Something's rotten in the state of Denmark

Saturday, November 13, 2004

Exit polls, continued

A reader writes in about the competition between the two exit poll studies -- the Freeman poll and the CalTech/MIT poll.



You are right. Caltech/MIT use different exit polls than Dr. Freeman. Dr. Freeman uses election night exit polls, and frankly he is right to do so. Caltech/MIT use exit polls that have been "corrected" to match the election results.

The Caltech/MIT study uses the "corrected" exit polls found on CNN.com.Because "corrected" exit polls are weighted to MATCH the election results comparing them to the election results to look for fraud is irrelevant, of course. The Final Exit Polls released on election night are not weighted to match the election results and so provide the relevant data.

For a pollster's explanation of exit poll methodology, Mark Blumenthal's Mystery Pollster blog provides excellent background.

Apparently, this mistake occurred because the authors are confused about the nature of exit polls. They do not seem to be aware that after elections exit polls are reweighed to match the election results. The study even fails to make this distinction when it uses election night national exit polls side-by-side with "corrected" state polls.

Dr. Freeman's study does understand this distinction. He describes exit polls as either "corrected" or "pure." Many other professors looking at the exit polls have become aware of this important distinction as well: Andrea Moro, and professor Sam Wang.

It is unfortunate that the Caltech/MIT study is misinforming the exit poll debate.


My take is still that the exit poll discussion might be irrelevant. I'm glad that the Freeman study reminds us of how rare it is for exit polls to be this wrong -- the oft-quoted 250,000,000 to 1 odds -- but I think that when the polling practitioners like Zogby et al are not standing by and defending their poll results, it makes it hard, real hard, damn hard for the rest of us to defend them for them. In the same way, I don't want to see people using the Caltech/MIT study as proof positive that there was no funny business/irregularities in this election.