For which characteristics is it valid to weight an exit poll?

In order to interpret an exit poll it is sometimes valid to
weight the data against known results. My favorite example is
the two precinct model:
Precinct A votes 60-40 for Kerry over *
Precinct B votes 40-60, i.e. in the opposite direction.
Exit polls are taken for each precinct, and are perfectly
accurate: there are 100 people polled at each, and A gives 60-40
and B gives 40-60.
From this we cannot derive a result: we need to know the turnouts
at the two precincts. Clearly turnout is a measurable quantity
(look at the number of signatures on the poll-books). If we
knew the turnout at A and B we could then predict the results.

There are some other measurable quantities which might also
prove valid to use to weight exit polls
---- please note that I am not saying to modify the polls: what
I mean is that the weights are used to interpret the polls in a
predictive setting ----
in particular, there are states in which one registers as D or R or I
and this can also be recovered from the polling books. Can gender
be recovered in any state?

I am trying to get a handle on *what* sort of weighting the
pollsters can have done on election night late in the evening:
there are some which are invalid in my book -- comparing outcomes
to polls to reweight the polls to predict the recorded outcome
is rather like going back over an exam with the answer key before
it is graded. In my biz, that is called cheating and gets the
student tossed out. On the other hand, weighting by precinct
turnout is statistically sounder.

Any thoughts/knowledge on this, anyone?

Thanks,
Breadbox.

You can use your predicted demographic turnout in all exit poll data until the final exit poll.

Here is how it can work: You have two pollsters on duty at each precinct. One does exit polling by talking to people who just voted ("Questioners"). The other records demographics like how many men vs. women exit the poll after voting without asking them questions ("Tally Takers"). At the end of the night, you weight the exit polling data of Questioners to the data collected by the Tally Takers, and your result will be different than the raw Questioners data not adjusted by the Talley Takers.

Talley Takers can also, at the end of the night, compare how many people voted by party based on the actual public records of who received a ballot and their registered party. They can then adjust the Questioners data by the party affiliation data they can collect after the poll closes.

Finally, since western states close much later than eastern states, there is often a prediction of voting patterns in western states that is used for a longer period of time than in eastern states, and only the final poll can adjust for western states actual data.

And, of course, NONE of this data can include absentee votes, as there is nobody to question or tally. If absentee voting is significantly different (in quantity or voting patterns) than in prior years, it skews the entire poll.

So, the final poll can weight based on Tally Taker data (such as gender and party affiliation) that is NOT available for a preliminary poll, and it can include western state and absentee voting data not available for a preliminary poll.

Again, I am not saying that this IS the weighting used. Just that it might be...and we can't really know until they release the raw data.