Wonder how much anchoring effect there is (sample numbers are given in the original question). I note the sample numbers are generated by javascript -- are they testing the impact of those numbers?

Also, I don't see why N-step arguments are invalid; everyone who went for the Nash Equilibrium is wrong (the average is not 3/2 * 0); you just need to predict how many step-thinkers people are :P

(I guessed 6, which is approx 2/3 of 2/3 of the winning answer... so presumably I'm two steps too deep :P)