Complete US Exit Poll Data Confirms Net Suspicions

Wednesday, 17 November 2004, 10:45 pm
Article: Alastair Thompson

Scoop.co.nz is delighted to be able today to publish a full set of 4pm exit poll data for the first time on the Internet since the US election. The data emerged this evening NZT in a post on the Democratic Underground website under the forum name TruthIsAll.

The new data confirms what was already widely known about the swing in favour of George Bush, but amplifies the extent of that swing.

Click for big version
Figure 1: Graph showing the "red shift" between 2004 US General Election exit polls & the actual 2004 US Election results

In the data which is shown below in tabulated form, and above in graph form, we can see that 42 of the 51 states in the union swung towards George Bush while only nine swung towards Kerry.

There has to date been no official explanation for the discrepancy.

Ordinarily in the absence of an obvious mistabulation error, roughly the same number of states should have swung towards each candidate.

Moreover many of the states that swung against Democratic Party hopeful John Kerry swung to an extent that is well beyond the margin of error in exit polls. Exit polls by their nature - they ask voters how they actually voted rather than about their intentions - are typically considered highly accurate.

Last week in an analysis of a similar, but incomplete set of data, Dr Stephen F. Freeman from the University of Pennsylvania calculated that the odds of just three of the major swing states, Florida, Ohio and Pennsylvania all swinging as far as they did against their respective exit polls were 250 milllion to 1. (See…"The Unexplained Exit Poll Discrepancy" – Dr Stephen F. Freeman - .pdf format)

Dr Freeman's academic paper contains a thorough description of why and how exit polls are conducted (in some countries they use them to prevent against vote fraud), and considers a number of hypotheses for why this year's polls could have been so dramatically wrong. He concludes that the reasons are unknown.

CAUTIONARY NOTE: The data that is released today shows the 4pm data run from the Edison-Mitofsky polling company. This run was based on 63% of the full 13660 sample in the poll. However as we also have a set of data from around midnight with which to compare this data, we can tell that the final exit poll results were not that far different than these early results. This in itself tends to suggest that the polling system did not have a systemic bias in its early data as suggested by some commentators in early reports on this puzzle.

(For a more detailed description of the limitations of this data and the claimed gender bias in the early data see.. EXTENDED FOOTNOTE ON THE LIMITATIONS OF THIS DATA - By Jonathan Simon )

BACKGROUND

Ever since the first analyses (See... "Faun Otter: Vote Fraud - Exit Polls Vs Actuals ") showing the swing in favour of US President George Bush between the exit polls and the actual results were published, the internet has been swimming with rumour and speculation about what the results meant.

These initial internet news reports were debunked in a report from the CALTECH/MIT Voting Technology Project which was widely distributed to the media in the days immediately following the election. The unnamed authors of this report stated:

"If we look at the 51 separate exit state polls, we see that 30 predicted more votes for Kerry than he actually got, while 21 predicted more votes for Bush than he actually got. Therefore, at the state level, the polls favored Kerry less than the sum of all the polls aggregated up to the national level. Furthermore, if we do a statistical test to see whether the differences between the exit polls and the official returns are significant, only three out of 51 are.5 Therefore, while it is fair to say that the exit polls predicted a significantly greater vote for Kerry nationwide than the official returns confirmed, it is not immediately apparent that any systematic biases are revealed when we take the analysis down to the state level."

This report was subsequently quoted in a November 12th New York Times front page article ("Vote Fraud Theories, Spread by Blogs, Are Quickly Buried") that purported to debunk Internet conspiracy theories and misconceptions about the 2004 election, including those about the exit polls. The New York Times stated:

A preliminary study produced by the Voting Technology Project, a cooperative effort between the California Institute of Technology and the Massachusetts Institute of Technology, came to a similar conclusion. Its study found "no particular patterns" relating to voting systems and the final results of the election.

"The 'facts' that are being circulated on the Internet," the study concluded, "appear to be selectively chosen to make the point."

However CALTECH/MIT's analysis had already been proved flawed on November 11 when Scoop.co.nz published the first iteration of a set of data that was fortuitously captured by VerifiedVoting.org activist Jonathan Simon in the early minutes of Nov. 3 (See… "47 State Exit Poll Analysis Confirms Swing Anomaly"). Dr Freeman's report was also based on this data. However Jonathan Simon had not managed to capture data for all states - hence the hunt for the full set of data continued.

Interestingly after the Simon data was widely circulated in the blogosphere the authors of the CALTECH/MIT report edited their footnotes (see footnote 2 & compare with the version cited above & hosted on Scoop) making it clear that the source of their data was the publicly available Exit Poll reports on CNN.com which were "rebalanced" in the early hours of Nov. 3. This data which has effectively been recast to fit the final results cannot really be termed exit poll data at all and has been the source of a great deal of confusion.

The complete set of New 4pm Edison & Mitofsky 2004 General Election exit poll data follows in tabulated form (sorted in descending order of the magnitude of the "red shift"):

NOTE: red shift = the exit poll margin - final count margin

EXTENDED FOOTNOTE ON THE LIMITATIONS OF THIS DATA
By Jonathan Simon

Always keep in mind that this 4pm exit poll data is the early sweep, skewed for women (58% to 42%) and therefore towards John Kerry/Democrats.

While this sample may have been accurate for the time of day measured (more women voting because of workforce composition and schedules), it constituted only one component of the full exit poll (63% of the sample), which consisted of two additional sweeps which reflected increased participation of men in later hours.

Thus the later updates (c. 12:20 a.m.), which impounded all three sweeps but was posted before contamination with tabulated data, is the best results we have at this time for comparison with the final results.

The first sweep has been singled out as having poisoned the whole barrel; but this is ridiculous - you simply need to make a slight adjustment to the final gender breakdown (@ 12:21 a.m.), if you believe that to be skewed, which has very little effect on the results (read on to see exactly how this works in terms of actual numbers).

That is, whether or not the first sweep had some distortion relative to our expectations for that time of day - it all amounted to very little, and is easily and inconsequentially adjusted for in the long run.

Let's see how it works. Take Florida with the 54% women/46% men exit poll sample that was supposed to be "way" off - you get the following for the full sample of respondents (men and women): 49.8% Bush, 49.7% Kerry.

If we adjust the sample to 52% women/48% men (probably about right), you get the following for the full sample of men and women: 49.9% Bush, 49.6% Kerry - that's right, a whole glaring 1/10th of one percent difference.

If we go "all the way" to a 50% women/50% men sample, now it's 50.0% Bush, 49.5% Kerry, a whole 'nother 1/10th of a percentage point.

The reason for this is that the gender gap is just not that dramatic and neither is the gender departure from a perfectly weighted sample.

In fact, if we want to get Bush up to a whole 1% lead, we'd have to take an exit poll sample of 55% men / 45% women.

But Bush "won" Florida by 5% (52% to 47%), a "red flag" discrepancy from the exit poll, however the poll is weighted by gender. Here's a fun fact: to get the exit poll results to equal the tabulated outcome, you'd have to sample all men, that's right 100% men / 0% women, just like the good ol' days before they passed that blasted 19th Amendment.

The argument has also been made that the early (morning/afternoon) exit poll sweeps were gender weighted 58% women/42% men, thereby invalidating the exit polls in general.

© Scoop Media

Alastair Thompson

Scoop Publisher

Alastair Thompson is the co-founder of Scoop. He is of Scottish and Irish extraction and from Wellington, New Zealand. Alastair has 24 years experience in the media, at the Dominion, National Business Review, North & South magazine, Straight Furrow newspaper and online since 1997. He is the winner of several journalism awards for business and investigative work.

Contact Alastair Thompson

• Linkedin
• Twitter - @althecat
• Facebook
• Email - alastair@scoop.co.nz