that would certainly be true if IQ were independently and identically distributed (iid), but the point here is that IQ is hereditable (not independent nor uniformly distributed) and so should show, in the long run, measurable durability in those differences across lineages. it's likely not reversion to the mean (or more to the point, the mechanism is not simply statistical in nature).
You're confusing concepts. Regression to the mean doesn't depend on any IID assumption. Conditional expectations are right in the definition! In the bivariate normal case (height, intelligence, etc), it's a necessary consequence of a correlation coefficient < 1.
that would certainly be true if IQ were independently and identically distributed (iid), but the point here is that IQ is hereditable (not independent nor uniformly distributed) and so should show, in the long run, measurable durability in those differences across lineages. it's likely not reversion to the mean (or more to the point, the mechanism is not simply statistical in nature).