Reply #81: Saying what? [View All]

Obviously gender, being a visible characteristic, will be held constant in the weighting.

Vote in 2000, being an invisible characteristic, will not be. In fact it won't be considered in the weighting at all, because the data is only obtained from a subset of the respondents.
So there are three reasons why the fact that the Gore:Bush percentages in the reweighted ("massaged" is fine by me; my own metaphor was "subjected to metamorphic pressure")sample don't jibe with the 2000 numbers:

1. The re-weights were crude (which they necessarily must have been)


2. Recalled vote is not a reliable indicator of actual past vote (which we know, from panel data)


3. The count was corrupt.

I DON'T believe the crosstabs tell us that the election wasn't stolen. But NOR do I believe that it tells us that it was. We are talking about survey data, and the alternative hypothesis to the fraud hypothesis is that the poll respondents were not a purely random sample - that those who "should" have been in the poll, but weren't were "drawn from a different population", as we say in stats, from those who were. And as those who "should" have been in the poll but weren't comprised at least 47%* of those selected, then that leaves a fair bit of room for a systematic difference. If the non-participants were more likely to be Gore-Bush voters; or DNV2000-Bush voters, then that would affect the crosstabs, as I said.

I maintain this is plausible, not that it is fact. But because it is perfectly plausible (to me) I don't buy the argument that it violates reality. Biased samples, especially with face to face selection, is all too real a phenomenon.

My case is simply that the exit poll evidence is very poor evidence for a stolen election.

That is NOT the same as saying the election wasn't stolen (although I am not convinced it was) and it is certainly not the same as saying the election was just.


*47% assumes all non-participation was due to non-response; it ignores the possiblity that selection itself was non-random.