Well, you did it, America. You just elected Donald Trump.
In the midst of Donald Trump’s victory in Tuesday’s presidential election, many liberals like myself were left stunned and disillusioned. Many of us started looking for a way out, or someone to blame. Many turned their ire to the polling industry, which had Hillary Clinton ahead in many polling averages and lulled them into expecting a win for her. They were ready to indict the entire industry and swore they’d never trust the polls again.
But here’s the thing: the polls were actually more accurate than you think.
Take national polls first. Hillary Clinton had about a 3.2 percentage point lead in the RealClearPolitics polling average heading into election night. She currently holds a 0.2 percentage point lead in the popular vote (which might expand after California’s votes are fully counted, but let’s just use these numbers). That three-point error is…actually pretty standard. Every poll publishes a margin of error, to adjust for the fact that they can’t possibly poll everyone in the nation in the right proportions.
Still don’t believe me? Let’s look at 2012. Obama had a 0.7 point lead in the RCP average, and won the election by… 3.9 points. That 3.2 point error is actually bigger than 2016’s! 2004 polls were about a point off, and 2000 featured a polling miss about the size of 2016’s. This was also about the same-size error that pollsters missed the vote on “Brexit,” the UK’s decision to leave the European Union, by. Simply put, errors happen.
The state polls showed that Clinton might be vulnerable in the Electoral College, as well. Let’s examine the averages vs. the results in key battleground states.
Most of these averages were… pretty darn accurate. Polls in Ohio and Iowa missed the margin of victory, but got the winner right, which in my opinion is more important. Most of the other misses were by narrower margins than the national polls did.
Now, by defending the polling industry as I am here, I do not mean to suggest that they didn’t make any mistakes, and shouldn’t reevaluate some of their methods. For instance, at the bottom of the chart above, you’ll see the two biggest and most consequential polling misses, in Michigan and Wisconsin. Those errors were larger than the national error, and got the winner wrong. Nate Silver said in the FiveThirtyEight Elections podcast that he thinks that pollsters may have conducted polls that showed a closer race, but like many of us, didn’t quite believe that those two states’ margins had shifted so sharply in just four years. Maybe they changed their demographic weighting, or didn’t publish polls that showed a closer race (known as “herding”), or some other change, but they clearly didn’t see Trump wins in those states coming.
Speaking of Nate Silver, many people had criticized his model as convincing them that Clinton was a surefire winner. Silver’s FiveThirtyEight model had Clinton as a 71.4% favorite to win. Those were good odds for her, but far from a sure thing. Silver’s model, in fact, gave Clinton much lower odds than a lot of models that had her chances in the 90s. He didn’t have her as more than about a 55-60% favorite in most of the swing states above (with the exception of Michigan and Wisconsin, of course). Many people failed to realize that Silver’s model does not make deterministic predictions (“Clinton will win Wisconsin”), but rather probabilistic predictions (“Clinton has an 83.5% chance of winning Wisconsin”). The thing about saying that something has an 83.5% chance of happening means that, by definition, sometimes it won’t happen. Silver’s model also heavily relies on polling data, so if the polls are missing something, his model will too.
I think we need to look at other factors to understand why Clinton lost. Media headlines seemed to exaggerate the chances of a Clinton win, without looking deeply at the polling data. Like pollsters, many of them probably didn’t think we’d actually elect that guy. She didn’t visit Michigan or Wisconsin during the general election, which may have meant that she was a bit overconfident of a win in those states. Maybe there was a “shy Trump voter” effect where people wouldn’t admit to pollsters that they were supporting Trump, even if they were. Perhaps Clinton made a compelling case against Trump, but not a case for herself. But tarring and feathering the polling industry when they did a reasonably good job just isn’t productive.