One reader of the WHOIS Task Force’s report sent me a private comment about the statitsical considerations section of that report, pointing out that there is no reason to suppose that the statistics of a self-selected sample can be modeled by a Gaussion or any other random model. This is, of course, true: The fact that the respondents were self-selected can add any systematic bias, and we don’t know what that bias looks like. However, we know that this bias makes the results worse. Thus, error margins derived from a random model underestimate the errors. Put differently: If you couldn’t derive a conclusion assuming the sample is random, you certainly can’t derive it when the results come from a self-selected sample.