Reply #41: I Have Not Been Able to Find a Description of the Methodology [View All]

[On the Road]

used in weighting the exit polls, but I did find this from a discussion with Steven Coll, a Washington Post editor:

Washington, D.C.: I don't understand The Post's polling numbers, because they contradict each other. Looking at the math according to different demographic groups, it simply doesn't add up. According to your breakdowns with regard to gender, party affiliation, ideology, region and 2000 presidential vote, Kerry has a clear statistical lead. Yet according to religious affiliation and perhaps issues, Bush wins. What's going on here? Why the statistical contradictions?

Steve Coll: The exit poll numbers we have paid for and been provided simply do not add up. They are internally inconsistent in important ways. They also are out of whack with voting results in ways that are difficult to explain. One thesis being explored today is shorthanded as the problem of "female skew." This refers to the fact that women are more likely than men to agree to be interviewed about their votes outside of polling stations. In fact, in the exit polls we received yesterday, there were more women in the sample than we expected to see in the final turnout. But the analysts handling this data believed that this distortion would not change the general trend of the poll and had been weighed to some extent by the poll's managers. We need to review questions such as this one in greater depth, although one's confidence that it will ever be possible to conduct accurate exit polls in the heat of a campaign such as the one we just had has to be shaken.

http://www.washingtonpost.com/wp-dyn/articles/A13590-2004Oct31.html

Coll seems to be saying that different crosstabs and breakdowns of the data don't match. He does not mention a problem with revisions or reweightings during the day.