Polls are able to capture what is often called a "snap shot" of a political race, providing a good assessment of where the contest is at a particular moment in time. But they don't capture everything, as an illuminating discussion by the analysts at FiveThirtyEight.com reveals:
Hypothesis #1: The polls are underestimating Clinton because they don’t factor in her superior ground game.
Most reports (and we’ll have an article with some extensive data on this soon) suggest that Clinton is far better organized than Trump, with more field offices, for example, and a better get-out-the-vote (GOTV) operation. Isn’t that reason to think the polls are underestimating Clinton by a bit?
natesilver: That’s one of the better arguments, yeah.
harry (Harry Enten, senior political writer): Well, the question you have to ask yourself is whether or not the field game is being picked up by the polls. In a wonderful, perfect universe, it would be. In reality, I have my doubts. That’s why I’m not entirely sold on polls showing a large gap (in Trump’s favor) between registered voter results and likely voter results.
natesilver: The empirical literature on the ground game is messy because there are so many conflating variables. But what would worry me if I were Trump is that the race isn’t that close if everyone turns out — and I have the worse turnout operation.
micah: How would the polls pick up on a better ground game?
harry: A ground game is supposed to get people to vote. So, if Clinton’s campaign is getting more people engaged, then polls that are aimed at identifying likely voters should pick up on that. More of those voters should make it through the likely voter screen.
natesilver: Well, maybe they’ll make it through the likely voter screen, but a lot of likely voter screens rely on past voting history, and may or may not have adequate provisions in place to capture new voters.
There's more insightful analysis, and anyone not convinced the polls are telling the whole story should check out the full article.