Reply #3: it seems to me that you are simply ruling out an obvious possibility

[OnTheOtherHand]

You say here that the paper voters were biased toward Gore in 2000 but not toward Kerry in 2004. That's possible, if (for instance) the Bush campaign did relatively better in absentee voting in 2004 than in 2000, and did relatively the same in provisional/affidavit ballots. But I don't know of a reason to expect that a priori. if anything, I might have expected Kerry to widen the paper gap. (As you know, I would not have used paper ballots in New York as a benchmark in the first place: I would expect provisional ballots to skew strongly but perhaps variably Democratic, and the split of absentee ballots could well depend on campaign mobilization efforts.)

What about the possibility that the exit polls really were more accurate in 2000 than they were in 2004? Why do you seemingly reject that possibility out of hand? Is it really just unimaginable to you? It dovetails nicely with, for instance, the shift/swing results that I can't seem to get you actually to look at. (If I used change in WPE as the shift measure, would that help?)

I don't see your point about "the high correlation between state exit polls and late vote shares." If you're looking across states, pretty much any measure of partisan preference is likely to be highly correlated with any other: no matter how you slice it, Massachusetts is a lot bluer than Wyoming. I think I've addressed the other points previously.

ETA: Just by accident, I came across this article, which I had long since forgotten: http://www.mysterypollster.com/main/2004/10/the_bush_landsl.html . It's interesting that Mark Blumenthal was writing about what I call "false recall" (but he interprets as false reporting due to "social discomfort") a month before the 2004 election. You might construe this as a clue that survey researchers tend to think differently about these issues than you do. Or I guess you might construe that he was in on the plot?!