TL;DR
This paper introduces a novel computational method to rigorously quantify partisan gerrymandering by generating large ensembles of "fair" redistricting plans. They apply this method to the 2021 Georgia congressional districting plan, finding it to be highly non-responsive to voter preferences, reliably producing the same partisan outcome regardless of election results.
What Problem Does This Paper Address?
Auditing political district maps for partisan gerrymandering is crucial for democratic representation. A common approach involves comparing an existing map to a large ensemble of "neutral" or "policy-compliant" maps. The challenge lies in computationally generating these ensembles, especially when trying to match specific redistricting policies (like compactness) that are difficult to sample from directly. Existing methods often struggle with computational complexity and slow convergence, particularly at the scale of statewide precinct data. This paper tackles this by developing and applying an improved sampling technique to analyze the 2021 Georgia congressional districting plan.
Methods
The authors employ a Metropolized-sampling technique enhanced with parallel tempering. This approach aims to overcome the computational difficulties of sampling from complex distributions that reflect redistricting policies.
- Parallel Tempering: This is a standard MCMC technique used to bridge distributions that are easy to sample from with more complex target distributions. It involves running multiple chains at different "temperatures" (representing different distributions) and allowing them to exchange states.
- ReCom (Recombination): This is a specific type of sampling method for redistricting plans that uses tree-based approaches and diffusive boundaries. The authors combine parallel tempering with recombination, which they claim is a novel approach.
- Accelerated Sampling: They adapt the parallel tempering scheme to be more computationally feasible for statewide precinct graphs. This involves using a "heat bath" method where chains at the target distribution instantaneously mix with a base distribution that is easy to sample from.
- Compactness Tuning: A key aspect is tuning the sampling to produce ensembles with similar compactness (measured by the Polsby-Popper score) to the enacted plan, ensuring a fair comparison.
Datasets
The primary dataset used is the precinct-level geographic and voting data for Georgia, specifically for the 2021 congressional districting plan. The analysis also incorporates historical election data from 2016-2020 for various statewide races (Presidential, Senate, Governor, etc.) to evaluate the responsiveness of the districts.
Key Insights and Ideas
- Computational Bottleneck Solved: The paper demonstrates a practical solution to the computational challenges of sampling from policy-driven redistricting distributions at a large scale, making such analyses more accessible.
- Georgia's Plan is Highly Non-Responsive: The core finding is that the 2021 Georgia congressional plan is remarkably rigid. It consistently elects the same partisan outcome (9 Republicans, 5 Democrats) across a wide range of simulated election results, failing to reflect shifts in voter preference.
- Polarization as the Cause: The non-responsiveness is attributed to the polarization of voters in competitive districts. The plan appears to have "packed" Democratic voters into some districts and "cracked" them in others, solidifying partisan leanings and reducing responsiveness.
- Novelty in Methodology: The combination of parallel tempering with recombination for redistricting analysis, especially at the statewide precinct level, is presented as a novel contribution.
Key Results
- The 2021 Georgia congressional districting plan reliably elects 9 Republicans and 5 Democrats across a broad spectrum of simulated voting patterns.
- Only 0.12% of the generated ensemble of "fair" plans exhibited the same level of non-responsiveness as the enacted plan.
- The analysis identified specific districts (e.g., the 4th, 5th, 13th for Democrats and the 9th for Republicans) where partisan packing and cracking were most evident.
- The developed parallel tempering method successfully sampled ensembles with comparable compactness to the enacted plan, a feat previously difficult.
Limitations and Future Work
- Compactness Weighting: The authors note that their ensemble is still weighted towards plans with higher numbers of associated spanning trees, suggesting further work is needed to overcome this limitation.
- VRA Consideration: While they explored the impact of Voting Rights Act (VRA) districts, they acknowledge that their initial analysis did not fully account for VRA considerations, and further investigation into how VRA compliance might affect responsiveness is warranted.
- Spatial Analysis Tools: They mention a new tool for matching districts spatially between ensembles and enacted plans, which they plan to utilize in future work.
- Generalizability: The authors suggest their methods are portable to other states and have already been applied to Georgia's General Assembly plans, indicating future research directions.
Summary updated