But when we disaggegate down to 17 individual panels, the N becomes a little thin, and small sampling variance kicks in – also, per Wisdom of Crowds, we didn’t have the luxury of asking the type of questions that are needed to get a “good’ answer with that method; e.g. these were people responding about themselves, not ‘betting’ on how others might respond to a given concept, idea, statement, or prediction.

But, per Joel’s comments, yes we are looking at the effects of a multitude of panel charactertistics on outcomes – it is a high dimensional problem, but a great challenge. Thanks for your note.