They get a selective advantage of 1.5% for the European lactose tolerance allele: “We estimated the selection coefficient on the derived allele to be 0.015 (95% confidence interval; CI=0.010-0.034) using a method that fits a hidden Markov model to the population allele frequencies as they change over time. ”
That’s way lower than any other result I’ve ever seen, including runs I’ve done myself. I’ve seen estimates a full ten times higher. I don’t get it.