A strange thing happened to pollsters in 2010. They got 2 races pretty wrong, Harry Reid and Michael Bennet. Why? Adam Serwer over at Mother Jones makes a good argument that Latino voters don't respond to polls very well and can make a big difference in the results of an election.
One obvious reason for this, one that I've heard in other places, is the language barrier. Some pollsters don't poll using Spanish, especially the robo call polls. Serwer goes into some of the other reasons, below the fold.
... Polls in Colorado and Nevada make the states look like anything but sure bets for the president, but there's good news for Obama: By undercounting Democratic-leaning Latinos, those polls could be dead wrong.
...
Polls showed Reid trailing his Republican opponent, Sharron Angle, by an average of nearly 3 points.
...
Then something weird happened. Reid won—by almost 6 points. In Colorado, another state with a large Latino population, Democratic Senate candidate Michael Bennet eked out a 1-point win despite polls showing his GOP rival, Ken Buck, up by an average of about 3 points.
...
Latino voters, undersampled by pollsters and written off as unlikely voters, had made a huge difference for Democrats.
...
According to Matt Barreto of Latino Decisions, a polling firm that specializes in public opinion surveys of the Latino community, ll sample sizes of Latinos, ineffective or inadequate efforts to reach "Spanish-dominant" Latino voters, and failing to account for how many Latino voters would turn out...
Barreto argues that the polls often miss working-class Latinos, who are more likely to use cellphones, work long hours, and prefer to speak Spanish. Instead, the polls are more likely to reach more financially comfortable, "English-dominant" Latinos who are more likely to vote Republican. Robopolls, in which an automated prompt calls voters, tend to be particularly bad at measuring Latino public opinion, even when they have a Spanish-language option, Barreto says.
...
Barreto says it's happening again this year. "Pollsters who are missing that component of the correct proportion of Spanish interviews, they are completely underestimating a growing part of the electorate, and this is the part that is most heavily Democratic," he argues. "They're are operating on models that are at best 20 years old."
...
Why does this matter? Even if Romney wins Virginia and North Carolina, Barack Obama only needs 33 of the the remaining 82 swing state electoral votes to win, while Romney needs 51. The race is so close that a shift of a few points among Latinos in states with large Latino voter populations like Colorado, Nevada, and Florida could determine who gets those crucial remaining electoral votes.
...
Latino Decisions' polling has Obama poised to win more than 70 percent of the Latino vote (other surveys show similar numbers), but in several public polls of swing states, Barreto says, the Latino sample sizes are so small or so unrepresentative that they're likely overestimating how many Latinos are going to vote Romney.
...
I usually roll my eyes when I hear people talking about under sampling young voters because of cell phones, or Democrats on weekends... but I think there is some good logic in this article. It's a matter of degree of course, and the higher the Latino population in a state, the bigger the effect will be.
I'm sure the Obama campaign is all over this, so lets help make this analysis right.