The Thin Blue Wall

It comes up every election: The national popular vote doesn’t matter. We elect presidents through the electoral college.

It usually comes up when either (a) when we’re too far out to have good state polling, and then (b) when there is a disconnect between state and national polling. It has come up multiple times this election in the form of the alleged “Blue Wall” that protects Democrats against the majority vote. In September, Nate Silver took a lot of grief by talking about the national polling when state polling still looked good for Clinton. The national popular vote doesn’t matter, we were told, because states. Then state polling changed and the story changed to “It’s temporary.”

Which, of course, it was. However, we’ve gone through this cycle a couple of times now, with each side arguing that the national polls that were starting to look unfavorable were less important than state polls that looked better. This isn’t wrong exactly, as state polling can actually give a more accurate view of the national picture than national polls can. But it’s primarily a question of whether we should apply national polling to the states, or extrapolate state polling onto the national. Because if they persist in disagreeing, one of them is probably wrong.

If there is a disagreement come election day, we’re likely going to find out that the state polling is wrong, the national polling is wrong, or both are wrong. What’s not likely to happen is a wildly divergent result between national and state polling. As we learned in 2000, the popular vote and electoral college vote can produce an inverted result. However, it can only do so in the event of a really close race. If a Republican or a Democrat wins the popular vote by 2%, there is an overwhelming likelihood that they have won the presidency. Even if it’s just by 2%.

When people try to get specific with the states, they tend to get hung up on the stated odds in specific states at the expense of the big picture. They look at a 40% chance of winning one state, a 30% chance of winning another, and a 20% chance of winning a third, and think that the candidate has about a 2% chance of winning all three. In fact, it’s probably closer to 20%. Chances are, if a candidate wins the 20% chance state, they’ve already won the other two. Swing voters (who are deciding between R and D) and marginal voters (who are deciding whether to show up at all) in Pennsylvania and Colorado simply aren’t that different. The same things that effect the thinking of voters in one pool typically affect the other.

People look at the 2008 and 2012 elections and see that Obama cleaned up in the electoral college and assume that this projects a very strong electoral college advantage. The electoral college, though, is a result amplifier. It takes close elections and makes them look more lopsided than they are. Think of 1984 when Reagan won by an 18-point margin and won almost every state despite 40% of the country voting for Mondale. Or George Bush’s 1988 victory, which had a stout 7.7% margin (close to Obama’s 2008 margin), but netted him a whopping 426 Electoral Votes (compared to 365 for Obama). That makes it seem like the Republicans had a huge electoral college advantage. But they didn’t.

You can get an idea of the electoral college advantage by looking at the Tipping State. The Tipping State is the state at finish line for the losing candidate. In 2012, the Tipping State was Pennsylvania. Romney would also have had to win Florida, Ohio, and Virginia, but he lost those by slimmer margins than Pennsylvania and with a uniform swing it was Pennsylvania that would have put him over. While swings are rarely completely uniform, it’s a reasonably good baseline. The assumption of a uniform swing did a better job of predicting 2012 margins than state polling did.

In 2012, "2008+Uniform swing" outperformed polling averages by state, so can we cool down with the electoral college analysis?

— (((David Shor))) (@davidshor) May 5, 2016

The long and short of it is that national totals come from somewhere, and they don’t usually come disproportionately from solid-color states. If Mike Dukakis had a surge of support, more likely than not it would have come from states red, blue, and purple. The tipping state in 1988 was Michigan, which Bush won by a 7.9% margin. He won nationally by 7.7%. As the national margin gets closer, states start flipping. Often, many states at once. A two-point swing in 1988 would not have given Dukakis any more electoral votes, but a third would have gained Dukakis nearly 60 electoral votes, and a fourth point would have made it 120.

Historically, the Tipping State is usually not very from the national totals. Since 1950, the gap between the Tipping State and the national margin has only exceeded two points twice and is usually lower than 1%. Starting at 1950 is cheating a little bit, because before that divergences were more common. Why? Our national culture was less national then, for one. For another, most of the exceptions fall into the category of (a) extremely lopsided races or third parties (1912 and 1948). The same applies to elections after 1950, with exceptions appearing in lopsided 1964, 1972, and 1980, as well as the three-party 1968.

The most inconvenient counter-example is 2008, wherein John McCain’s national margin was 7.2% (not a blowout by the above standard) and the Tipping State, Iowa, was 9.5% away. Now, by the standards of the comparison, 7.2% is not a blowout. By contemporary standards, of course, it is. four years later, Mitt Romney cut the popular vote margin down to 3.9%, and came out behind in Tipping State Pennsylvania by 5.3%. In the event that a Republican were able to cut further into the national vote, there’s a good chance the gap that went from 2.3% to 1.5% would close even further (in 2004, the gap was .3% as Bush won Ohio by 2.1% and nationally by 2.4%). As of October 30th, Trump is down 5.2% nationally and 5.3% in Colorado. As his national fortunes have waned, so too have his swing state fortunes.

Ultimately, the Blue Wall likely only protects Democrats up to a point or two, maximum. In other words, it’ll break a near-tie. Most elections aren’t that close, and the only close election to produce a large margin was almost exactly 100 years ago. The only inverted outcome was 2000, wherein Gore won the popular vote by half-a-percent. A Republican that wins 51% of the vote wins.

As it pertains to this specific election, there are three things worth considering.

The first is that one way a campaign can make targeted gains in swing states is through good organization, a strong GOTV operation, and possibly (though I’m not convinced) advertising. This confers an advantage to Hillary Clinton. Second, to the extent that the swing from 2012 to 2016 is not uniform, there are reasons to believe that it could actually favor Donald Trump as he may shed unhelpful votes in red states to pick them up in purple ones. Back when the race was close (September 27th, to be exact), FiveThirtyEight showed him down by 1.2% nationally and down by 1.4% in the Tipping State of Colorado, which is a far cry from the 1.5% gap that Trump had. The third thing worth considering is that this election is not, in fact, close. So unless something changes, I hope you enjoyed this intellectual exercise.

Image by marc falardeau