Reply #13: ok

[OnTheOtherHand]

The way the exit poll projections work (basically) is: the pollsters start with a pre-election estimate and with exit poll interviews. When the polls close, the pollsters figure a projection based on the interviews (not just raw percentages -- the interviews are compared against results from past races, and some other subtle stuff happens) AND the pre-election estimate. That shows up in the Edison/Mitofsky evaluation report as the "Call 3 Composite Estimator." The pollsters also calculate tabulations based on those projections, which is what appeared on CNN.com.

Then, as election results come in, the pollsters update their projections based on the results. Every so often, they also update the tabulations. In theory, adjusting the tabulations to the official results should make everything more accurate. Of course this would not work very well if the official results were wrong. But whether the official results were right or wrong, the point of adjusting the tabulations wasn't to cover up the discrepancy. It is indeed standard operating procedure.

I think probably everyone would agree that the exit polls should have been more accurate than they were -- they were way outside the range of random sampling error. As far as I can tell, most people familiar with surveys assumed that the problem was with the polls rather than the count. But it was always possible that the problem was with the count. (Or both, of course.)

Exit poll interview data has been released and archived for exit polls back at least as far as 1972. These data can be hard to interpret, because as I said above, the projections aren't just based on raw percentages in the returns. (For instance, if the precinct sample oversamples heavily Democratic precincts, that should not throw off the projections -- because they compare to past races -- but it could throw off the percentages.) But you can download the 2004 data right now and verify that sure enough, most of the respondents said they voted for Kerry. Certainly nothing in the Carter-Baker proposal would be likely to change that. I would like to see more openness, not less.

Some weird things happened in 2004, and some bad things happened. And no matter what happened in 2004, verifiable elections make sense, don't they? There are actually some decent reasons, in the abstract, for thinking that a paper trail isn't all that great -- but really existing paperless DREs are downright awful.