- (some) polls were systematically off, mostly because they underestimated white non-college turnout and overestimated black turnout. Note that poll errors were actually lower than for any other election–only this time they actually got the winner wrong.
- Most of the poll aggregators operated under the assumption of statistical independence of state-level results. This was a serious mistake, and both Huffpost as well as The Upshot (to a lesser extend) should be somewhat ashamed. 538, by the way, did this right and had Trump at 3X% on election day.
- There actually was some significant change in voters' intent in the last days, possibly fuelled by all the news stories the FBI directors was generating. It's highly likely that an election two weeks earlier or two weeks later would have ended differently.
It's also important to note that newspapers aren't really in the business of predicting the future. If they "relied too much on social media" for any facts they reported, I'd love to see some examples.