Still more on the Berkeley study of Florida voting. I want to point to some of the other interesting work on this and to thank people who've written nice things about my posts.
A useful place to start would be this page from Jonathan Wand, assistant professor of political science, Stanford. Professor Wand has developed a "robust method for estimating overdispersed multinomial models of count data" and using it, he finds no evidence that electronic voting spuriously increased votes for Bush.
Professor Wand also links to the VerifiedVoting.org page that has amazing data on the voting method used in virtually every county in the U.S. (Oh, the graduate theses that can be written! Oh, the conspiracy theories that can be constructed with enough patience and econometrics software!)
Professor Michael McDonald, George Mason University presents a pointed attack on the study. Of particular interest is a graph of the data. Professor McDonald discussed the study with Professor Hout and one of the graduate-student co-authors on a Berkeley radio station. An audio file of the discussion is here. (Great fun: listen as two academics try to explain, and debate, including second-order interaction terms in a regression model while the host quips, "You've just lost 98% of our listeners.")
Patrick Ruffini looks at precinct data from Palm Beach County and estimates that "President Bush received a 14-16% swing among Jewish retirees specifically." That plays well with my hypothesis about Jewish voters accounting for at least some of the "anomaly" in the three South Florida counties.
Two comments on an earlier Patrick Ruffini post--thanks for the nice comment, Patrick--make this interesting point: the Berkeley equation suggests either that Bush's 2004 vote in Palm Beach, Broward, and Miami-Dade counties was too high OR that Bush's 2000 vote in these counties was too low. The local governments of these counties are heavily Democratic. So perhaps some of Bush's 2000 votes were stolen, but the introduction of electronic voting this year simply made stealing more difficult.
I don't believe this, but it sure is a cute way to illustrate the limitations of the Berkeley analysis.
Other interesting and useful analysis and/or comments:
Richard Baehr posted an article online on October 17 that discusses the importance of the Jewish vote in these three counties. He employs the Jewish vote as just one avenue of attack in this week's sharply critical review of the Berkeley study.
Brendan Nyhan, graduate student in political science at Duke, has two posts on the subject. The first points to a number of other studies and comments. The second links to me. Thanks, Brendan.
Alex Stashny, Ph.D. student in Social Science at U. Cal Irvine--thanks for the link, Alex--constructs a different model of voting behavior and it finds no anomaly related to electronic voting.
Mickey Kaus's buddy, the Mystery Pollster. Thanks for the link.
Kieran Healy of Crooked Timber.
King at SCSU Scholars. Thanks for the link.
The Econoclast pays me a handsome compliment. Thanks, John.
One last--maybe--word. Among the other factors Professor Hout, et. al. don't pay sufficient attention to is the change in local economic conditions. (They do look at median income as a regressor, but that is likely to be a poor choice for predicting the change in % voting for Bush, their dependent variable.) Given that, a recent report by the Milken Institute is interesting. It constructs an index for the largest 200 metropolitan areas that ". . . ranks U.S. metros based on their ability to create and sustain jobs. It includes both long-term (five years) and short-term (one year) measurements of employment and salary growth." In 2003, Fort Lauderdale--in Broward County, one of the three "anomalous" counties--was 29th; in 2004, it ranked 9th. Palm Beach County--another one of the three anomalous counties--ranked 4th in 2003 and 4th again in 2004. I can't comment on the validity of the index, but if it's even partially accurate, might the significantly above average recent economic performance in these two counties account for a shift in votes toward the incumbent, particularly a Republican incumbent?