Algorithmic Redistricting: Using Clustering to Construct Fair Alternatives to Partisan Gerrymanders

Abstract

The guiding intuition behind democracy is that every citizen should have an equal voice in the political process, but gerrymandering – the practice of drawing political boundaries to intentionally over-represent or under-represent some political group – threatens this ideal in the United States today. Drawing congressional districts that meet legal requirements and reflect good political values is both mathematically and philosophically difficult. Additionally, large, detailed datasets must be compiled and used to generate districts that could be seriously recommended for public policy. To date there is no absolute consensus on the mathematical tools or political values that should be used to draw congressional districts. This thesis uses k-means clustering on a novel dataset to form coherent communities inside congressional districts, measure the legality of these districts and project election outcomes in these districts. Results show that this method represents public opinion well on the national level and constructs some good districts but additions to the algorithm are necessary before it can completely solve the redistricting problem. Additionally, these results provide interesting insights into minority representation, natural gerrymandering and other related public policy concerns.