more before I make a decision. Anyway, you do not need a randomized study to look for an IQ and homicide connection. A matched case control study with cases drawn from prisons and controls from the national survey of youth would work for me.

    ReplyDelete
  6. Actually, any data is more valuable if there are prior hypotheses, aka "prejudice" to be in/validated. Another way of describing this prejudice is as a Bayesian Prior distribution. This value of observation in the presence of "prejudice", "bigotry", etc. is due to the familiar "data dredging" complaint against people who cite correlations: If you look at enough correlations without prejudice, the strongest ones that appear are likely due to chance rather than being a reliable guide in forming hypotheses. Of course, the most ridiculous of such complaints are those that include complaints of "prejudice" along with complaints of "data dredging" -- but it is illustrative of the imbecility of those who oppose the reliance on data that we often see these two complaints lodged against the same researchers.

    ReplyDelete

Are gun owners mentally ill?

  Some anti-gun people think owning a gun is a sign of some kind of mental abnormality. According to General Social Survey data, gun owners ...