some quick fact checks on the election

The election’s over. Trump won. The GOP won the Senate, and kept the House. That’s all the grim truth. No one fully knows how bad this will be for the next four years and for all the years to come, but we have some educated guesses and it’s bad.

Some of the other claims flying right now – about how this happened, who’s responsible, how we missed this, and so on – are misleading or false. This post isn’t a detailed analysis, it’s just a quick attempt to get everyone on the same page. It’s also my meager attempt to deal with the emotional fallout this morning by doing what academics do best: post-hoc analysis and contemplation.

  1. Trump didn’t win a majority of votes. The votes are still being tallied, but it looks like Clinton took the popular vote by a small margin. Trump won about 47-48% of votes cast.
  2. Trump certainly didn’t win a majority of Americans. Turnout was typical for an American presidential election. Something like 60% of eligible voters voted. Update: I’m seeing reports that turnout was down a bit, to perhaps 55%, closer to 2000’s turnout (55%) than 2012 (59%). Many adults who should be eligible are not, due to things like felon disenfranchisement.
  3. Trump did win a majority of white voters of basically every standard demographic breakdown. CNN’s exit polls are probably off by a bit (exit polls are never perfect), but they show the story in reasonable detail. Trump did not (just) win working class whites, or non-college educated whites, or old whites. He won white voters.
  4. Trump won white women, but 53% to 43%, not by larger margins I have seen reported widely.
  5. Clinton won college-educated white women narrowly, 51%-45%. That was the one exception among major white demographics. A particularly depressing statistic for me: Trump won among young (18-29) white voters, 48%-43%.
  6. The polls got it wrong, but not by a historically unprecedented amount. Someone who follows the polls even more closely than I do can offer a full account, but it looks like there was a small systematic error on the same order of magnitude as in past elections. The polls seem to have pegged Clinton’s support pretty closely, but slightly underestimated Trump’s support, in part by underestimating turnout for Trump of white men in swing states and by predicting higher Johnson support than we saw (edit: 538 had Johnson at 5%, his actual numbers are closer to 3%). The polling averages had Clinton up 3 points in the popular vote, she ends up winning by 1. That’s not a huge miss, but it is a miss that implies non-random error. The swing state polls were off by more, and especially polls in those states that Romney won, but again, I don’t think in a historically unprecedented way.
  7. The polling-based forecasters mostly got it wrong. Sam Wang’s vaunted precision was a myth. Nate Silver’s emphasis on the possibility for correlated errors was vindicated. I study and teach about the 2008 financial crisis. The two contexts are very different, but the underlying principle is the same: systematic correlated error means that aggregating a bunch of seemingly independent chances (polls, financial risks) doesn’t actually produce a more certain outcome. Silver, unlike Wang and unlike the quants working in finance, built “fat-tails” into his models to account for these correlations. Here we are.
  8. Political science-style fundamentals models predicted a close election. They look to be largely vindicated (or, at least, not falsified). Political science research is also largely skeptical of the importance of advertising and campaigns more generally. Those claims look pretty strong today, though it’s always hard to know how much worse it could have been. Certainly, Clinton underperformed her polling despite a seemingly large advantage in “ground game.”

What other claims have you seen flying around this morning? What other stories are we getting wrong in our collective (and well-justified) panic?

Author: Dan Hirschman

I am a sociologist interested in the use of numbers in organizations, markets, and policy. For more info, see here.

5 thoughts on “some quick fact checks on the election”

  1. I’m avoiding the “claims flying around.” Can’t stomach it. I do think that there’s a strategic story to be told. As my spouse said in September, the Clinton campaign played Four Corners, i.e. a purely defensive strategy of stalling. She/they thought she’d wipe the board and sought to gain Republican centrist support. In doing so, she walked away from the early moves to the left that could have inspired the left wing of the Dem party and possibly spoken to discontented Whites. On the Trump side, the campaign basically imploded in September, and the big money pulled out and focused on down ticket races. I think this matters. The Trump campaign had its “October surprise” in Wikileaks and the last-minute FBI email story. The Clinton campaign believed the polls and stayed defensive, although I will say that my email was full of Democratic messages that the races were closer than people thought. I’m guessing that all that down-ticket money (probably a lot going into talk radio, which is virtually unmonitored) plus the pull of the racist/populist messages AND the partisan political calculus of controlling the Supreme Court tipped the balance for White conservative voters.

    Liked by 1 person

  2. Hi Dan. What a day. I’m a little puzzled by your distinction between “The polls got it wrong, but not by a historically unprecedented amount” and then “The polling-based forecasters mostly got it wrong.” You probably have something in mind — but since most polling is for prediction-purposes, I didn’t quite see the distinction. My sense of what you’re saying is that some of the independent metrics (popular vote estimate, state vote estimate) that the pollsters arrived at stand up okay — but since these metrics were systematically correlated, it resulted in wrong predictions when the forecasters combined them. Would you like to elaborate? I’m still trying to puzzle through what exactly happened.

    Like

    1. Apologies for the lack of clarity. What I had in mind was that any individual poll showed a close race, with a central tendency of Clinton +3-ish in the popular vote, but margins of error that included a tie (which favored Trump, since Clinton rated to need more than 50% of the votes to win the electoral college). The question was how to interpret the mass of polls all saying similar things. Most forecasters (The Upshot, Sam Wang, etc.) saw the relative uniformity of the polling data as a reason to increase their estimate certainty. In other words, they treated the polls as un (or at least less) correlated, so that if two polls tell us Clinton+3 it means a lot more than if one poll tells us that. 538, led by Nate Silver, argued that the polls rated to be systematically wrong in the same direction – and that given the high amount of undecideds and third-party candidates, plus the usual way in which polls tend to be systematically off by a couple percentage points – it was very possible that we’d see a nationwide shift of a couple percent towards Trump.. which is basically what happened. That 2% shift did not falsify any single poll – none claimed that level of accuracy – but it did falsify some of the poll-based forecasts which suggested such a shift was implausible. Does that make sense?

      Andrew Gelman’s take is, as usual, very lucid and far more authoritative than my musing.

      Like

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s