From an econometric perspective evaluating these kinds of things is fun. I would guess that any productivity gains would be concentrated in the most difficult-

From an econometric perspective evaluating these kinds of things is fun. I would guess that any productivity gains would be concentrated in the most difficult-to-answer questions, that you can otherwise spend hours / days trying to debug or understand. "20 hours for 5% of questions" is probably a more accurate characterization, although it would work out the same.