Even the good polls wiff sometimes. It could be perfectly accurate in how the data is processed and analysed, but the sample could be skewed in some way. Not match the electorate. Older, younger, whiter, blacker, etc. than the general population. Or should just by chance you could happened to knock into a block of people that happen to be voting differently than the general population. Or the model could be borked. You’re obviously working on a smaller than everyone in the country sample size. So that data has to be analysed by fitting it to a model of the electorate/general population. If you include more of a particular demographic than actually exists, base your model on bad break downs of the country, forget to carry a 2, etc. Your good data, however accurate or predictive it might have been, won’t match reality. All of this can be exacerbated by small sample sizes or collection problems.
IIRC the major polling problems for '08-'12 were down to a few of these factors. Both the general population and the electorate had major demographic shifts faster than expected. Basically old white men are less of a proportion of the country and voters than they used to be, to more of a degree than those polls accounted for. Basing your model of voter turn out and the assumed electorate on the 2004 polls (and then 2008 polls for '12 and '14 midterms) didn’t give an accurate read on who would actually be voting where. That introduced a drift between certain polls and actual voting. And IIRC from around 2004 to at least 2012 a lot of your major polls failed to adapt in certain ways. Specifically they refused to utilize or properly weight calling to cell phones and polling by online tools. That had the effect of excluding disproportionate numbers of young, non-white, urban and poor voters. All of whom are much less likely to have a land line than old wealthy whites. This basically just magnified one of the long standing issues with polling here. Old white people are way more likely to actually respond to polls. Effectively retirees in rural, mostly white districts have nothing better to do than talk to that random dude who just called him. That always skews thing to one extent or another, so it needs to be accounted for in analyzing the data. The last few go rounds it skewed things a lot more than anyone realized.
All of which is to say that individual polls = bad. Even the good ones. They can have a bad day, or miss something entirely. IIRC the Gallop polls from 2012 were pretty on the ball. Except for a handful of late in the game releases that missed the mark rather badly, and they were sort of the only major poll that did.
The Gallop poll is generally a good one (as are most of their polls on nearly every subject). But traditionally the Quinipiac polls is where the geeks look. Smaller sample size, and less granular in their questions. Pre-Obama it was far more often to call things right, and its often considered more predictive. Though I don’t necessarily think that’s the case (I don’t like to think of any individual poll as predictive). That doesn’t seem to have held for the last few elections though. Better are polling averages and demographic analysis (especially those based on primaries and past votes, actual voter data rather than questions about what will happen later).
For polls Real Clear Politics seems to be the go to polling average, and they have a fun little tool for digging around and looking at various situations. Though I’ve seen a fair lot of references to Huffington Posts polling averages this year. For analysis of those polling averages, and good break downs on the demos Five Thirty Eight is the usual go too.
For the change in tenor in Trumparitaville? As of Thursday things were largely the same. Yeah Trump! Oh he bankrupted a bunch of small business men? He’ll do the same to Iran! That guy is a shark I tells ya!
The change is starting Friday and moving through the weekend. Curiously after Trumps speech. I’ve seen a lot more of these people scowling at their candidate when he crops up on the TV. And only hear mocking references to his speech. Its been fairly interesting to me that the vast bulk of the people in question have never voted before. Even as they spend huge amounts of time regurgitating Fox news and circle jerking each other over conservative values and who’s a real proper conservative American. The sudden dip in enthusiasm, if its actually an indication of a change and not the heat wave, could point to those people staying non-voters. I am in NY though, albiet in a pretty reliably Republican area. I don’t know quite how to interpret what I’m seeing on the ground. Particularly given recent trends in the polls and news.