Revisiting the 2004 election, part 8, Florida again

“Something went awry with the electronic voting in Florida”

Several organizations, groups of investigators and individual researchers were intrigued by the results of the 2004 election in Florida. Among them was a research team at Berkeley (U. Calif.).

The research team, led by Professor Michael Hout, announced its study of the results of the Florida election in a press conference Nov. 18, 2004. In short form:

“A research team at UC Berkeley will report that irregularities associated with electronic voting machines may have awarded 130,000 – 260,000 or more excess votes to President George W. Bush in Florida in the 2004 presidential election. The study shows an unexplained discrepancy between votes for President Bush in counties where electronic voting machines were used versus counties using traditional voting methods. Discrepancies this large or larger rarely arise by chance–the probability is less than 0.1 percent.”

Prof. Hout, a sociologist, walked listeners through the study in a short talk followed by questions.

The research team used quantitative methods. Salient points:

• Its released report showed 130K-260K excess votes to GWBush in Florida in counties where electronic voting machines were used, an unexplained discrepancy. This could occur by chance, Dr. Hout stated, less than 1 time in 1000.
• The discrepancy was most prevalent in the more heavily Democratic counties—Broward, Palm Beach, and Miami-Dade, in order of magnitude.
• Voting machines tallied a net gain of 81K votes for President Bush; 41K in Palm Beach; 37K in Miami-Dade.

U-Cal Berkeley

Dr. Hout reiterated that the discrepancy could not be explained by other factors, and there was virtually no chance of its arising by chance. “For the sake of future elections, officials should explore this.”

The genesis of the project was that after the election, two grad students shared their frustration at having so many rumors, re the outcome of the election, with a lack of statistical basis. Key concern was to start with a model that accurately predicted votes or vote breakdown; to knock down false claims. So they decided to start with measuring change between 2000 and 2004, and they downloaded data from Florida and Ohio. The results showed that something odd was happening. Further tests showed that “the anomalous effect of this e-vote held up. It couldn’t be made to go away.”

“The students and I went back and forth several times with a number of revisions,” Hout said, and “results held up.” They were also run by a number of different faculty members, who concluded that “something went awry with the electronic voting in Florida.”

Organizers of the press conference hoped to draw the authorities–“somebody”– “to examine the results in Florida.”

As he commented, “a statistical approach is just about the only way to find out if anything goes wrong with e-voting, except in Nevada which has a paper trail.”

Walking the audience through some of the results, Hout concisely explained that a multiple regression analysis was used in the study. Statistical technique was used to assign different weights to different variables, for example the Bush vote in 2004 compared to the Bush vote in 2000; raw percentages in 2004 compared to 2000. Researchers assigned weight to a number of reasonable factors or not-so-reasonable factors. Expectations were taken into account.

Three counties stood out–Broward, Palm Beach, and Miami-Dade.

Three metro counties in Florida

All three counties produced “a different pattern from what we would expect” given factors including their median family income, history of voting Republican, markers of age structure (senior populations), and racial ancestry.  Not that Metro Miami went for GWBush by large margins. But the popular vote showed some oddities.

With all the other variables statistically controlled, the e-vote variable continued to show a significant relationship with both change in votes 2000-2004 and the absolute level of support for Bush in 2004.

“Inability of all other factors to explain away this discrepancy,” Hout said, “has us convinced of an e-vote anomaly” of 1,018,010 compared to 1.16M votes actually tallied for Bush. “130,000 to 260,000 is our best estimate of the number.”

“How this happened beyond our ability to observe.” Researchers left it to Florida officials to explain.

Didn’t happen, of course.

The formal study in final form was published on the Berkeley web site, where it remains.

A short round-up of similar or related material on Florida is accessible here.

Side note:

This year I finally read Timothy Crouse’s book on the 1972 election and the press, The Boys on the Bus. Should have read it earlier. Ranging back and forth between the elections of 2000 or 2004 and this year’s primary season is illuminating, but going back to the obsession with Who-will-win in 1972 is even more so. How did it happen that hordes of hard-working (often) reporters could miss so much about Nixon’s plans to consolidate control?  How could the reporters fail to report the Nixon team’s ongoing control of the press?

Same way it happens now, apparently, with days of air time devoted to speculation over who will win—and perhaps one one-hundredth of that time devoted to cogent, accurate analysis of what GOP candidates’ policies would actually do to the U.S.

Needless to say, even less than one-hundredth of the coverage time goes to problems with voting technology. Little goes to vote-suppression efforts, either, although at least the open efforts to suppress the vote in a number of states have been mentioned, mostly on cable.

Now as ever, Bev Harris’s team continues to do good work on this crucial issue at Black Box Voting.