Spoiler: this is a summary of a failed experiment.

There’s a wonkish new measure of gerrymandering that’s influencing the courts called the efficiency gap. It is a measure of partisan bias of a districting plan, trying to quantify “wasted votes”. It’s not too complicated a measure. Suppose a single district election goes 53 voters for candidate D and 47 for candidate R. Then 2 of those D votes were wasted, because they were in excess of the 51 votes needed for a victory. All 47 of the R votes were wasted, since their candidate lost. If you count up all those wasted votes over all the districts you have a measure of how many votes were wasted in a state. For instance if the Rs consistently win 80% of a state’s districts 53/47 and the Ds win 20% of a state’s districts 20/80, then the Ds waste a lot of votes both in the winning districts (80% instead of 51%) and in the losing districts (47%). When you apply this measure to real elections it quantifies just how much the Republicans successfully gerrymandered a partisan advantage in states like Pennsylvania or Ohio.

It’s a simple measure, but it’s still complicated to explain to people. So I took a crack at quantifying something even simpler, the gap between popular vote and congressional seats. For instance in North Carolina last year Republicans took 53% of the popular vote for House of Representatives. However they took 10 of 13 seats, 77%. That’s a pretty remarkable spread of 24%. That outcome spread is a different measure than the efficiency gap, much less subtle, but it’s one I think people can relate to pretty easily.

What happens if I add that spread up over all 50 states’ elections and graph it over time? I scraped data from Wikipedia pages and came up with this:

There you go. It shows that in 2016 the Republicans got 55% of seats with about 50% of the popular vote, a spread of +5%. Back in 1990 they got only 39% of the seats with 46% of the vote, a spread of -7%. It’s the spread between the two that I’m interested in, so let’s graph the spread directly:

Note the graph of the spread is more or less correlated with the blue line above, the % of seats the Republicans got. I’m not sure what that means but it seems important.

### Visualization Failure: Oversimplified

My main conclusion from this exercise is it’s a failure. I’ve oversimplified. there’s no obvious story here. I think any fine point about redistricting gets lost in the broader story of swings between Republican and Democrat popularity.

I think it’s a mistake to add up the votes and seats across different states. Every state is a different story. North Carolina had a virulently partisan districting which probably explains its huge gap in favor of Republicans. California has an independent non-partisan commission designing its districts, so a different process, although the outcome in 2016 still seems somewhat to benefit Democrats (62% popular vote, 73% of seats).

It might also be a mistake to assume all Republicans and Democrats are the same. I mean as partisan as the United States has become, Representatives are still elected to represent a local community and in many cases are known to their constituents. Congresspeople have individual agenda and views and don’t just reflect their R/D affiliation.

So back to the drawing board. I should consider just redoing visualizations from the efficiency gap data Stephanapolous and McGhee collected.

### Update

There’s some nice visualizations of the efficiency gap in this paper from the Campaign Legal Center

Time series view:

Per-state view. (Darker color = bigger efficiency gap. Larger square = bigger state.)

.

California introduced the independent commission relatively recently, I wonder if these measures show any noticeable change compared to the pre-commission regime.