Voting irregularities in three Florida counties that used electronic voting machines in this month's election may have awarded as many as 260,000 votes more to President George W Bush than were expected, according to researchers at the University of California, Berkeley.

The Berkeley researchers claimed on Thursday that their findings raise questions about the accuracy of voting results in Broward, Palm Beach and Miami-Dade counties, all of which have more voters registered as Democrats than Republicans. According to statistical models, voters in those three counties delivered between 130,000 and 260,000 more votes to Bush than were expected by a post-election analysis, the researchers maintain.

"Something went awry with electronic voting in Florida," says Michael Hout, a sociology professor, who led the research effort.

Hout says that the odds of the Florida irregularities happening by chance were less than 1 in 1000, and he calls for an examination of the results. "It's like a smoke alarm and it's beeping," he says. "We call upon the voting officials in Florida to determine whether there's a fire."

The irregularities do not account for enough votes to give the state to Democratic challenger John Kerry, who lost to Bush in Florida by more than 377,000 votes.

The possibility of problems with e-voting was a topic much discussed before the election.

To obtain their results, the Berkeley researchers analysed publicly available voting data from all of Florida's counties using a technique called multiple-regression analysis, which accurately identified butterfly ballot problems in Palm Beach County during the 2000 election, Hout says.

The technique involves building a statistical model to predict voting patterns based on a number of factors, including history of voting, median family income, age, and race. Hout's team conducted its study using data compiled from the November 2 election.

"We noticed that three counties stood out from those expectations," Hout says. "These were counties that had a significant departure from what we would expect, statistically, given the patterns in all those other counties."

Using their statistical model, Hout's team forecast that Bush should have received 28,000 fewer votes in Broward County than he received there in 2000. However, Bush received 51,000 more votes than he did four years ago. In Palm Beach County, where Bush gained 41,000 votes, the Berkeley research suggested a loss of 8900 votes. For Miami-Dade County the research showed Bush should have gained 18,400 votes. In fact, he gained 37,000 votes.

The counties in question used e-voting machines manufactured by Election Systems & Software and Sequoia Voting Systems.

A spokesperson for the Information Technology Association of America, an IT vendor group, dismissed the Berkeley results, saying that the study appeared to ignore the political, social and economic factors that affected the vote. "It is unclear to us that the technology, which is the one factor the authors appear to have focused on for this study, should be viewed as causal above the many other factors that could affect a voter's decision," said Charles Greenwald, an ITAA spokesperson.

Greenwald also criticised the study for not being peer reviewed.

The Berkeley research has already been informally reviewed by academics at Harvard University, and will no doubt be scrutinised now that the results are posted on Berkeley's website, Hout says.

He declines to provide the names of researchers outside of Berkeley who are familiar with the results, saying they asked not to be identified.

Because there is no paper audit trail for the e-voting machines used in Florida, it may be difficult to ultimately explain the irregularities. "Our statistical approach is just about the only way we have to uncover what went on in Florida or in any other state that uses e-voting as it exists today, except Nevada where there is a paper trail," Hout says.

The model found an even larger discrepancy when certain factors weighing the data in Bush's favour were removed, bringing the total possible discrepancy to 260,000 votes, Hout says.

The team did not, however, find this level of irregularity in 12 other Florida counties that used e-voting machines, he says.

Hout is unable to explain why some e-voting counties would experience irregularities while others did not, but he says that the irregularities were more likely to occur in counties that voted for Democratic candidate Al Gore in 2000. "This becomes an important clue that investigators who know something about both the software and the hardware can use," he says.

The Berkeley study also appeared to debunk speculation about voting irregularities in several heavily Democratic counties that voted Republican in the 2004 election. After applying the statistical model to Dixie County and Baker County, both of which bucked party affiliations and voted overwhelmingly for Bush, Hout's team found nothing amiss. These counties, which used paper ballots that were optically scanned, have historically voted Republican in national elections, Hout says.

Hout's researchers also examined the election results in the hotly contested state of Ohio and found no irregularities there. "Our results do indicate that Ohio probably did get it right," Hout says.