Thoughts on that thing that’s happening

The most respected national pollsters all agree that, on the eve of Election Day, the presidential race is as close to a tossup as one can have. What to make of this?

One thing that several people have pointed out is that there’s a lot of evidence of “herding” — that is, of polls tweaking their modeling assumptions ex post facto to avoid being outliers. The best evidence for this is that, if the race truly is this close, there should be more divergence in the polls than we’re seeing, simply from a combination of different pollsters using different modeling assumptions, and even among pollsters using substantially similar models, just from margin error effects alone.

I don’t actually know anything about the technical side of polling, but count me extremely skeptical of any kind of after the fact tweaking of a poll’s initial methodological assumptions, once the raw numbers have come in. That sure sounds a lot like a form of data dredging — not in this case to generate statistically significant results, but to get results that “look better,” for reasons that have nothing to do with the original modeling assumptions.

So who knows what the polls mean at this stage? There’s a lot of evidence that the last week of the campaign has been really bad for Trump, and that this should hurt him with late deciders. That’s the empirical reason that I’m doubting that the election is actually an Anton Chigurh-style coin flip. The psychological reason is very simple: I find the idea that Trump could win unthinkable. Intellectually I know it’s very possible: even if the polls are herding and Trump has had a terrible last week, it would hardly be shocking in the statistical sense if Trump were to win.

But unthinkable in the sense that I can’t seem to face up to the facts.

Say something once; why say it again?

Open thread for election-related content.