[Homaniac]

Yes, but over sampling (over powering) a study causes issues as well due to increasing the chance of Type I Error (i.e. detecting false positives).

[tertius]

I agree and that's not really what I'm saying.

I think if you want to generalize (which is very rarely done in science) then the sample size has to be representative of your (general) population.

Now we know that media (i.e. NYT) likes to take the extreme specificity that makes studies significant and apply it broadly. That's really where my contention comes from.