Short version: if you think children's IQ will regress towards the mean of the population, you have to ask yourself "which population?" Humans? Primates? Mammals? PhD scientists?
The population consisting of (or closely representing) the offspring of the latest several generations of ancestors of the children sampled.
That would definitely make more sense than using the entire human race. I still think it might be interesting to use multiple-child families for one of the studies. There would be confounding factors that would have to be accounted for, such as first-sibling bias and I don't know what else, but there might be advantages too: availability of modern standard tests, rough equivalence of environment, closest genetic matching, etc.