A lot of contrarians and Trump supporters have been talking about how people who were surprised by Trump's victory clearly just don't get it and need to learn about how the world really is.
And the @nytimes malware installer is back up. It was a blinding passion for objective data & analysis that misled us. But not cheerleading. pic.twitter.com/WKQNIUqT9B
— Eric Weinstein (@EricRWeinstein) November 10, 2016
This mixes together two things that are actually quite different:
- Surprise at Trump's widespread support.
- Surprise at a systematic divergence between opinion polls and voting results.
In particular, the "they don't get it" claim tends to prescribe remedies appropriate for (1), for cases that are at least literally more like (2). This is understandable, because many people surprised at Trump's victory conflate the two as well, but it's important to be clear on what exact problem you're addressing.
Surprise at Trump's widespread support
Anyone surprised at Trump's widespread support was clearly not making reasonable inferences based on widely available opinion polling suggesting that a very large minority of Americans planned to vote for him. Slate Star Codex correctly pointed this out before the election:
538 predicts Hillary has a 65% chance of winning the election to Trump’s 35%. New York Times says it’s more like 84% Hillary and 16% Trump. Both sites agree both candidates will get somewhere between 40% and 50% of the popular vote, and that Hillary seems to lead Trump by 3%. The smart money is on Hillary, but at this point either major candidate could win.
Lots of things can happen tomorrow. Maybe it rains in Philadelphia, that city’s racially diverse and left-leaning voters stay home, and Pennsylvania goes for Trump, winning him the election. Maybe there’s a really good get-out-the-vote campaign among Hispanics, and Florida ends up being Trump 48 Hillary 52 instead of the projected Trump 52 Hillary 48. Maybe the Department of Agriculture announces that Hillary is under investigation for bringing exotic weevil species into the US, and the population turns against her en masse.
And someone is going to confuse this kind of stuff with deep insight into the state of the country.
Imagine that the deciding factor really is a rainstorm in Philadelphia. There was a rainstorm in Philly, therefore nationalism is one of the great motivating forces in human affairs? It was a clear sunny day in Philly, therefore nationalism doesn’t matter anymore? The difference between nationalism being all-powerful and irrelevant is whether there was a cold front over the mid-Atlantic region?
But with a race this close, any deciding factor is going to be about as random as a rainstorm over Philadelphia. Maybe the pollsters made some kind of big mistake and missed shy Trump voters, and the vote goes Trump 47% Hillary 45% instead of the predicted Hillary 47% Trump 45%. So what? The difference between a proof of nationalism’s vigor versus proof its impotence is which candidate gets 47% vs. 45%? Really?
If a Trump victory tomorrow would convince you that X is true, I suggest that you believe X is true regardless of whether or not Trump wins, because Trump’s victory almost certainly will depend more on noise than on X. If a Hillary victory tomorrow would convince you that Y is true, I suggest that you believe Y is true regardless of whether or not Hillary wins, for the same reason. If there’s some Z that you will believe only if Trump wins but not if Hillary wins, then I suggest you seriously reconsider what thought process has led you to decide that you will flip your views on politics and society depending on whether or not there’s a rainstorm or a 2% polling error or whatever.
Instead, I suggest people precommit to their views on politics and society now. We live in a country and a world where Hillary can be at about 47% and Trump at about 45%. This is pretty much all you need to know. It suggests that a lot of people are willing to support a nationalist candidate, and a lot of other people really hate that candidate. It suggests that political fundamentals are totally compatible with a situation where either Trump or Hillary could win based on noise in the electoral process.
The remedy for this sort of surprise is to figure out what might be true about the world, that would lead Trump to have the support of about half of voters. I proposed one model before the election results were known:
Clinton voters think that sincere honesty is reciting a list of factually accurate statements. Trump voters think that sincere honesty is poor impulse control.
My post-election thoughts were largely aligned with this model. The main update I made was about the future – that a Trump victory has shattered the illusion of elite opinionmakers' power, which will likely constitute a major step in the direction of already-existing secular trends that favor candidates like Trump.
One of the least contentious updates, I think, is that the Republican party's base is not ideologically conservative the way conservative elites have been. They were quite willing to support a candidate very different in many ways from the largely consistent candidates the Republican party has supported in the past. They seem not to have felt well-represented by their party. If you've been modeling American politics under the assumption that roughly half the country agreed with the Republican platform, then this is evidence that you need to think again.
There are, of course, other models. If you tell me about them, I'll add links to the ones I find interesting here.
- Ozy points out in the comments that there's a very simple model predicting widespread support for Trump: Most voters are partisans, and favor their party's candidate no matter what. This is itself a pretty interesting fact about politics, and consistent with voters tendency to adjust their views on policy to match the positions of their favored candidates rather than adjusting which candidate they favor to match their policy preferences. This does not explain why Trump won the primaries, though.
Surprise at a systematic divergence between opinion polls and voting results.
If someone is merely surprised at a Trump victory, and not at widespread support for Trump, then the above remedies are irrelevant. They will already have taken Scott's advice and updated their beliefs, to be less surprised by the large number of Trump supporters.
To say that such a person "just doesn't get it" on the basis of their surprise about Trump's victory is unhelpful and commits a major category error.
The basic premise behind opinion polling is that if you want to know how people will vote in an election, you go out and ask a lot of people. Since asking people you know is likely to bias your outcome towards your own opinion, you need some way of randomly selecting people, like randomizing entries in a phone book. Pollsters went out and asked a bunch of Americans, pretty regularly, whether they'd vote for Trump or Clinton or someone else. Pretty regularly almost half said Clinton and almost half said Trump. Then we had an election, and sure enough, almost half of American voters voted for Clinton and almost half of American voters voted for Trump.
The polls also consistently showed a trend where support for Clinton was substantially higher than support for Trump. This trend turned out to be a mismeasurement in some way. But while it would be wrong to say that the polls accurately forecasted the election results, it would be even more wrong to say that the polls were uninformative.
So, if someone was surprised by the election, not because they disbelieved in support for Trump, but because they believed in the opinion polls, what should they learn?
The most prominent poll-based model, Nate Silver's 538 model, predicted a 25% chance of a Trump victory. Models that were more confident in a Clinton victory correctly observed that if you cancel out the noise, averaging a bunch of different sample means gives you a fairly precise estimate of the true mean. They were wrong because they forgot about systematic error. Sometimes something structural about the way you do opinion polling biases all your polls in the same direction.
Here are some ways the polls could have been systematically wrong in favor of Clinton:
- People who reported that they'd vote for Clinton may have been less likely to show up at the polls than people who reported that they'd vote for Trump.
- People who reported that they were undecided may have disproportionately favored Trump, either at the time, or on election day.
- People might have been reporting an intention to vote for Clinton to punish Trump for some of his bad behavior, without a real intention to vote for Clinton.
- People might have been ashamed to express support for Trump to a pollster, or intimidated into not doing so, but felt free to vote for him in the privacy of a voting booth (or their home, filling out a paper ballot).
- The election results might have been rigged.
- A large number of Trump supporters might be inaccessible to pollsters in a way that was hard to adjust for statistically.
This is not an exhaustive list, just the options I can think of off the top of my head. They're not worked out in great detail, and I'd be interested in reading a careful analysis of exactly why and how the polls were wrong.
Three levels of wrongness.
My point is that there are three distinct updates that need to be made.
One is a comparatively narrow, technical update. If you are surprised at the polling error, then you ought to learn that polling errors are likely to be more correlated than you might have thought, and think about why and how that might have happened here. Ideally such an update will prevent you from making errors like Nate Silver's Trump's Six Stages of Doom argument, which tacitly assumed that the unlikely things Trump would have to do to win the Republican nomination were uncorrelated.
The second is a deeper empirical update about the nature of American politics. If Trump's widespread support seemed unlikely, that means that you were deeply mistaken at the nature of American politics.
The third is a matter of basic epistemology. If you followed the opinion polls, and were nonetheless surprised at the election result because you were surprised at Trump's widespread support, then you were not taking the opinion polls literally. You saw Clinton polling at 47%, Trump at 45%, summarized this as the country supporting Clinton, and forgot that this implied that 45% of Americans were saying they'd vote for Trump.
I've been working on that third thing over the last few months. I'm not sure exactly what the relevant skill or habit is, but it definitely involves something like taking literal descriptions of the facts literally.
To be fair to Nate Silver, he learned the right lesson after the primary and correctly predicted the general (at least if you judge the more detailed prediction "Hillary is more likely than Trump to win, but a normal polling error could easily cause Trump to win").
I think the most obvious model of the large number of Trump supporters is the one where most people are willing to hold their noses and vote for their party's candidate, even if they are unenthusiastic about him-- particularly given that Hillary Clinton has been one of the most widely despised figures on the right for thirty years. I think this easily explains the majority of Trump's support.
Agreed that Nate Silver learned his lesson well 🙂
This has been driving me crazy, thanks for writing up