New prediction: Trump's poll average falls to 12 points behind Clinton in a week's time, then slowly claws back until he loses by 8 or 9 percentage points. Clinton's number of EVs falls between 346 and 399.
Okay, that didn't happen. I should have stuck with my "Trump Cycle" theory rather than expect that the Access Hollywood tapes/debate losses would have done anything other than crash him briefly and then allow him to claw back so that he'd be at the top of the cycle again by November 2 and quite possibly Election Day. Here's a new set of predictions. They're mostly just hunches.
1. Non-Cuban Hispanic turnout increases more than expected*.
2. Non-college-educated white turnout is higher than expected.
3. Black turnout is lower than expected.
4. Cubans don't shift enough relative to higher NCEW turnout to make Florida more Democratic-leaning than the nation as a whole.
5. Some sort of "shy Trumpist" effect appears to the tune of 4 or 5 ppt among college-educated whites, but is statistically zero among any other demographic.
The expected net effect would be a nail biter of an election. The most likely outcome based on current polling would be the following map**, with Clinton eking out a narrow win by carrying her 272 EV "inner fortress" and Nevada being unpredictable between #1 and #2***:
*I just realized "expected" is a weasel word, so I'll lay out a couple of different versions of "expected" for predictions 1-3 and my level of confidence relative to each. I am fairly confident that these will be true relative to the 2012 election. I am very confident that these will be true as a proportion of all voters, adjusted for total turnout, relative to the 2012 election. I am much less confident that these will be true relative to most current "likely voter" models, which I don't even know the assumptions of. I'll call the latter the "strong form" of my prediction and stick with it anyway.
**Note that all the predictions except this map are in terms of the turnout rates and margins relative to expectations of the nation as a whole. If Trump shoots himself in the foot at the last moment, has a November Surprise, or if the race just turns out to be a more lopsided win for Clinton, this map would shift without affecting any of the other predictions. Likewise for Trump doing better than expected.
***Choosing between #1 and #2: this election in a nutshell. Both are gross, but one is much grosser than the other.
edit: added two footnotes, because it's hard to make a precise and well-explained prediction without long, detailed footnotes.