87 — Gerrymandering and Computational Redistricting

Excited to share: @DrLoveBC and I have been working on a big data science project on Gerrymandering and Computational Redistricting and it is finally out! 🌎Please visit site & give feedback if you're from the US.
Website: https://t.co/5oWgLgHIq6
Preprint: https://t.co/BoaaBV5pxY

— Olivia Guest (@o_guest) November 15, 2017

This paper reviews redistricting algorithms that aim to avoid the common pitfalls (or intentionally malicious effects) of gerrymandering. Gerrymandering, which is an extremely good word but extremely bad practice, is an 1812 term for redrawing district lines to manipulate regional vote results. In short, if you know how people will vote, it’s possible to draw new district lines that can amplify the effects of a smaller population (“packing”), or effectively cancel out the votes of a larger population (“cracking”).

While other countries such as Mexico have already designed districts using an impartial algorithm, districting is still performing by (partisan) humans in the USA.

Guest, Kanayet, and Love present a novel algorithm for drawing district lines that both acknowledges federal requirements for districting (disticts should each hold approximately the same number of people) and maintains some level of sanity (people in the same district should probably live close to each other, or have something in common).

Using density and proximity of co-district’d voters as an optimization metric, the authors develop a protocol based on weighted $k$-means clustering. This algorithm dramatically outperforms current districting, and performs even better in larger states with higher populations.

The paper states the (in my opinion, perhaps too generous) hypothesis that human spatial reasoning is partially to blame for the abysmal state of district design; that beyond partisan efforts, human brains are just Not So Good at mapping tasks.

Or maybe impartial tasks just shouldn’t be left to humans.