Nate Silver’s Lessons for Big Data from the Unpredicted Trump Victory

Individual state polls had more variance than the national polls, but those were wrong in both directions, both over and understating Trump's support, mostly depending on the number of less educated white people in the state, who were apparently hard to poll, or to estimate their likelihood to actually vote.

Both you and the other poster seem to be trying to create a narrative that Trump's popular support was underestimated, when really it's just the interaction of where his support was and the electoral college system that fooled most models, though notably not the 538 model, which gave Trump a decent shot at winning. Btw, you almost certainly don't know what an "oversample" is, you should read up on that, as you use the phrase as if it is some devious way to skew the polls in one direction. Actually, while I'm on the topic, you seem to have got the wrong end of the stick about what a push poll is too. Those are devious, but do the really apply here. Just a Wikipedia summary should be enough to set you on the right path.