I’ve got a new set of electoral college predictions. I’m using the same method as before, but with these differences: a) I had to use less than ideal polls (c rating, a few that overlapped with days prior to POTUS debate III) on the last run, this time no such polls are used; and b) there are some new polls added in this time.
The difference is interesting, and somewhat concerning (compare to this result). For example, in this run, Arizona, Virginia, and New Hampshire go for Trump. Most people think of that as unlikely. Personally, I don’t see Virginia doing that. New Hampshire is conservative and is very white (thus potentially Trump-leaning), but is in transition. However, these states are all within a very small fraction of the 50-50 cutoff. Oddly, North Carolina is not that close.
I did a second map (using 270 to win) with the same data but adding ca 3% correction for ground game. Trump seems to not have much of one. I asked a number of colleagues what percentage correction they might use for a good vs. bad ground game. These are people who have ground game experience and a good record. They were all over the place in their suggestions, and noted that any such guess would be iffy this year. So, I picked 3%. North Carolina is actually slightly more tha 3% of the 50-50 line, but I included it anyway in this latest run, which adds New Hampshire, Arizona, Virginia, Ohio, and North Carolina.
I want to remind you of a couple of things about this model. It is indifferent to your opinion as it might be derived from polls. That is the point. This is meant to account for some of that potential observational bias, or at least, ignore it. Also, this model tends to work ver well. However, it is accurate mainly with respect to the percentage of the vote assigned to each candidate in a unit area (a state), not whether the candidate takes the state or not. In other words, we look at this and freak out about a state being blue or red, and the model says, “Who cares about that, I’m trying to tell you the PERCENT of the vote per candidate. So, 49 vs. 51 are two points off, and 81 and 83 are two points off, they are the same, silly human!”
The real meaning of this particular prediction, which uses BETTER DATA POINTS than the last one but FEWER OF THEM, is that it is not a) closing in on a Clinton landslide — that isn’t going to happen and b) it shows the kind of crazy variation over time that should keep us up at night. On the 8th. But not so much other nights, because it is, essentially, impossible for Clinton to lose.
And, to underscore that point, here are the states that my model currently says will go to Clinton, on the stronger Clinton side of the distribution, that are the minimal needed to get 270 votes:
So, that’s how Clinton can win without Pennsylvania or Ohio. And, again, this is the quirky nature of variation near the 50-50 line. Clinton will probably win Pennsylvania (according to everything) and couple of other states, and does not, therefore, need Florida. Probably.