I've been working with the author of autoredistrict on an academic paper on gerrymandering, which we have just finished. You can find a preprint copy of our work here along with all the source code and data for our analysis. In this post, I'll give a quick summary of what we did. First, we developed a statistical framework for assessing gerrymandering based on Bayesian analysis. Given the results of at least one election, we are able to estimate a statistical model that represents plausible outcomes that could have occurred if we were able to re run elections as many times as we wanted. We generate sample elections and compute the bias in each one, which gives us a pretty clear idea of the tendency of the redistricting plan in a state to disadvantage one party or the other.
There are a number ways to measure bias using election results that we have discussed previously, but for the paper we proposed a new metric called the specific asymmetry. The specific asymmetry measures the discrepancy in seats won by either party under a reversal in the statewide popular vote. For example if one party wins 75% of the available seats with 60% of the vote, the result would be symmetric if that same party would have won 25% of the available seats if it had gotten 40% of the vote. The specific asymmetry increases when one party is packed into a few districts where they win by large margins while the other party wins its districts by narrower margins, so it measures a structural disadvantage to one party regardless of whether or not a statewide reversal in the popular vote is something that might actually occur.
More discussion on the methods can be found in the paper but now I'll jump right into an example. Below are histograms of the specific asymmetry from 10,000 simulated elections for California in the 2000s and the 2010s. Negative values for the asymmetry favor Democrats, while positive values favor Republicans. In the 2000s, the average value of the specific asymmetry was about 5 seats favoring Democrats, with a very small probability that the specific asymmetry would be zero or favor Republicans. This is indicative of some pretty aggressive gerrymandering. For the 2010 cycle, an citizens commission was implemented in California as Arnold explains here, and there was a significant reduction in the asymmetry. The California map still favors Democrats on average, an ideal map would have an average asymmetry of zero, but the citizen commission was a big step in the right direction.
We looked at asymmetry in U.S. congress from 1972-2016, and the amount of gerrymandering we found was pretty staggering. Since 1972, about 6% of available seats in congress were won by an unfair advantage caused by gerrymandering in every election. This number has held fairly constant over time, but that is not to say that the way gerrymandering has affected the composition of congress and the rights of voters has also held constant. In the 70s, 80s, and 90s, the Democrats were the primary culprits, while in the 2000's both parties gerrymandered about the same amount in their respective territories which caused the net effect on congress to roughly cancel.
In the current cycle, a number of states saw modest but meaningful reductions in gerrymandering compared to what they had in previous cycles, while a few other states have gone in the other direction. The combination of the concentration of Democratic voters in urban areas and the fact that voters don't change their minds very much create the perfect conditions for sophisticated redistricting algorithms to carve out extremely aggressive gerrymanders, and many states have done so. In terms of average asymmetry as a percentage of available seats, Republicans have gerrymandered Pennsylvania, North Carolina, and Georgia more right now than any other state since 1972. Honorable mentions also go to Virginia, Michigan, Ohio, and Wisconsin, which are extreme gerrymanders by historical standards, and also favor Republicans. In other words, the notorious REDMAP program paid off bigly.
Below are maps of the average asymmetry simulated for every state and each redistricting cycle. In the first gif, the asymmetry is shown in terms of seats. In the second, the asymmetry is shown as a percentage of available seats.
Average asymmetry in each state and cycle in terms of seats:
The average asymmetry as a percentage of available seats:
The technology and data needed to draw fair maps or unfair maps are exactly the same. The changes in gerrymandering from the 2000s to now show that there is hope for reform, but there is serious risk that gerrymandering will get worse if it isn't put in check. If you want to lobby your state legislators to turn over the power of redistricting to an independent commission, you can use this form letter to help you do that.