[I hooked up to the dial-in press conference at UC Berkeley today. Naturally they had a technology failure, so dial-in participants were cut off during the Q&A period and couldn't ask questions after they had been restored to the conference. But I was able to get this (which will be supplemented by online publication of the paper whenever I can find it). Here's what I learned.]

Michael Hout of the sociology department announced the results of a statistical study of Florida voting patterns. The researchers found that statistical models indicate that Bush received an excess of from 130,000 to 260,000 votes in certain counties using e-voting equipment.

Hout said that there was only a "trivial possibility" that the discrepancy could be explained by other factors—effectively less than 1 in a 1000 chances. He added that adjusting for other factors such as ethnicity and income of the voting population "only made the results more robust."

The effect was concentrated in the large Democratic-leaning counties of Broward, Palm Beach and Miami-Dade, but was not limited to them. The effect was actually correlated with the total of votes for Al Gore in the 2000 election—i.e., the greater the number of votes for Gore in the 2000 election, the greater the excess of votes for Bush in the 2004 election!

On the other hand, the group did not find a statistical discrepancy in the so-called "Dixiecrat" counties of Florida that have high Democratic registrations but tend to vote Republican. (See my post "A slender reed".)

The research group used a statistical method known as multiple regression analysis that began with a model that compared support in Florida for Bush in 2000 with the votes tallied for Bush in 2004. Hout said their model was able to account for over half the variance between the two elections, but in accounting for the remaining variance the three Florida counties stood out.

Hout emphasized that the model is a standard tool of professional researchers and that it has been used previously to identify problems in prior elections. He noted that in assessing the results of the election, "our statistical approach is about the only way we have without a paper trail, as in Nevada."

The research team also checked the results in Ohio, but did not find the kind of discrepancies that occurred in Florida. Hout added, however, that "there are other types of problems that our model cannot pick up."

History of the research

In the wake of the November 2 election, two graduate students began to discuss ways to test election results. They looked at some of the claims of voting irregularities being made on the internet and were able to "knock them down." But when they looked at voting changes taking the 2000 election as a baseline, they noticed certain anomalies.

The students contacted their professors to see if their results could stand up to scrutiny. After verification of the initial results, Hout became involved. Hout said that he was initially dubious, taking a "show me" attitude. But after further review of the data, Hout said "I converted from skeptic to the person you see before you now."

Hout refused to offer any speculation as to how the discrepancies occurred, noting that he is neither an engineer nor a programmer. He also emphasized the apolitical nature of the research, saying that he was not part of a political action committee or a lobby.

What's to be done?

According to Hout, "our results give credibility to those who found fault with the results in Broward County." As for the reason for holding the press conference, Hout said, "Something went awry with electronic voting in Florida. We want to make sure someone examines these results!"

He concludes, "We're not saying the result would change"—that is, who actually won the election, "but the tally would change."

---

The data are online here.

Note: The reported vote tally in Florida was Bush 3,911,825, Kerry 3,534,609 for a net difference of 377,216 votes. The vote irregularity found by the Berkeley study is not sufficient to overturn the Florida outcome.

Related post
A slender reed