What an upset! No, no, I don’t mean for what’s-his-name, I mean for Ann Selzer. I feel badly for her, actually, because she was trying her best to damp down all the “Ann Selzer is the greatest pollster in the history of the world” hype in the full knowledge (as a statistician) that it was more likely she’d get it wrong at some point than it was that her winning streak would be uninterrupted.
I’m confident that over a series of elections, she’ll get it right far more often than I will. Her methodology makes sense, whereas mine’s like using Paul the Octopus. But I confess I’m surprised. My Iowa poll methodology either worked better than I expected, or it was a total fluke. Only one way to know: Let’s see if I can do it again.
Like ToryWarWriter — who called this well before I did — I started with the knowledge that most of these polls weren’t well-constructed. But I started with the assumption that some of them were very well done indeed. So I used the aggregate of the best polls as a baseline, and then added a few ingredients:
- I took into account the known unknown: How many Trump supporters were apt to be Trump voters. Trump’s an unpredictable candidate; I figured, so his supporters were apt to be unpredictable voters. Since we had no data suggesting they would turn out to vote the way more traditional candidates’ supporters do, I assumed that they were less likely to turn out. This assumption was shaped by a) what I wanted to believe; b) Trump’s failure to construct a get-out-the-vote apparatus in Iowa (conservatives believe that things become traditional for a reason); and c) My instinct that Trump’s a protest candidate apt to be used by American voters as French voters use the the National Front: to torture and punish its politicians with the threat of nuking the whole country, but in the end decide, “We’re not suicidal.”
- I defaulted to the belief that Iowa Republicans resemble me a lot more than they resemble the left-wing stereotype of Republicans. I did so partly because I want this to be true; but more so because I’m American: My intuitions about my own country are based on a lot of personal experience.
- I took some knowledge of Iowa and its culture — not deep knowledge, because I’ve never lived there, but the knowledge most Americans would have — and went for a virtual walk in a town in Iowa. I chose it because it was the first town I found with its own newspaper. I was looking by alphabetical order. Close to random, in other words.
- I read the newspaper carefully and got to know its local columnists. I checked out what’s been happening in Albia over the past year or so. I looked at the local schools, the crime rate, the cost of living, and the health care system. I concluded that Albians were probably pretty similar to Ricochet members in the way they viewed the world: They definitely didn’t come across as rude, angry, populists who’d go for a rude, angry, populist candidate. Nor did they seem naive: They get the cynical ethanol pandering, and they get that this vote is about America, not ethanol. Everyone who worked at the local paper seemed to agree that a bit of humility is a job requirement for the presidency. Finally, Albia obviously liked Cruz a lot.
I don’t think this methodology is sound at all, because it’s based on “What I want to be true,” and a completely inadequate sample size. I got lucky. But now I’m curious, so I’m going to keep doing it throughout the primaries. The day before every election, I’ll start with the best polling data I’ve got, add a heaping dose of my own political intuition, spend a day taking a virtual walk in one city (no more), chosen at random, and see if I can match or beat Ann Selzer more often than half the time.
I’m betting against myself, but I’m curious.