Reply #41: changing the weights throws the MOE off [View All]

[chorti]

Once you change the weighting of the poll, the whole thing, all the questions, need to be re-weighted. This can not be explained by the margin of error. Once you do the re-weighting, the reported results will be outside the margin of error of the exit poll. In other words, the national popular vote total is impossible. It is far outside the possible margin of error of the national exit poll survey.

Let's say that at most, 97 percent of Bush voters from 2000 returned to the polls in 2004, due to 2 percent passing away and another 1 percent who could/did not vote for some reason. This would mean that at most, 40.85% of the 2004 electorate were folks who voted for Bush in 2000, rather than the 43% used in the national weighting. This is a significant difference and means that the entire exit poll needs to be re-weighed. It would throw off all the questions, such as 'Are you a liberal, moderate, conservative?' etc.

And what does it do to the question of how folks voted in 2000? Let's split the 2.15% subtracted from Bush-2000 voter portion of the electorate, placing an additional 1 percent in the new voter category (those who didn't vote in 2000) and an additional 1.15% in the Gore-2000 category. (This seems like a conservative adjustment favoring Bush). Then we have the following, looking only at the Bush and Kerry portions:

How did you vote in 2000?
..............Portion of..........................Bush......Kerry
..............electorate....Bush...Kerry......partile....partile
Did Not Vote...18%........45%.....54%..... 0.081.... 0.097
Gore ...........38.15%.....10%.....90%..... 0.038.... 0.343
Bush ...........40.85%.....91%..... 9%..... 0.372.... 0.037
Other ........... 3%.........21%.....71%..... 0.006.... 0.021

Total ............................................... 0.497... 0.498

Kerry wins 49.8% to 49.7%. In other words, even in the best-case scenario for Bush, where he gets 97 percent of his 2000 voters to return to the polls in 2004 (but loses 9% to Kerry), he basically tied Kerry in the popular vote. With a margin of error of 1%, this would produce a range of -1.9 to +2.1 for Kerry over Bush.

The actual reported margin of 2.7 percentiles in favor of Bush falls well outside this margin. In fact it would produce a T-score of 3.33, which has a 0.09 percent probability of happening. This means that there is a less than one out of a thousand chance that the reported popular vote total is accurate. And this is using the "fixed" exit poll data which had already been adjusted in Bush's favor and using assumptions along the way that favor Bush.