Statistics Seminar

Abstract: Redistricting is essential to representative democracy.  In the United States, Congressional district lines are often drawn by partisan actors, raising concerns about gerrymandering.  Traditionally, political scientists evaluated a redistricting plan by comparing it with plans from other states or previous plans of the same state.  Such comparison across states and over time, however, suffers from confounding bias due to differences in political geography and redistricting rules.  To overcome this problem, I will introduce a new Monte Carlo simulation algorithm to detect gerrymandering and evaluate redistricting plans.  The proposed simulation approach allows one to compare a redistricting plan against a representative sample of alternative plans that could be generated under a set of specified redistricting criteria.  I will discuss how this simulation-based approach has influenced court cases, including the racial gerrymandering cases in Alabama and South Carolina that reached the Supreme Court of the United States.  I will also present the simulation-based evaluation of the 2020 Congressional maps across 50 states.  The analysis shows that partisan gerrymandering is widespread in the 2020 redistricting cycle, but most of the bias cancels at the national level, giving Republicans approximately two additional seats.  The proposed methodology is implemented as part of open-source software packages that help researchers with data ingestion, algorithm implementation, and visualization of results. This project is part of the Algorithm-Assisted Redistricting Methodology (ALARM) Project (https://alarm-redist.org).