Cross-posted at Latino Decisions, Latino Vote Matters, and Daily Kos:
Final LD Vote and Turnout Predict Projections: Record-breaking Turnout and Lopsided Support for Obama among Latinos
Over the final several weeks of polling (Figure 1), the LD Vote Predict estimates for presidential vote choice have been quite stable, leading us to our final projection for a breakdown among actual Latino voters who choose one of the two main candidates and no undecided voters:
- Obama 73%, Romney 27%
Uncertainty in model parameter estimation and polling means that Obama is very likely to best Romney by somewhere between 40 and 52 points, nearing or surpassing the record proportion of Latino vote for a Democratic presidential candidate, set in 1996, when 72% of Latinos backed Bill Clinton. Given the 95% bayesian interval of +/- 2.6, this means Obama will win between 70.3 – 75.5 of the Latino vote.
Our final projection of turnout among Latino registered voters (Figure 2) is 73.2% (+/–3.6%). By our own assessment, this would narrowly improve upon the record established in 2008. Estimates of turnout percentage are somewhat difficult to calculate. Census numbers are based on self-reported voting, and tend to overestimate validated turnout by between 12% and 20%.
So, for example, the U.S. Census reports 84% of Latino registered voters claiming to have voted in 2008, but the true figure is likely in the neighborhood 70%. Going on self-report alone, we would anticipate around 85–88% of Latinos turning out to vote this year. A wildcard, however, is the number of newly registered voters; these individuals have extremely high turnout, especially as many states allow one-stop registration and voting. Thus, if registration drives among Latinos are more successful than in past years, the turnout numbers could be higher than predicted.
The prediction models developed here are discussed in much further detail in my original blog post on September 6th, and for turnout predictions here. The models take advantage of the full weight of 11-weeks worth of national polling data, and create a best fit a statistical model. My estimate from October 4th is posted here.
