If you look at the places where he's holding rallies, that's what he seems to be doing. He's choosing places where his message has the potential of near-unanimous agreement.
Totally agree. Plus, if the sample aims to reflect the national popular vote, it misses the point of the electoral college altogether. Other have pointed out that there may be a hidden Trump vote, a large cohort that does not participate in polling. If these various "pockets of support" aren't properly included in the sample, then the sample average will have a baked-in (and consistent, reproducible) error.
It has been discussed for years what the reduction in response rate has on the reliability of polls. PEW research published an article in 2012 showing that their response rate (actually making an interview) was now down to 9% (nine percent). I have seen another article indicating 8% response rate in 2014.
This is a problem for the pollsters since it makes it more expensive for the pollsters to sample a large enough group. But there is another problem: What if there is a difference between those who reply to the poll and those that don't? That could induce a large, an overwhelming bias in a poll.
Some research has gone in to answer that question, and in data published in WaPo comparing accuracy and response rates in polls between 1999 - 2014 it did not appear that the accuracy had declined, despite the fact that the response rate had gone from just below 35% to below 10%.
However, for the polling companies this is like skating on bad ice. Just because it worked yesterday, and an hour ago, it may not work now!
In some previous threads I noted that if there is such a low response rate then a single percent difference in response rate will cause a large difference in the polling result even though there may be an underlying 50/50 split in the preference for the candidates.
In the old days when the response rates were well above 80% then you needed much larger differences in response rates between candidates to skew your poll. Such large differences may have been picked up elsewhere.
So to this election: It is quite obvious that the media bias has been worse than ever before. One has to go back to the election 1980 to find anything close - and that was not an election which the pollsters remember with fondness.
Now, is it very unlikely that people who are thinking of voting for a candidate that is vilified almost universally are more (and remember we are talking of a difference of a few percent only) reluctant to answer a poll?
I think the answer to that question is pretty obvious, and therefore I am much less interested in polls this election than previous ones.