by billg » Tue 15 Jan 2008, 22:44:25
"The Diebold Effect": Hillary's Votes Higher From Diebold Machines
Even Controlling for Demographics (education, income, population, etc)
Posted on: January 15, 2008 5:42 PM, by Chris Chatham
$this->bbcode_second_pass_quote('', '
')In contrast to exit polls, the final vote tally from the NH democratic primary shows a surprise victory for Hillary Clinton. People quickly noticed an anomaly in the voting tallies which seemed to show an advantage to Hillary conferred by the use of Diebold machines.
However, there was an easy explanation: towns with Diebold machines are more urban on average, and Hillary was always thought to have more support in urban areas. So, like many others, I was supremely irritated by the lack of analyses which statistically controlled for this obvious factor.
So I got a copy of the vote counts, and thanks to Brian London at BlackBoxVoting, the demographic information from each town (most notably, the % holding bachelor's degrees, the median household income, and the total town population). Now, Mark LaBonte at BlackBoxVoting has provided estimates of the mileage for each district, allowing for the calculation of population density.
To my complete (and continuing) amazement, the "diebold effect" on Hillary's votes remains after controlling for any and all of those demographic variables, with a p-value of <.001: that is, there are less than 1:1000 odds for this difference occurring through chance alone, and that's after adjusting for variability in Hillary's votes due to education, income, total population, and population density.
While this "diebold effect" varies in magnitude depending on the exact covariates used, it seems to center around an additional 5.2% of votes going for Clinton from Diebold machines. The same analysis shows a Diebold disadvantage for Obama of about -4.2%, significant with a p<.001, using the same covariates.
Due to the cooperative grassroots nature of this effort, I cannot guarantee the accuracy of the data file - the information has come from a variety of sources and I won't claim to have verified it all. Furthermore, I'm not a statistician - I'm waiting on the Social Science Statistics Blog (Harvard) and the Statistical Modeling Blog (Columbia) to weigh in. However, my analysis seems in line with this paper about the 2004 NH democratic primary.
http://scienceblogs.com/developingintel ... rys_vo.php