Sampling the Latino Vote

[phantom power]

I think the moral of this story is, polls can definitely be wrong, but if you don't have a very clear story about why and how they are wrong, it's safer to assume they are correct.

I am really skeptical of any analysis that is predicated on the polls being off in one direction or another. In my experience, the average of polls is rarely off by a whole lot. Maybe a couple of points, but when the polls say you are going to lose, you are almost definitely going to lose. In 2004, we tried to console ourselves that the pollsters weren't picking up enough voters who only used cell phones. That was probably true, but we still lost and the polls still pretty much predicted the result. So, when I see someone arguing that the pollsters aren't surveying enough Latinos or the right kind of Latinos, I immediately have a negative reaction. However, one example where pollsters were badly wrong was in predicting Harry Reid's demise in 2010. Using a very similar methodology to what he is using now, Nate Silver gave Reid less than a one-in-six chance of winning reelection.

The presumption that Reid was unelectable rested on a series of public polls, nearly all of which showed Reid behind Angle. Indeed, in October alone the Nevada press reported on 14 surveys, only one of which showed Leader Reid ahead.

It turns out that those polls made three major errors. They did a poor job of separating likely from unlikely voters, they didn't work hard enough to poll hard-to-reach people, and some of them were precluded by law from contacting cell phones. All three of these errors featured prominently in a major underestimation of the Latino vote. It turns out that people who were resistant to answer phone surveys on the first or second attempt skewed heavily for Reid. And it turned out that Spanish speakers were most likely to use cell phones exclusively and the least likely to respond to phone surveys. And it turned out that Latinos were more likely to vote in general than the pollsters had assumed. As a result, the pollsters missed the result badly and Harry Reid won by six points.

Even within the Latino sample, the pollsters missed something. Naturalized Latinos supported Reid more heavily than native-born Latinos, and they were harder to reach because of a stronger language barrier. This led the pollsters to underestimate how strongly Reid would perform with the Latino community even as they were underestimating their percentage of the electorate.

How many of these mistakes have been addressed by the polling firms in the field right now?

http://www.boomantribune.com/story/2012/10/26/111542/53