Reply #43: Sometimes this is like trying to conduct [View All]

a conversation in a blizzard.

The numbers I am talking about are not the "final exit poll numbers" which were indeed weighted to the vote returns, as they announced they would do in advance, as they do every election, and as has contributed to their reputation for "uncanny accuracy".

What I am talking about are the raw, precinct exit poll returns, before any kind of processing whatever has been applied. These numbers show that the actual raw returns, when compared with the precinct count made mostly at the precinct (60%) but in some cases from the county tabulations (40%) were substantially discrepant from the count, with discrepancies in both directions, but with the mean discrepancy substantially "redshifted" ie. on average, the percentage of poll responses for Kerry was higher than the percentage of Kerry votes in the count.

Two hypotheses could account for this: bias in the sampling of voters or bias in the count of votes, and I am trying to distinguish between these two. We cannot do this by looking at the "final exit poll", and I am not even looking at the "final exit poll" numbers. I am looking at the correlation between the precinct level discrepancy between the two sets of numbers (raw poll; precinct count) and the degree to which Bush's vote share changed between 2000 and 2004 in that precinct.

OK?