Jim Lampley’s concern over the gap between the exit polls and the final results in the 2004 election is sensible. Exit polls are, after all, used by independent election observers to monitor elections. When large gaps between the exit polls and the final results emerge, then there is a chance that the election was stolen. While we live in a mature democracy, the lust for power could conceivably lead criminals to try to steal an election.
Since the election outcome and the exit poll are two possible views of the “correct” result, one can not prove that one is right and the other is not. One can only construct hypotheses other than fraud that might explain the differences, and confirm or reject these with the data. If no reasonable alternative hypotheses can be supported by the data, then the suspicion of fraud intensifies.
For the 2004 election, there are two main hypotheses consistent with the disagreement between the two numbers. The first is that Bush stole the election. The second is that Republican voters were unusually unwilling to chat with exit pollsters, and the exit polls were wrong. While fraud can take many different forms, making the first hypothesis quite diffuse in its implications, the latter hypothesis suggests a number of possible patterns in the exit poll data that we can look for.
While investigation of fraud is not the main objective of the Edison/Mitofsky study, it does allow one to collect evidence that sheds light on these issues. My read of the study is that it is quite supportive of the second hypothesis. While there are many patterns in the data that suggest something other than fraud, for me, the most interesting finding is that exit poll completion rates were strikingly low for polls taken by college and graduate students. When college kids did the interviewing, a higher percentage of interviewees refused to cooperate. Since conservative speakers have a hard time speaking at a college or university these days without getting a pie in the face, it is plausible that conversations with these interviewers were not attractive to many Republican voters. (And note that this observation does not depend on assuming that the actual vote was correct).
Perhaps the main action item from the study relates to this finding. In the future the firm will “use augmented recruiting methods to reduce the proportion of students and young adults as interviewers.”
Since the underlying patterns in the data support the alternative hypothesis, one need not appeal to fraud to understand the difference between the exit polls and the actual outcome. I side with Byron. Jim can keep digging if he wants to, but my guess is that no evidence of election-changing fraud will ever be discovered because there was no fraud.
But if the same thing occurs in the next election, and the exit pollsters are all high-school-educated retirees, then I will be highly suspicious.