Holy hell. What a turn in U.S. history. And we didn’t even see it coming—because the polls were way, way off, as they were in 2014. We found ourselves facing vote margins in the presidential race that were 5, 10, even 14 points off from our final Daily Kos Elections polling average margin. The miss, on average, was a little under 5 points too rosy for Hillary Clinton on the margin. That will probably improve a little as more ballots are counted, but not that much.
What’s remarkable, as the graph at the top of this post shows, is that the polling average miss was almost universally in Trump’s favor—those are the dots below the green line. Polls often miss, but almost always, some misses favor one party and some favor the other. The uniformity of the misses here is striking. Want further details to obsess over and distract you? Read on.
First, an expansion of the graph to show all 50 states (with DC excluded so you can see the data better). I’ll be using the Daily Kos Elections final model numbers, but you’d see similar patterns no matter how you average the polls. You’ll notice that there were still some states where Clinton outperformed the polls—NV, NM, WA, CA, and HI (and DC, again not shown).
What was the polls’ major failure? Underestimating Trump’s share in almost every case. Below, every point above the line is a state where Trump outperformed the polls. That’s 49 of ‘em—all except Hawaii and DC (sorry, still not showing it—it would be impossible to fit into a normal-sized chart).
The polls did a lot better with Clinton’s vote share. Here, every point above the line is where Clinton outperformed the polls:
You may have noticed a pattern in the graphs above. It’s our old friend showing up: the relationship between the partisan lean of the state and its polling errors. In the next graph, you’ll see that polls overestimated Clinton’s margin by a large factor in the deep red states (well below the line), and a moderate amount in the more neutral territory (mostly below the line), while mostly underestimating it in deep blue states (above the line). This pattern has been around for a while, and seems likely to stay with us.
So what’s going on? I don’t know. One hint is that compared to 2012, pollsters really blew it with the white vote. Final polls in 2012 had Romney ahead of Obama by 17 points among white voters; exit polls had it at 20 points. This year, the final polls had Trump ahead of Clinton by just 13 points among white voters, but the exit polls? They showed Trump winning by 21.