It seems unlikely that the distribution of 100x-1000x impact people is *exactly* the same between your “network” and “community” groups, and if it’s even a little bit biased towards one or the other the groups would wind up very far from having equal average impact per person. I agree it’s not obvious which way such a bias would go. (I do expect the community helps its members have higher impact compared to their personal counterfactuals, but perhaps e.g. people are more likely to join the community if they are disappointed with their current impact levels? Alternatively, maybe everybody else is swamped by the question of which group you put Moskovitz in?) However assuming the multiplier is close to 1 rather than much higher or lower seems unwarranted, and this seems to be a key question on which the rest of your conclusions more or less depend.
It seems unlikely that the distribution of 100x-1000x impact people is *exactly* the same between your “network” and “community” groups, and if it’s even a little bit biased towards one or the other the groups would wind up very far from having equal average impact per person. I agree it’s not obvious which way such a bias would go. (I do expect the community helps its members have higher impact compared to their personal counterfactuals, but perhaps e.g. people are more likely to join the community if they are disappointed with their current impact levels? Alternatively, maybe everybody else is swamped by the question of which group you put Moskovitz in?) However assuming the multiplier is close to 1 rather than much higher or lower seems unwarranted, and this seems to be a key question on which the rest of your conclusions more or less depend.