These calculations originally appeared on the website of Air America and have been widely circulated via email. This data has not been independently verified. Note how the exit poll data roughly matched the results in states with paper ballots, but was widely off in states that used unverifiable electronic voting machines, particularly Florida and Ohio. Assuming these data sets are correct, what could account for this discrepency?
Related Links and Updates:
A subsequent story cited at the same website contains a report that a former MIT math professor analyzing a larger data set has concluded that there was no statistical significance in the final count in favor of Bush in states that used electronic voting vs. paper ballots if data from New York is included. The same former professor also reportedly calculated the odds that Bush would make an average gain of 4.15 percent among all 16 states included in the media consortium's 4PM exit poll at 1 in 50,000 or .002 percent.
The Democractic Underground website has a table of raw election night figures (see top of page) indicating that in eight battleground states with no paper ballot backup Bush had an average net gain of 6.25% over the 4PM exit polls on election day compared with an average 0.8 percent Bush gain over the 4PM exit polls in states that do have a paper ballot backup. There appears to be some confusion, however, as to which states used paper ballots and which did not.
A November 19, 2004 report from a team of scholars at the University of California led by sociologist Michael Hout claims statistical evidence of an unexplained boost in votes for Bush tied to the use of electronic voting machines in Florida (but not Ohio). Researchers who have contested or clarified Hout's paper include Kieran Healy and Andrew Gelman; a nice summary of their findings can be found in Kevin Drum's piece in the Washington Monthly.
The best hope for clearing up this controversy appears to rest with the recently established National Commission on Elections and Voting, a project of the Social Sciences Research Council.