Reply #49: Weights....

[Febble]

FEBBLE
TIA, all your analysis assume that exit poll sampling is unbiased, and that people correctly report what they did four years ago (e.g. whether they voted; who they voted for). While it is possible that these two assumptions are correct, there is no reason to believe that there are, and a fair bit of reason to believe they are not.

TIA
Once again, you avoid the FACTS about the 2000 and 2004 elections.

I have stated the following over and over again. We already know the MAXIMUM "How voted in 2000" weights because we KNOW how many Bush and Gore voters were still alive and could vote in 2004.

I have no idea, TIA, what you mean by "weights". If you had bothered to download the actual questionnaire data, from which the crosstabulations you use were actually derived, you might have a better understanding of the term. But to explain briefly: in that spreadsheet - the spreadsheet containing all the answers to the exit poll questions with a line for each respondent, and a column for each question - there is also a column marked "weights". Each respondent has a positive number in this column. A number less than 1 means that that respondent's answers will be downweighted in the crosstabulation, and indicates that respondents like him/her were deemed to be over represented in the poll, while a number greater than 1 means that that respondent's answers will be upweighted in the crosstabulations, and indicates that respondents like him/her were deemed to be under-represented in the poll.

A number of factors are taken into consideration when computing the weights (and if you had downloaded the spreadsheet while it was available you could have computed the crosstabs with and without weights and you could even have experimented with your own weights). One factor is the geographical stratification of the sample. Another is the reported values for the age, race and sex of non-respondents. If, say, white middle aged men are better represented amongst non-responders than in the sample, then white middle aged men will be upweighted in the sample.

These weights are changed at intervals on election night (as you know) as the vote count comes in. I will not debate here the rights or wrongs of this process, merely that it is done.

What is NOT a factor in computing the weights is answers to the question "how voted in 2000". I do not know what you mean by "weights" in this context. You are either confused as to how the weights are computed, or you are using the term in a different sense. If the latter, you need to make it clear what you mean. It is not at all clear to me what you mean.

Yet you still assume that the weightings are the result of a flawed sample. That is flawed reasoning. WE ALREADY KNOW THE WEIGHTS SINCE WE KNOW THAT ALL MEN ARE MORTAL.

THEREFORE, it is IRRELEVANT whether Gore 2000 voters forgot or lied to the exit pollsters IF they said they voted for Bush. What IS RELEVANT IS WHO THEY VOTED FOR FIVE MINUTES BEFORE THEY WERE EXIT POLLED - and 91% said it was Kerry.

And so these two statements make no sense. It is true that the fact that the weights are not equal to 1 for every respondent indicates that the pollsters do not consider that the sample bears a 1:1 relationship with the electorate. Indeed, they know for sure that it does not (from their age, race and sex data on non-respondents) and they also base their stratification on past vote-count data which they do not assume is a flawless predictor of patterns of vote-count in the current election. And if, after reweighting their sample in the light of what they think (on the basis of data from many sources) is a better apportioning of the electorate, the proportion of those who indicate that the voted in 2000 scales up to a larger number than could have possibly voted, then there are three possible inferences that could be made. One is the one you draw, that the vote count itself was inaccurate. A second is that the weighting process itself was too crude. And a third is that some people did not report accurately whether they voted in 2000, or who they voted for.

You may choose to consider the second and third possibilities implausible, and that is your right. But it does not come out of the math. It comes out of conviction. The math works just fine. Even if we assume that the weighting process was broadly correct, all one needs to do to square the resulting cross-tabulations of the 2000 vote with what could actually have happened is to postulate that a small proportion of voters misreported their 2000 vote (whether they voted or for whom). And we know, from the work Mark Lindeman has done on longitidinal data (data that actually tracks answers from the same voters over time), that this actually happens, in the pattern required to solve the problem. Why should they do this? Who knows. Can it happen? Yes, and Mark's evidence is strong evidence that it does.

The 2000 and 2004 recorded votes and the annual MORTALITY rate are historic and documented FACTS. They are both NECESSARY AND SUFFICIENT to determine the MAXIMUM number of Bush and Gore voters who COULD HAVE VOTED in 2004.

Indeed.

Only Bush and Gore 2000 VOTER TURNOUT in 2004 is UNKNOWN. But turnout can be ESTIMATED in order to determine PLAUSIBLE WEIGHTS (95% is the base case used in the True Vote Model). The model indicates that Kerry won by 66-58 million votes, a 52.56-46.43% vote share.

Well, until you say what you mean by "weights", this statement is not comprehensible. I will therefore snip a bit...

THE MAXIMUM NUMBER OF GORE, BUSH AND NADER 2000 VOTERS WHO COULD HAVE VOTED IN 2004 (I.E. THE WEIGHTS) IS A FUNCTION OF 1) (F)ACTUAL 2000/2004 RECORDED VOTE DATA, 2) THE ANNUAL 0.87% MORTALITY RATE AND 3) AN ESTIMATED 2000 VOTER PERCENTAGE TURNOUT IN 2004. THAT'S IT.

Do us all a favor and accept the reality of the above statement. Your avoidance of this COMPELLING evidence for almost two years is truly astounding,being that you are an extremely intelligent individual.

Gee, for a ghost you shout pretty good.

"THE MAXIMUM NUMBER OF GORE, BUSH AND NADER 2000 VOTERS WHO COULD HAVE VOTED IN 2004" does not equal "THE WEIGHTS". It is, however, a function of the actual recorded vote data and the mortality rate. And if you want to throw in an estimate of turnout into your estimate of the maximum, that's fine (although I suggest you read the piece I linked to when estimating turnout).

So your assertion that I "avoid this COMPELLING evidence" is false. I am perfectly happy to accept that there were a large number of voters in 2004 who could not have voted (and did not) in 2000. What I don't find is that this is "COMPELLING" evidence of fraud. It is simply evidence that there were more voters in 2004 than in 2000. Which is not in dispute.