A New Methodology for Evaluating Redistricting Plans

Abstract: Decennial redistricting is a critical element of American representative democracy. Previous studies have found that redistricting influences turnout and representation. From a public policy perspective, redistricting is also potentially subject to partisan gerrymandering. To address this concern, researchers have proposed numerous remedies including compactness and partisan symmetry requirements. However, there exist severe methodological challenges for the scientific evaluation of redistricting plans as well as the rules and constraints that govern the redistricting process. In particular, the number of possible redistricting plans grows exponentially as researchers consider smaller geographical units such as precincts and census blocks. To overcome this difficulty, we propose a flexible new methodology for simulating a representative sample of redistricting plans from a target population. We reformulate the task of drawing district boundaries as the problem of partitioning a graph into separate units. Based on this insight, we extend and apply a Markov chain Monte Carlo algorithm developed in the field of computer vision. Through simulation and empirical studies, we demonstrate that the proposed algorithm outperforms the existing approaches. Applying this new methodology, we examine the ability of various proposed criteria to limit partisan manipulation of redistricting.