A few notes on my piece in 10/21/2012 New York Times.
1. The "recent estimate" on misreporting voting cited is Rogers and Aida Why Bother Asking. The 2/3 figure comes from using Bayes' Law from Table 2. Also see Ansolabehere and Hersh Pants on Fire, which finds that 64.2 percent of non-registered voters report that they are registered.
2. Most of the Google numbers are publicly available from either Adwords, Google Correlate, or Google Trends. However, some of them I had to use the methodology I described in my racial animus/voting paper because of some data restrictions. In particular, I had to do this for "Paul Ryan shirtless," "Willard Mitt Romney," and daily "McCain life expectancy" data.
3. The methodology for predicting voting is using change in search volume for "vote" or "voting," compared to four years earlier.
I will aim to post a full paper within a couple days -- and updates as more data arrive. I will do both here. I would love to hear if anyone has comments/suggestions/criticisms. I can be reached at firstname.lastname@example.org.