Waylaid Dialectic

November 10, 2016

Trumping the polls

Filed under: Random Musings — terence @ 8:47 pm

I’m still trying to wrap my head around the US election results. What will happen now? Trump frightens me. I didn’t love Clinton’s liberal interventionist approach to foreign policy. But if Trump is as impulsive as he appears to be, and if he truly holds the beliefs he professes to hold (and also isn’t a quick learner), we will soon be living in a much more dangerous planet.

Alienating Muslims won’t help win the war on terror (you too France). Acting irrationally around China brings mushroom clouds closer. Denying climate change ruins future generations’ lives.

Beyond the big stuff, as someone who uses polling data, I’m pondering how the polls and their various aggregators got it more or less wrong (more or less because they simply predicted the chances of a Trump victory were low, which is not the same as saying he wouldn’t win).

And so for future reference: Nate Silver one, two, three; Natalie Johnson, John Sides.

[Update: Gelman, Gelman, Gelman]

First, national polls weren’t that wrong: they probably overestimated Clinton by a couple of percentage points (on average). And their prediction that she would win the popular vote will likely be borne out.

Second, even state level polls weren’t, on average insanely wrong: they were out by (guestimate) 3-5%. When you look at the binary: Clinton predicted to win, but she loses, that’s very wrong. When you look at a continuous variable – predicted vote share vs actual it’s wrong, crucially wrong if this is your business, but not that wrong.

Still wrong needs explaining.

I don’t think it was social desirability bias (people being ashamed to admit they were going to vote for Trump). One of the posts above links to attempts to test for this, which appeared to show it wasn’t an issue. And Trump voters struck me as loud and proud, and unafraid to admit it to pollsters (and remember in many instances you’re just admitting it to a computer). But perhaps those Hispanics or women who voted for Trump were more reluctant to admit it (even to themselves?).

I do think it could be that Trump voters were simply more likely–or the portion of them who were frustrated and alienated–to hang up on pollsters.

I also think the final FBI headlines in the paper (all you needed was the reminder of why you weren’t so hot on her; so the subsequent exoneration may not have done much) may have, rather than shifting Clinton voters to Trump, simply depressed turnout amongst tepid Clinton supporters.

Indeed, one thing that complicates polling is that you have to not only figure out who people support, but also which people actually turn out and vote. I want to learn if the latter was more wrong than the former. Though I’m not sure how you do this.

On the turnout points above — one of the links above links to something about exit polls getting it wrong too. If this is the case it can’t have been turnout and is more likely social desirability bias or aversion to being polled.

These are just guesses: I’m interested to see how they will be borne out by subsequent analysis.

A final thought: at least some of what is to come (particularly midterm elections) will depend on how many people voted for Trump because they liked him (remember his unfavourably ratings were low) and how many voted for him as a kind of protest vote. The second category of voter may peel away from him quite quickly. Likewise, white working class voters might peel away quick enough when his economic policies (if he follows through on them) start hurting them. Although things might also just get ugly as he then does other stuff to keep them in his camp.



Blog at WordPress.com.

%d bloggers like this: