The question is whether that assessment task demand more than trivial cognitive ability. In that case, monetary rewards will do no good, as discussed in this research presentation.

The top quartile from the 1999 paper, at least, looks dominated by “regression to the mean” to the point where it even overcomes the effect of positive illusions. Much more than what I would call a smidgen, at least for the 1999 paper! Or is there a better alternative explanation?

That’s fair; you’re right, I was a bit careless in my example. I wasn’t really making any deep point about depression per se, but simply pointing out that changes in behavior that often get interpreted substantively by researchers may not reflect anything important about the course of depression itself, but are simply a statistical artifact. But I concede that this is viewpoint dependent to some extent; if you’re someone who studies clinical depression (which I’m not), you might well want to separately consider veridical changes in people’s mood from regression to the mean induced by measurement error.

Thanks for the link, I see your point now and I’ve learned something about regression to the mean.

I do think that your last reply viz measurements as stochastic variables changing for any number of immaterial reasons is basically the position of the first commenter and me, but quite different from the position in your blog post where you devote a whole par to explaining regression to the mean for depression in terms of natural history. It’s the attempt to identify the immaterial and unidentifiable that caught our attention. And both the examples you give in the post are repeated measures. Clearly you understand the phenomenon better than I do, but it doesn’t come through in your post for those reasons.