Monday, December 12, 2016

The 2016 Polls Were OK

Because Donald Trump won the electoral college and because it takes weeks for the full popular vote totals to be counted, the perception has set in that the polls were wildly wrong this year. But they weren't! Here's my layperson's understanding of what happened.

To get a rough sense of how the national polls performed in comparison to recent presidential elections, let's look at the Real Clear Politics national poll average for the last four elections.

RCP Average
Kerry v Bush 2004
Bush + 1.5
Bush + 2.4
Obama v McCain 2008
Obama + 7.6
Obama + 7.3
Obama v Romney 2012
Obama + 0.7
Obama + 3.9
Clinton v Trump 2016
Clinton + 3.2
National polls were only slightly less accurate this year than in 2004 and 2008, and significantly more accurate than in 2012. Ok, but what about state polls? Sam Wang's Princeton Election Consortium used only state polls and calculated that Clinton's margin was 2.2%. This was highly accurate. Wang missed Clinton's actual winning margin by only 0.1%.

Ok, but if the polls were fairly accurate, why did so many data journalists miss what was coming? Part of the answer is that they didn't. We've had over 50 of these presidential elections, and it turns out winning the popular vote is a really good way to become president. Winning it by millions of votes is an even better way to become president. This is what Clinton did, but she lost the election. Every once in a while, the candidate most people vote against manages to thread the needle and win.

This is what Trump did. He had very small margins in Pennsylvania, Michigan, and Wisconsin, getting just the right amount of votes in just the right places. Given the size of Clinton's lead, the aggregators were probably right to believe she was the heavy favorite to win the election. Losing the popular vote by millions means you need everything else to fall perfectly into place. For Trump, the stars aligned.

But that doesn't mean there weren't clues before election day that this could happen. Many of the aggregators seem to have missed signs that there was an unusually high chance of a popular vote/electoral college split. Polls showed Clinton outperforming in sunbelt red states that she was unlikely to win in any case, like Texas, and underperforming in important midwestern states like Ohio and Iowa. This opened up the possibility that Clinton would have an unusually high number of "wasted" votes. And that's exactly what happened. The possibility of Trump losing the popular vote while winning just the right number of key swing states was the main reason Nate Silver's fivethirtyeight gave Trump a significantly higher chance than the other aggregators.

Even though aggregations of state and national polls ended up being fairly accurate, there were larger errors in a few swing states. Few observers anticipated that such a large gap would open up between swing states and the rest of the country. Ohio, a state Obama carried twice, ended up being 10 points more Republican than the national vote; Iowa, 11 points more Republican. Those are surprising numbers. But Trump's strength in those states was already apparent before election day, even if the magnitude of his victory there was unexpected.

So the polling industry did not collapse. The polls have had better years, but they've done worse too. They were ok. And that means they continue to be useful instruments. But the aggregators were overconfident in the conclusions they drew from them, underestimating the range of uncertainty created by the interaction of polling, the popular vote, and the electoral college. The pundits failed more than the data failed. The shock many of us felt on November 9 was moral rather than empirical. It was less about what the data indicated was possible, and more about our own inability to imagine that such an indecent outcome could really occur.

All of this might seem like a moot point, but it has implications for political strategy going forward. The more popular candidate lost, and Democrats should act accordingly. They should resist the urge to try to draw profound lessons from this election. They got unlucky, and yes, the electoral college is dumb. They don't need to become more Trumpian in their appeals or move toward his policy positions. He's an unusually unpopular president-elect. They need only resist him and they will reap political rewards while doing the right thing.