I think it's quite likely a lot of people are lying to pollsters, being ashamed of voting Trump but intending to vote Trump anyway
yeah it's called the Bradley Effect (or in the UK, the Shy Tory factor). by its very nature it's hard to prove to what extent it exists.
so is nonresponse bias, i.e. Trump voters being less likely to respond to polling at all for whatever reasons, and so underreported
so both probably factors in inaccuracy of 2016 but no one knows to what extent
the main thing is that the polls would have to much more inaccurate than they were in 2016 for a Trump win, or as 538 puts it
Trump Can Still Win, But The Polls Would Have To Be Off By Way More Than In 2016
polls in 2016 actually weren't
that inaccurate. national polls were basically right, Hillary did win the popular vote. state polls were the issue, and they still weren't terrible. what Trump did was outperform them by a few points in a few key states to take the Electoral College win.
there's obviously no way of knowing if pollsters have improved their modeling - through attaching more importance to education in targeting likely voters or whatever - until after the election but you can at least say they'd have to be not only as wrong as last time, but much more wrong.