I had recently discovered that that people who took both the old GRE and the old SAT (circa 1990) had a verbal old SAT distribution of:

mean: 510.1 (SD = 107.7).

Because the SAT distributions if all 17-year-old Americans had taken the SAT would have been 376 (SD = 102), this implied GRE takers had a mean IQ of 121 (SD = 15.4) (U.S. norms).

Unfortunately this led to ridiculous results like the 95th percentile of the GRE population being at the 99.9 percentile of the general U.S. population. Clearly college admission tests are not normally distributed so we must look at the observed distribution, not the theoretical normal one.

And so I looked at Ron Hoeflin’s Omni sample norming of the Mega Test, where seven Mega Test takers reported scores on the old LSAT.

Then by pairing the Mega and LSAT scores by equal rank in the sample, we get the following equivalencies.

Now to put these numbers in perspective, a 630 was 92nd percentile (among LSAT takers) in 1960 and 725 was the 98th percentile, in 1974 (source: Law School: Legal Education in America from the 1850s to the 1980s, by Robert Bocking Stevens).

From here we might conclude that the 95th percentile of the LSAT population was around 700, which equates to an IQ around 138 (99.5 percentile in the general population). (Note: I’m being 8 IQ points more generous than Mensa, which equates the top 5% of the LSAT population with the top 2% of the U.S. population and thus IQ 130).

This kind of makes sense, because if we assume that roughly 10% of Americans pursue post-grad degrees, and almost 100% of the very brightest do so, then the top 5% of those taking post-grad admission tests should be the top 5%/10 = top 0.5% of Americans as a whole.