« We Agree | Main | Those Leaked PDFs »

January 03, 2005

More on (Sigh...) Exit Polls

Looks like my New Year’s resolution to change the subject won't last through my first day back. As those who read the comments on this blog probably know already, there were a few developments in the ongoing exit poll controversy during my break:

  • Someone posted a batch of pdf files that appear to be actual internal Election Day reports generated by Edison/Mitofsky for the National Election Pool (NEP). Each file shows cross-tabulations of the vote -- similar to those posted by the networks online -- that also includes an overall tabulation of the vote (marked "not for on-air use"). The site lists many different files; some are for the presidential vote, some for the vote for Congress. Each file has two versions, one showing percentages tabulated horizontally, one vertically. The files of greatest interest are those whose file names end in "NONE_H_data.pdf." These show the presidential vote preference on the "national" survey tallied at three different times (all presumably EST): 3:59 p.m. and 7:33 p.m. on Election Day and 1:24 p.m. the next day (11/3).
  • A new paper on the exit poll controversy is making its way about the Internet, this one by Jonathan Simon and Ron Baiman. It focuses on comparing extrapolated "results" from the national exit polls to the national popular vote.
  • Several emails and comments have noted that Steven Freeman has released a new batch of papers on his website, discussing his hypotheses for the "exit poll discrepancies" [although as of this posting, those links were not working].

Unfortunately, my time is up for today. Resolution aside, I will have more to say about all of the above over the next few days.

Related Entries - Exit Polls

Posted by Mark Blumenthal on January 3, 2005 at 05:27 PM in Exit Polls | Permalink



I haven't had much time to dig into the actual PDFs that were posted, but what I found troubles me so far.

First, in order to get the final 11/3 weighted exit poll to match Bush's final tallied count of a 51%-48% margin, Mitofsky's sample of 13,660 respondents had to be pulled largely (60%) from the South and Midwest (over 8100 of the 13,660 came from these two regions) while the West and East regions were suppressed in the final weighted exit poll that Mitofsky did the day after the election.

Second, if you want to accept the South's regional final (11/3) weighted exit poll results that were used by Mitofsky to construct his final conforming exit poll with a 51%-48% result for Bush, you would also have to accept that the South went 58%-42% for Bush. Yet the later exit poll on 11/2 done near poll closing (7:33 PM eastern) had only a 54%-45% Bush advantage, but it had the same gender breakdown as the 11/3 final poll. So in order for Mitofsky to weight the final exit poll to reflect the tallied vote results of a 51%-48% Bush victory, he had to find a way to weight the South with an even larger tilt to Bush than it had the day before while keeping the gender breakdown consistent with what it was on 11/2.

And he had to pull 60% of his final exit poll sample from the South and the Midwest in order to make his final exit poll reflect the tallied results.

I ask you: how plausible is all of this?

Posted by: Steve Soto | Jan 4, 2005 3:00:51 AM

Sign the petition to demand a revote in Ohio and Florida


Sign the petition to stop social security privatization, increase the minimum wage,and repeal the faulty Republican prescription drug benefit and replace it with a simple 80 percent coverage of medication under Medicare Part B.


Posted by: buckfush | Jan 4, 2005 7:01:07 AM

Get over it. The question is not why the vote was wrong, but why the exit polls were so wrong. You are living in a fantasy world if you believe that Mr. Kerry really did win in either Florida or Ohio.

Posted by: Bruce Hayden | Jan 4, 2005 9:39:29 AM

Steve Soto - You've hit the nail on the head, I'm afraid. There simply is no plausible way to reconcile the exit poll results with the actual election results. There are only two possible explanations. Either the exit polling company was incompetent and didn't know what they were doing, or else there was widespread fraud committed in this election.

Posted by: Jayson | Jan 4, 2005 12:57:15 PM


Why is it not possible that there was a combination of: 1) random survey error (inherent in all polls based on a sample); 2) systematic bias due to non-sampling error (coverage and/or non-response); and 3) innacurate election tally?

Posted by: Rick Brady | Jan 4, 2005 1:35:53 PM

Rick: No. 1 doesn't account for the magnitude of error exhibited. No. 2 and No. 3 are definitely plausible -- No. 2 would fall into my "pollster doesn't know what they're doing" possibility. No. 3 would be a symptom of my voter fraud possibility.

Posted by: Jayson | Jan 4, 2005 2:21:35 PM

Actually, #1 "could" account for a rather large percent of the magnitude (up to 1.96 Zs at 95% CL, or 2.575 Zs at 99%CL), therefore it should be considered as a possible explanation for "part" of the error. However, the random sampling error could also have occurred in the opposite direction, meaning that the discrepancy due to #2 and #3 is even larger than the exit poll shows! We simply do not know, but it must be considered as possibly part of the explanation.

Also, how is it incompetence on the part of the pollster if people refuse to participate in the survey and this non-response is for some reason biased? I think that some portion of any coverage error (which is technically non-sampling error from what I know) could be attributed to pollster incompetence, but non-response? How so?

Notice I am not making any value judgements about #1, #2, or #3; I'm only arguing that the observed discrepancy "could" be described by a combination of these factors.

Posted by: Rick Brady | Jan 4, 2005 3:17:56 PM

Jayson, you're missing the obvious. Exit polls are conducted only in certain precincts, predetermined to be bellwether precincts by the pollster. If the bellwether precincts turn out not to be representive of the entire state, then the exit polls will fail. John Kerry won all three bellwether counties in Ohio by narrow margins. But the election wasn't decided in those counties (as one would normally expect). High Republican turnout in partisan Republican precincts decided the election. The same was true in Florida (particularly in the panhandle counties). The polling outfit didn't cover these precincts and missed the high GOP turnout. Thus, Kerry had a better showing in the exit polls.

Posted by: ErikK | Jan 4, 2005 4:09:02 PM

Why are people so committed to propping up the exit poll results. Lets face it, the exit polls have been ridiculously wrong in 2000, 2002, and 2004. Either there was major nationwide vote fraud in each of these elections, or the exit pollsters are friggin' incompetent idiots. I will take the latter, its the most logical scenario.

Posted by: Marc | Jan 4, 2005 7:29:25 PM

Those exit polls had South Carolina "too close too call."

Bush won by 16 points. Anyone think that was the result of massive fraud? Does anyone think that it was ever too close to call in SC?

The answer is that the exit polls were wrong.

Posted by: blaster | Jan 4, 2005 8:44:06 PM

ErikK states:

"Exit polls are conducted only in certain precincts, predetermined to be bellwether precincts by the pollster."

Precinct selection does not work like that.

Precincts are chosen randomly within states. Try MP's exit poll FAQ to read about the exit poll methodology.

Posted by: Alex in Los Angeles | Jan 4, 2005 11:10:33 PM

The comments to this entry are closed.