ELECTION POLLS - Misinterpreted

"The Presidential Polls Weren't As Wrong as the Interpreters" by Robert Faturechi, ProPublica 11/21/2016

The polls, of course, turned out to be wrong.

In the days since the election, reporters and pollsters have been mulling why most projections missed the extent of support for Donald Trump in the Rust Belt and the weaker-than-expected turnout for Hillary Clinton among some voter groups.

Was there a “shy Trump” phenomenon, with some of his supporters reluctant to tell pollsters they were voting for him?  Were projections thrown off by the unusually high percentage of undecideds on the eve of the election?  Was the importance of the country's demographic shift exaggerated?

Before the election, we talked to Harry Enten, a senior political writer and analyst for FiveThirtyEight, about what the polls were projecting.  We have him on again this week for a postmortem, and a discussion of how polling -- and the interpretation of polls --can be improved.

A few highlights from the conversation:

Robert:  How much did party identification, what party a voter considers himself or herself to be a part of, how much did that matter in this election and has that changed over time how much that plays a role in which candidate a voter chooses?

Harry:  So basically, party identification is the name of the game.  Donald Trump got about 90 percent of self-identified Republicans.  Hillary Clinton came pretty much doing exactly the same with Democrats.  That means that if you know how self-identified independents are going to vote and you know the percentage that Democrats and Republicans are going to make up the electorate, your poll is going to be pretty much right on.  There was a lot of talk before the election that Donald Trump would suffer among self-identified Republican voters.  That didn't really seem to be the case.  Yes, perhaps he did a little bit worse than House Republicans did among self-identified Republicans but he did well enough, and indeed, this is part of a continuing trend in American politics where party identification basically tells all.

It's a drug, as I like to say.  Self identified Republicans are pretty much always going to vote with the Republican candidate and self-identified Democrats are always going to vote for the Democratic candidate.  That's far different than it was 35 to 40 years ago when the winning candidate regularly got a lot of support from the other side.  For instance, Ronald Reagan, when he won the 1980 election, got well over 20 percent self-identified Democrats.  That's just not the case anymore.  You pretty much get your side's supporters and then some independents and you get very little support from the other side and that is part of a continuing polarization in this country of Democrats winning Democrats and Republicans winning Republicans.

Robert:  Have there been other instances in modern history where the polling has been so off?

Harry:  Well, I mean yes.  It was just not that they necessarily made the difference who was going to win and lose.  I mean in 2000, for instance, the final average of national polls taken over the final week of the campaign had George Bush winning by about three percentage points.  He ended up losing by nearly one percentage point.  That error was basically the average error that we saw across state polls.  It just so happened that George W.  Bush ended up winning in the Electoral College so it didn't end up mattering, but also was the fact that I think a lot of people understood that the race was close in a way that I think many people especially in urban centers and on the coast just couldn't imagine Donald Trump winning, and there have been errors for sure that have been even larger than what we saw this time around on the state level.

For instance, in 1980 the state polls were much further off, but again, the final state polls in 1980 indicated that Ronald Regan was going to win by a small margin.  He ended up winning by a large margin, so I don't think people were able to connect those back.  They said, "Oh, the person who was leading won, so no big error."  This time around, though, we had actually a slightly smaller error, it just went in the other direction and that's why I'm very careful not to necessarily talk about errors in binomial senses, that is, wins and losses, but rather in terms of margins.  And this time around, even if the margin wasn't as large as it was, say, in 1980, it was large enough and went in the correct direction -- or the wrong direction, depending on who you are -- to be able to flip the winner in the key swing states and ultimately the election.