Trump underperformed his polls, Clinton generally over-performed (MI being the obvious exception) - outside of the bizarre victim mentality of Trump supporters (the irony of which is sad and adorable), I see NO evidence of a genuine shy Trump effect, there is however, evidence of a shy Clinton effect.
IIRC, Trump basically matched his polling #s in the primaries, while his GOP opponents tended to beat their numbers. Which I’m assuming is because undecideds broke towards Cruz/Kasich/Rubio/etc. And that makes sense, as supporters of the other candidates tended to have a negative opinion of Trump. So there could be lots of voters undecided between Cruz and Rubio or Kasich and Rubio, or some other combination. But few were undecided between Trump and another candidate. Thus he would obviously pick up few undecideds. Whereas in the Dem. race, since it was just a 2-person contest, any undecideds were undecided between Clinton and Sanders. So Clinton picking up some of those folks is not unexpected.
But that’s not a “Shy” effect either. Undecideds breaking for one candidate at the end isn’t a “Shy” effect. As I said, the idea that the main source of polling error is people being reluctant to tell pollsters who they really support is weird to me. I don’t know why this meme is so popular here. There are plenty of reasons why polls might underestimate or overestimate a candidate’s support, and I don’t see why people are convinced that voters lying to pollsters would be that significant an effect compared to other reasons.
EDIT: I guess I can see the argument some have made that
if there is a real "shy effect" for either candidate (and I'm not convinced there is), then it would be more of a problem for live phone polls than robopolls and internet polls, because people too embarrassed to admit their true voting intentions would be more likely to lie to a person than a machine. So OK, if robopolls end up being a lot more accurate than live phone polls, then I'll concede that there's at least a little evidence for a shy effect, as opposed to polling error being present for other reasons. But I'm not counting on that happening. Live phone polls are (in aggregate) more reliable, and I have no reason to think it'll be different this time.