One interesting thing about the paper is that it seems that the Wikipedia article incorrectly describes the procedure: respondents were not asked to guess whether the majority would agree with their position. They were asked to guess what per cent of other respondents would agree. I think that's a pretty severe difference in the method.