Commenter Mug of Pee was mentioning professor James Lee, who was recently interviewed by Spencer Wells and Razib Khan.
The interview describes a study where a genomic formula predicted 11-13% of the variance in educational attainment (highest degree or diploma obtained).
Taking the square root, it implies that known common genetic variants correlate 0.35 with education, however Lee cautions that population stratification can inflate these numbers. He cites the cliché that chopstick use would seem highly genetic if the sample were a mix of Chinese and non-Chinese people, but the heritability would be misleading because it’s not that many genomic variants are causing chopstick use, but rather they’re signaling Chinese ancestry, which in-turn causes chopstick use.
To avoid the problem of population stratification in the study Lee co-authored, they looked at within family data (I guess because siblings all belong to the same sub-population) but found that the effect size of their predictors were 40% smaller. I guess that means instead of common SNPs correlating 0.35 with education, they correlated 0.21, now explaining only 4% of the variance?
I wonder if these numbers are distorted by range restriction because families have less variance than the general population and that’s known to depress correlations.