Reply #56: The exit polls

[Time for change]

I made four points against believing that exit poll bias explains most or all of the red shift. It sounds to me that you agree with me factually about points b. and d. 2004 exhibited the largest red shift (discrepancy between the polls and the official vote) since Mitofsky began polling in 1988, though there was almost as much discrepancy in 1992, as you point out. And the precinct selection favored Bush, which means that there would have to be even more exit poll bias within precincts to account for the difference between the official vote and the polls -- though as you point out the precinct selection factor was small.

I don't understand your response to my point c. I don't know what you mean by "design effect", or what that has to do with the fact that 5 of 11 swing states were outside the margin of error with regard to red shift.

Now, let's consider my first point about the exit polls, which is that Mitofsky in his report presents little or no evidence that exit poll bias accounted for the red shift. You disagree with that statement, and to make that point you say "bias was greater in precincts where some factor which might be expected compromise random sampling, and thus allow for the introduction of bias". Then you give two examples, including distance from the polls and low interviewing rate.

I have three problems with this explanation.

First, as you correctly point out, there was no information given that would allow the calculation of statistical significance.

Secondly, in the two examples that you give (and I believe that this applies to any other example that could be given), there nevertheless remains a red shift in every single category, including the ones that supposedly would be most favorable for accurate interviewing. For example, the median WPE with the smallest interviewing rate (1) is still -4.5, and the red shift for the smallest distance from the polls (inside the polling place) is -4.2. In either case, the red shift even under the most favorable of circumstance is still great enough to overturn the election if we postulate that the red shift is due to fraud rather than to exit poll bias.

And thirdly, not all the red shifts are even in the direction that you might expect to be consistent with your hypothesis that factors that might compromise random sampling might lead to bias. For example, highly educated interviewers were actually associated with a greater red shift, and I would expect that the more educated the interviewer is the less likely it would be that the interviewer wouldn't be able to conduct the interview properly.