Thursday, November 18, 2004
The Berkeley Study on Florida Voting in 2004
Michael Hout, Laura Mangels, Jennifer Carlson and Rachel Best are the authors of a new study on the Florida presidential elections in 2004. The authors gave a press conference today, and their article is available online. I have read through it once, so my comments here will have to be regarded as preliminary. In particular, my brain has a delay button on critical thinking, so I probably come up with more questions later on. But I hope I have enough to explain what this study argues.
The authors use a statistical model that tries to explain why the support for Bush might change in the 2004 elections as compared to the 2000 elections. Possible factors affecting this change in support are historical voting patterns in an area (such as past support for Republicans, including Bush), voter turnout and changes in it, size of area, its income and the ethnic composition of its population.
In addition to these factors, the authors included variables that measure the use of electronic voting machines. Some areas don't have them, some do. If electronic voting has no impact, we'd expect the factors in the preceding paragraph to explain why some areas have more growth in Bush votes than others. If electronic voting turns out to have an impact, over and above the other factors, then an audit is indicated.
The results indicate that electronic voting did matter. Translated into real numbers, the authors argue that Bush might have gained 130,733 extra votes from this, possibly as many as twice this amount if the votes should have been assigned to Kerry instead.
Interesting. Note that this is not about those precincts where Dixiecrats voted overwhelmingly Republican. The voting machine effect was especially large in Broward, Palm Beach and Miami-Dade; all Democratic areas of the state. A similar study done in Ohio didn't show a voting machine effect at all, although the authors didn't have all the data they used in the Florida study on factors other than voting machines.
If I wanted to subject this study for criticism I'd ask what other factor might account for the voting machine effect. Is there something that was not included in the list of factors in this study but which should have been included? This variable would have to correlate with the presence or absence of electronic voting machines.