Source? or calculation? The guy giving the figure is an evolutionary psychologist.

If the difference between "best case" and "worst case" ranges from 5--15, then that's the average best-worst difference. So the range is (x-2.5, x+2.5) to (x-7.5, x+7.5). Given it's a "lottery," the average person would expect x and could improve to x+2.5--x+7.5.

But I'm certainly not a psychologist.

Calculation: Kids differ by 5 to 15 IQ points on average. Assuming a symmetric distribution, that means that the smartest kid would be 2.5 to 7.5 IQ points smarter than the average kid, while the dumbest would be the same amount dumber than the average.

Actually, I'm not sure whether "5 to 15" isn't meant to describe the standard deviation, in which case screening lots of embryos might even produce much higher gains.