Low Marks For Bad Stats
A UC Berkeley study by sociology prof Michael Hout and his grad students showing anomalies with Florida e-voting used a faulty equation, reports Kim Zetter in Wired:
The study, released three weeks ago by seven graduate students from the University of California, Berkeley's Quantitative Methods Research Team and sociology professor Michael Hout, presented analysis showing a discrepancy in the number of votes Bush received in counties that used touch-screen voting machines versus counties that used other types of voting equipment.But Bruce McCullough, a decisions science professor at Drexel University in Philadelphia, and Binghamton University economics professor Florenz Plassmann released an analysis (.pdf) of the Berkeley report criticizing the results.
According to the Berkeley study, the number of votes granted to Bush in touch-screen counties far exceeded expectation, given a number of variables -- including the number of votes those counties gave Bush in 2000 -- while counties using other types of voting equipment gave Bush a predictable number of votes.
The analysis was not peer-reviewed, although Hout and the students said that seven professors examined their numbers. They would not speculate about what occurred with the voting machines, but voting activists on internet forums seized the study as proof of faulty voting machines or election fraud. Drexel University's McCullough, however, found fault with the study.
"What they did with their model is wrong, and their results are flawed," McCullough said. "They claim those results have some meaning, but I don't know how they can do that."
McCullough said they focused on one statistical model to conduct their analysis while ignoring other statistical models that would have produced opposite results.
"They either overlooked or did not bother to find a much better-fitting (statistical) regression model that showed that e-voting didn't account (for the voting anomalies)," McCullough said.
Charles Stewart, an MIT political science professor, called the study "the type of exercise that you do in a graduate data-analysis class" rather than as an academic paper.
"If I were to get this article as (an academic) reviewer, I would turn it around and say they were fishing to find a result," Stewart said. "I know of no theory or no prior set of intuitions that would have led me to run the analysis they ran."
Some red faces up there in Birkenstockville, I think.
Arguably the finest public artwork in LA county: It *offends* rubes.
Cridland at December 10, 2004 4:47 AM
I shall not jump to conclusions
I shall not jump to conclusions
I shall not jump to conclusions
I shall not jump to conclusions
(with all those people sharing the same intelligence that leaves very little intelligence per capita)
viktor at December 10, 2004 4:16 PM