Redistricting forever

Grading the continuing stream of maps, partisan and otherwise

It’s Groundhog Day for redistricting, over and over…

Summer has been flying by. Our Electoral Innovation Lab workshop on the science of democracy reform will pass on some great work to us perma-Labbers.

In the area of redistricting, our work was punctuated by new incoming events: the refusal of Alabama Republicans to comply with a Supreme Court order to draw a new majority-black Congressional district, and partisan redistricting lawsuits in New York, Utah, and now New Mexico. Despite the record heat, it’s an unending winter of offenses - Groundhog Day.

But we’ve adapted. To tell how, I turn the reins over to Ethan Arsht. Ethan is visiting from the University of Chicago, where he is a master’s student in computational public policy. He’s been working on the Princeton Gerrymandering Project’s analytics, examining the last wave of maps passed in 2022 and 2023.

Here’s his take on partisan and racial gerrymandering, especially in South Carolina, where the two offenses converge. -Sam

This past month, I’ve spent nearly every day dissecting and analyzing district maps for the House of Representatives and state legislatures. These efforts continue the work of the Princeton Gerrymandering Project to analyze redistricting plans. Hard metrics give advocates, journalists, and concerned citizens an accessible and robust tool to understand and grade maps.

As readers of this blog are well aware, partisan gerrymandering has declined somewhat in the current redistricting cycle since its peak in the 2010s. Yet even after a combination of voters, redistricting commissions, and courts pushed back, a few extreme gerrymanders persist. And depending on one’s definition, as many as ten “little gerrymanders” have popped up.

To evaluate gerrymanders, an increasingly widespread tool among analysts is to compare a proposed or enacted district plan to an “ensemble” of simulated maps. An ensemble is a collection of hundreds of thousands of alternatives generated using a computational tool called a Monte Carlo Markov chain. The goal of this approach is not to draw the perfect map, or even to draw a fair map. Rather, the ensemble provides an approximation of the entire range of possible maps that are allowed under federal and state law. Within this ensemble I can place a real human-drawn map to see if it is extreme by one measure or another. If it’s extreme, the human may have prioritized one goal (for example partisan advantage) to the exclusion of others (for example, respecting where communities live). The simulation algorithm can be held to the same criteria as the human such as not drawing crazy shapes or making sure there are minority-representing districts as required by federal law. 

Let’s take South Carolina, where there is a current lawsuit, as an example. I simulated 100,000 maps of South Carolina’s congressional districts using computational tools from the ALARM Project. I told the algorithm to:

• make each district contiguous and equal in population to others;

• avoid county splits when possible; and

• draw a map that contains a Voting Rights Act-compliant district.

Now I’ll analyze the enacted Congressional map for partisan fairness. In a partisan world, limiting the power of either party is a leading motive for gerrymandering. And in the 2022 South Carolina election: Republicans won six out of seven House seats (86%), even though Republican candidates only won 66% (about two-thirds) of the votes.

But percentage comparisons are too naive an approach. In a district-based system, the majority party tends to have a natural advantage, and geographic concentration of voters by partisan preference can enhance this advantage. This is where ensembles come in handy. By comparing the real districting plan to computer-drawn ensemble, we have a way to assess whether the plan is likely to have arisen from party-blind principles. 

This chart, which arranges districts in order of partisan preference, reveals an unusual occurrence in Democrats’ best two districts. Their second-best district is less favorable to them than in any of the simulations. And their best district is extreme in the other direction. This suggests that the South Carolina’s state legislature might have designed a map to pack Democratic voters into the top district, leaving the next district down to Republicans. 

The geography bears this out. Using the invaluable Dave’s Redistricting App, we can see that South Carolina’s Sixth District captures Democratic-leaning areas in Charleston and Columbia.

However, combining these cities also reduced the Black voting age population (BVAP) of the Sixth District, a possible gerrymander because it dilutes their voting power. Indeed, one court expert has found that the Sixth District has a far smaller BVAP than simulations did.

The following simulated map, which I picked pretty much at random, has a Black-majority district and leaves Columbia and Charleston whole (and in different districts).

Thus a comparison with an ensemble of simulations shows that South Carolina’s enacted Congressional map minimizes overall Democratic representation, while reducing Black voting power in a key district.

While partisan fairness is a current emphasis area in court cases, we can also evaluate more traditional measures, much as compactness and county splits. These traditional measures echo people’s intuitions about strange shapes. However, they are also easier criteria to meet, even with a gerrymandered map.

Compactness can be measured many ways, and ideals often focus on smooth borders and a close-to-circular shape. At the Princeton Gerrymandering Project, a main measure of compactness is the Reock score, which compares the area of a district with the area of a circle that is drawn to just fit around it. A circle, which defines a Platonic (i.e. impossible; who would be represented in the hypocycloid between three districts?) ideal, has a Reock score of exactly 1.

I compared the average Reock score of the enacted South Carolina map to the ensemble. The map is in the middle of the pack, not exceptional.

County splits are even easier to calculate: how often is a county split into two or more districts? Some county splits are unavoidable in order to achieve equal population. Compared to simulations, the ten county splits in the enacted plan are in the middle of the pack. However, the simulations also show that it’s possible to draw a plan with half as many splits.

All this is to say that by traditional measures based on shape, the South Carolina congressional map looks okay. It goes to show that behind a pretty map there could still be ill intent. Two of our summer mentors, Bernie Grofman and Jonathan Cervas, have called plans like this stealth gerrymanders.

>

In the closing weeks here, I’m now working on completing the Gerrymandering Project’s report cards - including states with just a few Congressional districts. That’s relevant in Utah and New Mexico. I’ll be glad to show you that in a few weeks over at the Gerrymandering Project website!