Let’s look at some figures, which, since this is a blog post and not a school paper, I got from the Wikipedia 
2004 Bush votes: 62,040,610
2004 Kerry votes: 59,028,444
2008 McCain votes: 59,948,323
2008 Obama votes: 69,498,516
2012 Romney votes: 60,933,504
2012 Obama votes: 65,915,795
2016 Trump votes: 60,350,241
2016 Clinton votes: 60,981,118 
(Since there’s a right-wing theory floating around that Trump got more popular votes, I include a reference to the 2016 voter figures)
Point being, Republicans have a pretty stable turnout model, but Democrat turnout fluctuates a lot more.
Now, the problem with polls is that most people, these days, plain simply do not answer the phone and take the time to answer a poll — people have been inundated by telemarketers to the point they just don’t answer the phone if they don’t know the number calling them — and, with the entire culture of free, there is less money to do a proper poll.
So, when someone does answer the phone (or take the online survey), they then ask questions about gender, race, ethnicity, age, and likelihood of voting.
Since different races and ages have different rates of answering the phone, and since said groups have different rates of voting (with some groups being reliably Democratic, and other groups being reliably Republican), what the pollsters then do is skew their numbers so that their sample is (hopefully) the same as the number of people who go out to vote.
However, if you look at the figures, Democratic turnout is highly violate. If a polling firm uses, say, 2012 voter turnout models to determine who is going to vote in 2016, they will predict the kind of decisive loss Romney had in 2012.
The issue here is not that Trump energized some base of voters; the issue here is that reliable Republican voters held their nose and voted for Trump, while 10%-20% of Democrats stayed home instead of voting for Hillary.
When pollsters called me, the first questions were screening questions. One time, I said I was fairly (but not very) likely to vote; that resulted in them ending the call. If this is how polls are run, they are not getting very good demographic information about people who stay home instead of vote.
It’s the Democrats staying home which gave the election to Trump. Pollsters did not get good information about who was staying home, and that is why they saw Hillary having the same kind of victory Obama had in 2012.
December 2016 update: The current numbers are 62,800,198 Trump — 65,443,689 Clinton. This in mind, the issue is not only the Democrats who stayed home (but that’s a factor; if Clinton had gotten as many votes as Obama did in 2008, she would have won) but also the fact that Trump was able to get about two million more voters than Romney and McCain did; he energized conservative voters in key swing states that previous candidates did not — voters who the pollsters probably did not account for.
To post a comment about this blog entry, send me an email. I may or may not post it, with or without editing.