Bringing Spatial Interaction Measures into Multi-Criteria Assessment of Redistricting Plans Using Interactive Web Mapping

TL;DR

This paper introduces a novel spatial interaction measure, the interaction ratio (IR), to evaluate how well redistricting plans capture communities of interest using mobility data, and presents an interactive web tool for analysis.

Contribution

It proposes using spatial interaction communities and the IR metric for redistricting assessment, along with biasing methods for plan generation and an interactive web mapping platform.

Findings

Higher IR correlates with increased plan proportionality.

Biasing ReCom can produce plans with desired IR levels.

Spatial interaction measures complement traditional criteria.

Abstract

Redistricting is the process by which electoral district boundaries are drawn, and a common normative assumption in this process is that districts should be drawn so as to capture coherent communities of interest (COIs). While states rely on various proxies for community illustration, such as compactness metrics and municipal split counts, to guide redistricting, recent legal challenges and scholarly works have shown the failings of such proxy measures and the difficulty of balancing multiple criteria in district plan creation. To address these issues, we propose the use of spatial interaction communities to directly quantify the degree to which districts capture the underlying COIs. Using large-scale human mobility flow data, we condense spatial interaction community capture for a set of districts into a single number, the interaction ratio (IR), which can be used for redistricting plan evaluation. To compare the IR to traditional redistricting criteria (compactness and fairness), and to explore the range of IR values found in valid districting plans, we employ a Markov chain-based regionalization algorithm (ReCom) to produce ensembles of valid plans, and calculate the degree to which they capture spatial interaction communities. Furthermore, we propose two methods for biasing the ReCom algorithm towards different IR values. We perform a multi-criteria assessment of the space of valid maps, and present the results in an interactive web map. The experiments on Wisconsin congressional districting plans demonstrate the effectiveness of our methods for biasing sampling towards higher or lower IR values. Furthermore, the analysis of the districts produced with these methods suggests that districts with higher IR and compactness values tend to produce district plans that are more proportional with regards to seats allocated to each of the two major parties.