Hey, I participated in this. I guessed 0 the first round, and IIRC I raised my estimate to 33 in the second round. I suppose the lesson is that I'm a poor judge of human nature.
I tried this experiment a few years ago with a roomful of actuaries, who I suspected would be likely to "iterate" their response a few more times than the average person, and the answer came out at 14.
Even in a room full of game theorists, will the winner be at one. There are a few considerations. One is that one person who slips up either through a misunderstanding or spite will throw off the average. Another consideration is that if you win with zero or one, you will have to split any prize pool with many other players.
Yes, but if you guess higher than 0 (in, as you say, spite) then you only hurt yourself, since everyone else ties for the win. The choice of 0 in this game is a Nash equilibrium - no rational agent will guess otherwise, at least if it assumes all other agents are also rational.
This is true, but I was specifically addressing the case where a "spiteful" agent attempts to guess high and throw off the average, and just pointing out that this would actually hurt no one but the spiteful agent himself.
The results around 22 aren't too surprising: most people will probably quickly figure out that the target cannot possibly be above 66.67, but few will realize that this implies that the target will be below 44.44. If you assume the guesses will be randomly distributed from 0 - 66, you get a target of 22, which isn't a bad guess for most of the times this experiment was performed on the proles.
