Monday, July 31, 2006

Better state IQ estimates

Intrigued by Tickle's IQ test results by state, I created a regression formula for estimating state IQs based on infant mortality and life expectancy, as these two variables correlate strongly with national IQ (.84 and .85, respectively). It was roughly plausible, with a few exceptions (most notably in that Hawaii came out on top), but not satisfactory.

This time around is better. Taking the regression equations produced by running the numbers in the data table put together by Richard Lynn in Race Differences in Intelligence where he correlates IQ scores with international math and science test scores (pp 173-175) and then adjusting the nominal test score values (by running an IQ of 98 through the regression equation produced by Lynn's numbers) on the international tests to the NAEP math and science tests in the US, I gave it another shot. From the process described above and by applying equal weight to the science and math test scores by state, here are the results (rounded to one decimal place and color-coded by 2004 Presidential election results in the spirit of the IQ hoax):

1. Massachusetts -- 101.5
2. North Dakota -- 101.4
3. Vermont -- 101.2
4. Montana -- 101.1
5. South Dakota -- 101.1
6. New Hampshire -- 101.0
7. Minnesota -- 100.8
8. Wisconsin -- 100.3
9. Wyoming -- 100.2
10. Iowa -- 100.0
11. Idaho -- 99.9
11. Maine -- 99.9
13. Nebraska -- 99.7
13. Virginia -- 99.7
15. Washington -- 99.6
16. Ohio -- 99.6
17. Colorado -- 99.4
18. New Jersey -- 99.3
19. Kansas -- 99.2
20. Oregon -- 99.1
21. Utah -- 99.0
22. Michigan -- 99.0
23. Conneticut -- 98.8
23. Delaware -- 98.8
25. Missouri -- 98.7
26. Pennsylvania -- 98.6
27. Alaska -- 98.6
28. Indiana -- 98.5
29. Kentucky -- 98.3
30. New York -- 97.8
31. Illinois -- 97.8
32. South Carolina -- 97.5
33. North Carolina -- 97.4
34. Maryland -- 97.2
35. Texas -- 97.2
36. Oklahoma -- 96.9
37. Rhode Island -- 96.8
38. West Virginia -- 96.7
39. Tennessee -- 96.6
40. Arkansas -- 96.5
40. Georgia -- 96.5
42. Florida -- 96.1
43. Arizona -- 95.9
44. Nevada -- 95.2
45. Louisiana -- 95.0
46. California -- 94.7
47. New Mexico -- 94.5
48. Hawaii -- 94.4
49. Alabama -- 94.4
50. Mississippi -- 93.3
51. DC -- 88.0

Seems quite reasonable. I used the international tests for children aged fourteen and the NAEP results for eighth graders, so age discrepancies are not an issue. Lynn found that international math results correlate with IQ at .89, although that may be an error (or due to rounding), as running the exact numbers he has entered in the table yields a correlation of .87. He found that international science results correlate with IQ at .81 (replication yields the same). Accounting for attenuation, Lynn actually argues that the correlation is virtually 1.00. So these appear to be pretty accurate estimates.

Notice the states that come closest to realizing the open border crowd's quixotic utopia--Texas (#35), Florida (#42), Arizona (#43), Nevada (#44), California (#46), and New Mexico (#47). Compare them to states like North Dakota (#2), Montana (#4), and South Dakota (#5), where Americans still do the jobs that Americans won't do. Our future looks dumber.

(Human biodiversity)


Anonymous said...

Interesting. Believable, unlike so many pseudo-estimates out there.

Anonymous said...

Hmm, except that the NAEP is in English, and many of the students at the time of testing may not be fluent in English, but will acquire that knowledge later (by the time they are adults -- but when they are no longer included in the NAEP figures). You have no idea what those students IQs are in their native language.

This is similar to the situation that caused a lawsuit years ago in San Francisco, because Spanish speaking students were being placed in Special Education classes after being IQ tested in English and labelled "retarded". IQ testing of elementary students now is done with a non-verbal test for this reason.

This is a pseudoestimate and not at all what this data was intended to be used for.

crush41 said...


Even though I used math and science results, VCU's Professor McDaniel used reading results. Our estimates correlate at more than .96.

Are you arguing that the disadvantage these ESL students face via English comprehension equal affects their performance on reading and math? Students do not take the tests until they've had at least three years of Enlgish instruction--what elementary mathematics question comes close to requiring more than that?

Also, I wonder how this figures into the vast majority of test takers that are native English speakers--states like Mississippi and DC have no fluency issues, yet they score even lower.

Anonymous said...

are the errors normally distributed? If not, you cannot apply the "regression" method without normalising the sum of squares with something else

Audacious Epigone said...

I used the least squares method and it's a linear equation.

Anonymous said...

He probably has enough data that violation of normality probably doesn't matter that much, if at all. IQ tests on the other hand, are generally problematic.

Anonymous said...

By definition average IQ of the population is 100. Based on the population of states below 100 and the population of the states above, this information cannot be correct mathmatically.

Anonymous said...

I suggest the possible reason the "average" IQ is slightly less than the expected 100, is due to the sample population.

The least intelligent people are not likely to have gone anywhere near an IQ test in all of their life. Since they are not sampled when determining the average, the sampled average may be artificially high when compared to the true average.

Audacious Epigone said...


The 100 IQ average is for whites of northwestern European descent, not of the US as a whole. In aggregate, estimates for the US put the mean at around 98. Consequently, the estimates for most states are less than 100.

Watcher said...

IQ tests are generally controversial and disliked because certain minority groups tend to score lower than average on them. Cultural bias or whatever, doesn't matter in this discussion why. However, it's been an observed fact for a long time. Specifically Native American, Hispanics, and African Americans tend to score lower on IQ tests.

So bottom on your list is D.C. What are the demographics of D.C.?

50.7% African American
38.5% White

Bottom state is Mississippi.

37.3% African American (The largest percentage of any state in the US.)
59.1% White

Top state is Massachusetts.

7.8% African American
84.1% White

I think it is safe to conclude that this study can be determined to be unconsciously racist.

Anonymous said...

If you're using only public schools data, you're using biased sampling. Need to use sampling that isn't overweighting lower income groups in states that have charter schools (public schools = lower income). IQ outcomes tend to be correlated to childhood enrichment opportunities and lower income kids have fewer such opportunities. Hence bias.