Reply #18: I give up.

[Febble]

Well, one last go.

Reweighting does not mean adding new samples. It means reweighting the data you have in the light of new information. It is, as you point out, mathematically impossible for the final sample to have been so different from the previous two. Quite apart from anything else, the MoE would go through the roof as the between sample variance would vastly outweigh the within-sample variance. So what happens is that all the data is reweighted. Not in the light of the final sample, but in the light of other information. Mostly turnout, I would guess. It's a dynamic process apparently. It's not a secret that its done. It's not like a horse race where the bets are off after the starting gun. They keep adjusting the odds as it becomes clearer and clearer which horse is going to win. Especially after 2000 where they called Florida for Gore then Bush (or was it the other way round). They want to get it right.

But I don't expect exit-poll companies like to have to do a lot of reweighting as it means their initial weights must have been "wrong".

The extent of the reweighting just tells you the extent of the problem you have to account for.

The reason could be sampling bias or fraud. Take your pick.