Tuesday, May 23, 2006

State SAT Stats   posted by Jason Malloy @ 5/23/2006 09:32:00 AM
Share/Bookmark

A few papers out right now, like Jackson and Rushton, revolving around Frey and Detterman's formulas for converting SAT scores into IQ scores. First, gnxp's own A.A. Beaujean has a paper in the latest Personality and Individual Differences, providing additional evidence that the SAT is a reasonable measure of g, matching it up reliably to another set of undergraduate scores; though it is suggested that the conversion equations need to be improved, which will take additional research samples. Second, the versatile evolutionary psychologist Satoshi Kanazawa provides yet more IQ and the Wealth of Nations-inspired work with his new, in-press paper IQ and the Wealth of States. (PDF here) Chapter 5 of IQ&tWoN set the foundations for the book's larger cross-national comparisons, by first showing that regional, sub-national IQ scores also predicted differences in economic prosperity between regions. For instance, within-city IQ differences between districts/boroughs in New York City and London in the 1930s were highly correlated with intra-city economic differences. Similarly, economic differences between a number of 1930s American cities showed a significant relationship with childhood IQ differences between these cities as well. A study of IQ differences between US states circa 1950 (using large-scale military data) found a correlation of .81 between IQ and state income. Other data is provided for regions of France, Britain and Spain.

Using Frey and Detterman's equations, Kanazawa now attempts to update the 1950s study by comparing converted state SAT scores with state economic indicators. But a simple comparison is complicated by the fact that SATs are not taken by representative samples of the population, but by an upwardly biased group of college-bound high-school graduates. Also the percent of people taking the SAT differs dramatically by state. Kanazawa attempts to mathematically "correct" for this and his success seems mixed. The results go in the predicted direction; for instance the correlation between the converted state IQ scores and the Gross State Product (GSP) per capita is .50, with median family income is .57, and with % in poverty is -.35. These are fairly high associations (higher than the relationship between IQ and individual economic success, in fact), but it's possible that the even higher 1950s state correlation is more accurate. The selection bias of the SAT is already missing at least one large part of the story - as Kanazawa notes, even unconverted state SAT scores do not correlate with % of the population that is black, despite the large, well-confirmed black-white IQ/SAT gap. He doesn't attempt to explain this, but it could mean the number of blacks actually taking the test is fairly similar in each state regardless of the total number of blacks in each state. Perhaps readers have a better understanding of what's going on.

Another anomaly is that the reported state IQ levels themselves are implausible, which is probably a function of both the Detterman equation and the Kanazawa conversion - or to put it another way, the imperfection of converting the SAT into IQ (or SAT as IQ) or to adequately "correct" for the SAT selection bias. For instance, the highest state IQ is New Hampshire with an IQ of 110.3 and the lowest is Mississippi with an amazing 62.7! That's a spread of over 3 standard deviations - almost 50 IQ points. Fully 12 states are reported with IQs lower than 80, and this is with race not playing a detectable role. We get Utah with an IQ of 75.1 and Wisconsin with 78.4! We can compare this against other data to see if it's compatible. For instance Henry Harpending provided Steve Sailer with Project Talent IQ data for 366,000 high-schoolers from 1960. This representative data only has a spread of about 1 standard deviation: from Montana with an IQ of 105 to Alabama with 89. The Encyclopedia of intelligence also provides Wechsler standardization differences by U.S. region. (by lumped states) which gives the modest spread of 101.6 for the Northeast, to 98 for the South (3.6 point difference). Kanazawa's numbers lumped in the same way provide much different results, with regional differences exceeding 1 SD.

Kanazawa is aware of all this and suggests:

. . . while the state IQ estimates do correlate very highly with the macroeconomic performance measures and thus appear to have some validity, it is difficult to take the estimates at the face value. . . Until more accurate estimates of the absolute levels of state IQ appear (derived, for example, from actual IQ tests administered to large, representative samples of state populations), perhaps it is best to treat the current estimates as reflecting the relative standings of states. . . than estimating the absolute levels of state IQs.


Finally, it was worth applying this information to the infamous blue state/red state IQ hoax from 2004, and Kanazawa's data were not kind to the red states. It matters not if we accept his relative or absolute scores, the states stack blue side up. In the link above, Steve Sailer showed only small differences on the NAEP achievement results, but with Kanazawa's SAT aptitude results we find that 8 of the top 10 scoring states were Gore voters in the 2000 election (average blue state IQ = 99.3) and 8 of the bottom 10 scoring states were Bush voters (average red state IQ = 90.2). So, apparently, those ornery hoaxers were on to something after all.