Estimating Scott Alexander's IQ
There is only one good piece of evidence regarding Scott's IQ, which is his SAT score of 1540, according to a self-report. A great verbal score and a mediocre math score seems to be in line with the life evidence he presented in this article, so I feel comfortable taking the estimate at face value. I originally estimated that the mean and SD within the general population was 980 and 213 based on some calculator fiddling - I now know that, based on the NLSY97 (elite sample due to selection), that the average among those who took the SAT was roughly 1000 with an SD of exactly 200. To be fair to him, (invoking rule 8), I will use a mean of 980 and SD of 200. This would correspond with a z-score of 2.8.
Beyond this, he is very well paid - he has about 1000-10000 paid subs who pay him about 10$ and he is rumoured to be paid ~250k to be exclusive to substack. In addition to his normie job, I would estimate that he’s paid roughly 950k per year, which puts him in the top 1%.
Assuming a correlation between IQ and income of .35, and between IQ and SAT of .84 yields an estimate of 138 with a standard error of 8.
It’s difficult to find percentiles for the SAT verbal at this time - the NLSY97 data suggests that is a z-score of 2.7, but even newer editions of the SAT suggest that a perfect reading score corresponds to roughly the ~99.8th percentile (difficult to judge due to the way they publish the percentiles), which is closer to a z-score of 2.9. Taking into a combination of regression to the mean and the fact that the older SAT was harder, I’d peg his verbal ability at 3 SDs above the mean.
g <- rnorm(60000000, mean=0)
iq <- 0.84*g + rnorm(60000000)*sqrt(1-0.84^2)
cs <- 0.35*g + rnorm(60000000)*sqrt(1-0.35^2)
subby1 <- data.frame(iq, cs)
subby1$g = g
subby2 <- subset(subby1, (subby1$iq > 2.7 & subby1$iq < 2.9) & (subby1$cs > 2.25 & subby1$cs < 2.45))
mean(subby2$g)*15
sd(subby2$g)*15