According to PEW, (no time to dig up the references now) the response rate to polls was 9% in 2012, and 8% in 2014. That is more than 90% of people called refuse to answer or take part in the poll.
Some research was published in the WaPo, showing no increase error rate in polls from 1998 to 2012 (iirc) despite the response rate going from +30% to less than 10%. That may well be true, but the pollsters are skating on very thin ice. With such a small proportion of the population actually taking part in polls, there needs to be only a very small difference between the responding and the non-responding cohort for the polls to be totally skewed.
PEW published a recent experiment (a FR-thread a few days ago) where they had asked republican voters in three different ways if they were voting for Trump. It was clear that a total anonymous internet poll gave larger response for Trump than an automated phone call, which was more positive than a personal phone call.
Interestingly enough if the same question was asked about Jebb Bush there was no difference between the different ways of polling. (!!!!!!)
I believe this makes the polls almost completely useless this time around.
The best succinct summary of the technological issues that have developed in polling with modern methods of communication screening and attitudinal changes related to them.
I wish your commentary here was required to be posted on every poll thread (good or bad).