Thanks to SCOTUS, gerrymandering is back in the news. The term is a century old, so designing districts that are safe — but not too safe — isn’t exactly a modern invention. However, as explained in an excellent episode of the The Daily, big data and machine learning have weaponized our bad intentions.
An obvious response is to fight algorithms with algorithms: take district drawing out of the hands of humans and turn it over to computers.1 The results are aesthetically pleasing, but that Washington Post article identifies Voting Rights Act compliance issues and that optimizing for compactness could make packing even worse.
The plaintiffs in the aforementioned Supreme Court case proposed a less radical, but still technocratic, solution: they define a gerrymandering metric to test the fairness of district boundaries. Jay Cost describes a number of its issues. The most compelling rebuttal is that it doesn’t work well for states with fewer than seven or eight districts; unfortunately, thirty-three states have eight or fewer congressional districts, and any approach that doesn’t work for two-thirds of the country isn’t a serious proposal.
Luckily, we don’t need a gerrymandering-detection metric. A more fundamental solution would work to reduce the influence of district boundaries. But before discussing solutions, we need to understand why things are broken in the first place. CGP Grey explains the issues with first-past-the-post voting. It leads to gerrymandering and two-party systems, while actively disincentivizing third parties. FPTP is part of the problem, so instant-runoff voting would help, but consolidating districts would go even further.
For example, here in Seattle there are two at-large city council seats.2 Candidates select a seat to run for, and thus a subset of opponents to run against, but nothing practically differentiates the seats. This is stupid. It would be much better for all at-large candidates to be in a single pool, with some sort of transferable vote system to select two victors.
The same system could be used at the state level to draw congressional districts. Nothing would change in one-seat states like Montana, Wyoming, or the Dakotas; of course, there’s no gerrymandering problem there either. For states with seven or fewer representatives, all of the districts would be consolidated into a single district that elects multiple representatives. This prevents any district from having fewer than four seats. A second district is added at the eighth representative, and another for each additional multiple of four.
Washington, for example, would have two, five-seat districts, and a natural division would probably follow obvious geography, like everything east of the Cascades plus the counties along the western Oregon border. Given this map, a moderate democrat could coalesce 20 percent of the vote from this otherwise pretty conservative region and get elected. It would also allow the relative conservatives in the Seattle metropolitan area to band together and elect at least one congressperson more representative of their views.
The end result would be a significantly more moderate congressional delegation because it would empower centrist conservatives in liberal bastions and centrist liberals in conservative strongholds. In time, races could break along axes orthogonal to party. Imagine a candidate that represented all the libertarians in SoCal, or someone running on a non-partisan, f**k-Seattle platform in the western Washington district.
I’m sure there would be unforeseen issues, and it would be prudent to start slow — state-by-state — but I can’t imagine this system would be worse than the one we have now.
It has a strong political bent that advocates for basic income as a response to the increasing automation of human jobs, but politics aside, @HumanVsMachine is an interesting follow on Twitter. ↩
There are also geographically aligned seats meant to ensure each neighborhood is represented. ↩