Saturday, April 03, 2010

More on the IQ regression to the mean

In the recent post on IQ regression to the mean, reader Levi Johnston asked if there were enough black respondents to calculate IQs for those with two highly educated parents, not one or two. The n for blacks drops from 47 to 16--a small sample, indeed--but the results are a mean of 109.6 for whites and 105.3 for blacks (the difference is not statistically significant). Compared to the earlier results, scores went up a couple points for whites but five points for blacks. The gap in standard deviation (sd) terms shrunk from around half to about three-tenths.

I should note that with the full sample of blacks and whites, the IQ gap based on the GSS vocabulary test in only about 10 points--not the usual 15. This is probably due to the fact that the test taps verbal IQ, a form of intelligence with a smaller racial gap. For the full sample, the black-white difference is only two-thirds of an sd--not the one sd difference that researchers usually observe when looking at general intelligence.


  1. Anonymous10:28 AM

    I thought vocabulary was one of the more heavily "g-loaded" sections of a standard IQ test. No?

    And the more g-loaded the task, the bigger the racial gap.

  2. Doesn't that test only have 10 questions? With that range and imprecision I suspect it can only resolve in the 80-120 range, which is grand for large social questions involving big populations but not so much for individual questions many folks here might have (i.e., what's the probability distribution function of outcomes for my kids given my family background and that of my wife?).


Meta-analysis of clinical trials: Eat walnuts

I am always looking for easy eating choices that are good for you. This new meta-analysis of 26 clinical trials looked to see if walnuts ma...