Reply #82: Shorter Febble:

[Febble]

Presumably, "the metamorphosis" changes "how many", but not "of what" (or "whom")

Right.

It won't change:

a White Middle-Aged Male reportedly-Bush 2000-voting Republican who voted for Kerry, into
a White Middle-Aged Female reportedly-Gore-voting Democrat who voted for Bush.

But it might apply equal upweights to:

a White Middle-aged Male reportedly-Gore-voting Republican who voted for Bush (WMMGRB), as to
a White Middle-aged Male reportedly-Bush 2000-voting Republican who voted for Bush (WMMBRB)

because the re-weighting algorithm doesn't have access to data on whether these two groups of voters were equally under-represented or not (because there isn't any). And if they weren't, it won't get the re-weighted crosstabs right, even if it gets the 2004 vote proportions right, and the age, race, sex characteristics right, even if the voters correctly reported their past vote. If, for example, WMMGRB voters were more under-represented in the poll than WMMBRB voters, then the weight would be too low for the first group, and result in an underestimate of Gore voters in the reweighted cross-tabs.

Well, not a lot shorter, but I hope clearer. And on top of that, there's the mis-reported vote thang.

As I said, a lemon, not a clinch.

(edited for typo)