Quantifying Gerrymandering: A Case Study of Redistricting in North Carolina

North Carolina has a storied history with gerrymandering. The 2016 congressional redistricting was accused of being a partisan gerrymander, in part due to the discrepancy between the statewide popular vote (close to a 50/50 split between Republicans and Democrats) to the actual outcome (9 Republicans and 4 Democrats) in 2016. Given the geographical nature of the problem, this is not clear: can we quantify the extent of the gerrymandering in an interpretable way?

The analysis present in this work came from a 2016 summer project for Duke’s Data+ Program, which itself was a continuation of other undergraduate research efforts previously performed. The original goal was to develop rigorous evidence to determine whether a given redistricting was a partisan gerrymandering; to do this, an MCMC scheme was used various redistrictings of North Carolina based on a Gibbs measure that took into account traditional redistricting criteria: as equal as possible population, geographic compactness, VRA satisfaction, and preservation of municipal (county) boundaries. This work introduced a number of important ideas in the study of partisan gerrymandering , namely:

• Basing quantitative analysis with respect to a sampled ensemble of redistricting plans rather than global baselines, as done with past measures such as the efficiency gap.

• Interpretable plots that can easily illustrate whether a given redistricting plan is a partisan gerrymander

The initial version of this work can be found here, and the updated version in preparation for journal submission can be found here. My primary role in this paper was as a summer mentor for Han Sung and Justin.