Reply #58: Well, that's coherent!

[Febble]

I still wouldn't put it at the top of your summary of evidence, though. The one I'm interested in is your experience as a poll-worker. Even if the machines weren't tampered with, the fact that the seals were broken means that surely using them was illegal (not that I am a lawyer). Did you do a chi square on the proportion of Republican votes on the unsealed machines? Presumably the assignment of voter to machine should have been random.

Yes, I take your point re the analysis of the exit poll discrepancy, although the work I did on the data demonstrated that about 50% of the variance was sampling variance. But that, as you imply, would have no net bias (or not after we'd figured out how to get rid of the inbuilt bias in extreme precincts). But the reason I put a high probability on the discrepancy being accountable by bias is not simply that methodological factors were a strong predictor, but because of the complete absence of any correlation between redshift and swing to Bush. If fraud variables were being masked as methodological variables, then it follows that the fraud had variance. And yet that variance was not shared with any benefit to Bush, even after baselining the discrepancy to state mean levels.

That certainly does not mean that there was no fraud in 2004, nor even that it didn't contribute something to the discrepancy (it may have done). But in terms of effect size, it was completely swamped by other factors, which tells me that the probability that it was on a scale of millions is low. And if 2004 had large bias and fraud below the scale of millions, then I find it hard to follow the inference that 2006 had small bias and fraud on scale of millions. You may be right. But I simply don't think it amounts to more than a footnote that the exit polls are consistent with fraud on some scale. And I certainly don't think the generic poll discrepancy is much of a corroboration. I was following them, and the debate about what they meant, pretty closely, and the pundits I read who attempted to translate them into House seats got the prediction about right. And the close races where there were actually polls of specific races again, seemed to get them about right.

So if I were your editor (ha!) I'd put the stuff at the top of your OP in small print in the footnote to an appendix, and get on with taking the people who refused to take the machines with tampered seals to court. And start finding out who tampered with them.