A good article appeared this morning on the discrepancy between hand counted and machine counted votes in New Hampshire. The author, Beverley Wang, does a decent job of covering the statistical issues involved. The gist of the story is exactly what my post below discusses, that the use of machines to count votes is not a random event but is in fact correlated with underlying demographics that themselves determine the vote.
My colleage John R. Lott Jr. has a paper, joint with Kevin Hassett, titled "Voting Technology and Voter Fraud: A Test Using Exit Poll Data." In this paper, the authors use exit polling data to show that exit polls had a similar pattern between machine counts and hand counts, using national 2004 election data. Since the technology to count final votes could not have influenced exit polls, this is pretty conclusive statistical evidence that the counting technology was not behind voting patterns.
I still would like to see if additional demographic variables in my vote prediction model would reduce the significance of the machine count variable. If I can get the data, I will run the models and report them.
Of course, since there will be a recount, any claims of irregularities will be quickly dispensed with.