I think a common mistake by people, including many liberals, who believe the polls were way off in 2016, is that people think polls predict winners and losers. They do not. They provide data from which others make predictions.
Consider a hypothetical poll which says D+1 where the outcome turns out to be R+1. The "poll" got the winner and loser wrong, right? Shit poll! No, the polling error was a margin of 2, which is OK. In another hypothetical, the poll says D+9 and the result is D+2. Everyone thinks this poll was great because it accurately "predicted" the winner. But in reality, the error margin there is 7, and polling was much worse than in the first case.
It's why no one seems to realize that polling in 2012 was actually a little more off than polling in 2016, because they think the 2012 polls accurately "predicted" the winner, but in reality, they showed Obama with a much smaller lead than he actually had.
I think lots of people just do not get this.