The discrepancy, reported Thursday, is insufficient by itself to sway the outcome of the presidential race in Florida, but the UC Berkeley team called on Florida elections officials for an investigation.
"This is a no-vote-left-behind kind of project, not a change-the-president project," said UC Berkeley sociology professor Michael Hout, who oversaw the research. "We're as interested in the next election as the one just over."
Broadly speaking, the UC Berkeley team found that Bush received tens of thousands more votes in electronic-voting Democratic counties than past voting patterns would have suggested. No such pattern turned up in counties using optical scanning machines.
The UC Berkeley report has not been peer reviewed, but a reputable MIT political scientist succeeded in replicating the analysis Thursday at the request of the Oakland Tribune and The Associated Press. He said an investigation is warranted."There is an interesting pattern here that I hope someone looks into," said MIT arts and social sciences Dean Charles Stewart III, a researcher in the MIT-Caltech Voting Technology Project.
Stewart isn't convinced the problem is electronic voting. It could be absentee voting or some quirk of election administration. But whatever the problem, it didn't show up in counties using optical scanning machines. Rather than offer evidence of fraud or voting problems, the UC Berkeley study infers they exist mathematically.
Frustrated at the lowbrow, data-poor nature of allegations of election fraud flooding the Internet, three Berkeley grad students decided to apply the tools of first-year statistics class.
"We decided, well, you might as well test it properly instead of sitting around speculating," said first-year sociology grad student Laura Mangels. She and two colleagues downloaded voting and demographic data, ran them through statistics software and in the first night had results that produced a collective "Wow" among the students, she said.
They shopped their results to faculty and finally to Hout, a well-known skeptic who is chairman of the university's graduate sociology and demography group.
"Seven professors later, nobody's been able to poke a hole in our model," Mangels said. "Our results still hold up."
Hout agreed. "Something went awry with the voting in Florida."
They found nothing out of the ordinary in Ohio. But in Florida they discovered a small, unexplained boost in Bush support in three heavily Democratic counties compared to how those counties voted in 1996 and 2000.
The counties Broward, Palm Beach and Miami-Dade were at the eye of Florida's 2000 election storm. All traded out their reviled punchcards for touch-screen voting machines sold by either Omaha-based Election Systems & Software or Oakland-based Sequoia Voting Systems.
The Kerry-Edwards campaign and allies concentrated most of their Florida effort in those three counties.
In Broward County, the students found, Bush appeared to have received 72,000 more votes than would be forecast based on Broward's past voting patterns.
The UC Berkeley study estimates that all 15 electronic-voting counties in Florida produced at least 130,733 and as many as 260,000 "ghost votes" for Bush votes that either weren't cast by voters or were registered for a candidate other than the one intended by the voter.