Silver said most of the major network polls — NBC/Wall Street Journal, ABC/Washington Post, CBS/New York Times — are “pretty good," while the polls that faltered made too many assumptions about the population.
I think that high quality polling really differentiates itself now because if you take an automated poll, you miss people who have cell phones, which is about a third of the population now. And they're mostly younger urban demographic, mostly Democrats, so you will undersample Democrats, to use that buzzword, if you don't call people who have cell phones. And lo and behold, those polls had a Republican bias this year. Not because the pollsters are evil partisan, but because, hey, if you miss a big chunk of the population that's Democratic-leaning, you're going to have problems…
The best pollsters, Silver said, “let the sample tell you what it is by themselves.” The ones who didn’t “put their finger on the scale” and didn’t make assumptions about the voting public were most successful.
Just going with what the data said instead of making assumptions is usually the best practice whenever you're doing any kind of scientific survey and that worked again this year…. I’m a pro-horserace guy. I’m more interested in diagnosing 2016 now than Benghazi, for example, because that’s where my bread is buttered. But if you’re going to do horserace, then do it the right way because it can be more data driven.