### How Inefficient will the Electoral College be this time Around?

Hang around on the internet for long enough, and you'll read article after panicked article, about how this time around, Trump could lose the popular vote by 10 million votes (or some other very large number) and still win the electoral college.  Could he?

Let's stipulate that the electoral college is bad and terrible, and it's still likely to benefit Trump this time around, but I'm going to put my marker out there that the fear-mongering about the electoral inefficiency are a bit overblown, and by and large, early indicators are the inefficiency is going to be less significant this time around than in 2016.  And I'm going to base this on data, not on simulations. You know the old expression: GIGO (garbage in, garbage out).

First, a couple of important definitions.  The tipping point state is the one where, if you listed the states in order of margin, it's the one that would give the Democrat their 270th electoral vote.  For example, from 2016:

 State EV Clinton Trump Margin Cumulative EV Minnesota 10 46.4 44.9 1.5 229 New Hampshire 4 46.8 46.5 0.3 233 Michigan 16 47.0 47.3 -0.3 249 Pennsylvania 20 47.5 48.2 -0.7 269 Wisconsin 10 46.5 47.2 -0.7 279 Florida 29 47.4 48.6 -1.2 308

So, looking at the list, Hillary would have had to win Wisconsin, with a margin of -0.7, as well as all states that were closer.  (Side note: because of the ridiculous system where a 269-269 tie in the Electoral College is broken by the House of Representatives, but each state gets a vote, chosen by the majority of their delegations.  Even though Dems control the House, the smaller states run rural, so the ties would inevitably be broken in favor of Trump).

We then define the Electoral College Inefficiency as the difference between the popular vote margin and the swing state margin.  In 2016, the popular margin was D-R=2.2 (two party vote share), so the EC Inefficiency was -0.7-2.2=-2.9, meaning that Hillary could have won by as many as 4 million votes and just barely won the EC.  Negative EC Inefficiencies mean that it benefits the Republican, and positive mean that it benefits the Democrat. And, even recently, it hasn't always benefited the Republican.  Here are the Inefficiencies from all of the elections since 1980:

 The Electoral College Inefficiency, 1980-
2016 was the biggest inefficiency in more than 40 years, though it's worth remembering that Obama also could have lost the popular vote by about 3 million votes, and still won (at least in 2008).

There's a lot to keep in mind, here. After all, things are changing and faster than you might think. In one of my earliest posts, I estimated that simple demographic changes (read: older Republicans dying off and are replaced by younger, more diverse, more liberal voters) account for an electoral shift of about 0.6%/year toward the Democrats, or about 2.4% per cycle.

Turns out a) that estimate was very good, and b) I didn't need to make it so difficult.  Quick shoutout: the good folks at Daily Kos Elections have an outstanding spreadsheet of every presidential election, state by state, which I'll be using extensively here.  You can use it yourself if you don't buy my analysis.  The first thing you'll note is that if you do a simple plot of presidential popular votes over, say, the last 40 years, there's a very clear trend toward the Dems:

 The popular vote margin (D-R) in every presidential election since 1980.
Even without drawing a line through it, the trend is clear. There's a huge scatter (around 8%), but on average, Dems are gaining nationally, to the tune of about 0.5%/year (a touch less than I estimated above).  This means, by the way, that if all of the still living voters from 2016 came out and recast their votes exactly as they did before, Trump would lose.  Indeed, the main point is that no matter how you slice it, there are few scenarios, even with voter suppression, where the Republicans will be a viable national party for much longer.

But the country isn't the only thing that's swinging.  We can also look at the trends of each state relative to the national.  For example, here's Minnesota:

 The D-R margin in MN presidential elections, relative to the national margin.
Oh no! It's becoming more conservative! Only, it's not.  It's only moving relatively rightward at the same rate the country is moving leftward, so, on average, MN is staying just where it is.

Indeed, it's worth noting that if you simply extrapolate current trends (always a dangerous decision), only 11 states are becoming redder in absolute terms: SD, MS, MO, LA, OK, WY, AL, AR, TN, KY, and WV, while states like UT, FL and AZ are getting blue faster than the national average.

We can do the same for all 50 states+DC and ask if we could have used those trends to predict the Inefficiencies in any particular election year:

 Actual historical EC Inefficiencies, with their estimates from approximating from linear trends.
It predicts, on average, that we might expect Trump to have about a 1 point EC advantage in November, but the model doesn't really follow the data.

For what it's worth, we can also use those trends and compare them to the actual Electoral Vote outcome:

 Electoral College outcomes from the last 40 years (solid circles), compared to a simple model of states and the nation predictably driving (green triangles).
This part of the prediction isn't terrible.  It gets 1992 and 1996 wrong, along with 2016, but the rest are pretty close, not just in terms of the winner, but in terms of the margin.

The final wrinkle is that the differences between a real election and the smoothed linear model are highly correlated.  The 2016 map saw certain states which seemed to move rightward more than most.  We could call it a "Trumpy" movement.  Here are the relative shifts during a Trumpy electoral year:

 In a Trumpy year, Dem does better than average in the South, Texas, and California, and the Republican does better in the Rust belt.
Not for nothing, but this is almost identical to simply comparing the 2016 map and subtracting out the 2012 map:

We could ask how "Trumpy" each election was, as a correction.  2016 was 100% Trumpy, by definition.  Every other election except for 1980 and 1984 had a negative Trump coefficient (even the ones that various Bushes won), while Reagan's maps were more Trumpy.

If we know the Trumpiness of an election (along with the national margin), we can predict the Inefficiency very accurately.  Here's the predicted Inefficiency:

where the scatter of blue on the right show a range of possible Trumpiness. Even if the map is just as Trumpy as 2016, the Inefficiency will be about a half point less.  And if the Trumpiness is zero, the Inefficiency will be as low as about -1.

However, there are better ways of estimating the Inefficiency (or the Trumpiness) for any given year. Look at the polls!  State polls, in particular.  We know where the states should be in a "typical" year.  Are the states results Trumpy or Untrumpy relative to the expectations?

For instance, Sanders currently has significant and reasonably recent Head-to-Head state polling in CA, NV, NM, AZ, IA, WI, and NH.  You can see all of this at my 2020 Tracker, where higher opacity corresponds to more recent polling:

We can use those states to do an estimate (independent of national polls), and it turns out that the relative Sanders to Trump performance points to a ~5 point lead in the popular vote.  But more relevant to this discussion, the data points to a Trumpiness parameter of nearly 0, and thus an EC Inefficiency of -0.8.

The same is true for almost all of the leading candidates. The EC Inefficiencies are current:
• -1.1 (Bloomberg)
• -1.2 (Biden)
• -0.8 (Sanders)
• -1.4 (Warren)
• -0.7 (Klobuchar)
In other words, though this is just a snapshot, we're nowhere near the EC Inefficiency=-4 nightmare scenario where Trump loses by 5 million votes, but still wins.

So breathe easy, but work your ass off.  If we do things right, we won't be close enough for the Electoral College to screw us this year.