Mathematically Quantifying Non-responsiveness of the 2021 Georgia Congressional Districting Plan

Zhanzhan Zhao, Cyrus Hettle, Swati Gupta, Jonathan Christopher Mattingly, Dana Randall, Gregory Joseph Herschlag

TL;DR

This paper introduces a novel computational method to mathematically assess partisan gerrymandering by generating and analyzing large ensembles of "fair" redistricting plans. Applying this to Georgia's 2021 congressional map, they found it to be remarkably unresponsive to shifts in voter preference, reliably producing the same partisan outcome regardless of the popular vote.

What Problem Does This Paper Address?

Auditing political district maps for partisan gerrymandering is a complex challenge. A common approach involves comparing an enacted map to a large ensemble of "neutral" or "policy-compliant" maps. The goal is to see if the enacted map's partisan outcomes are an outlier compared to what would be expected from a fair process. However, generating these ensembles, especially when trying to adhere to specific policy preferences (like compactness), is computationally very difficult. Existing methods struggle with sampling from these complex distributions, leading to slow convergence or an inability to explore the full space of possibilities. This paper tackles this computational bottleneck, aiming to make the analysis of gerrymandering more robust and scalable, particularly for statewide redistricting plans.

Methods

The core of this work lies in developing an "accelerated sampling" technique. They combine Metropolized ReCom chains (a standard method for generating redistricting plans based on preferences) with parallel tempering. Parallel tempering is a technique used to bridge the gap between distributions that are easy to sample from and more complex target distributions. The authors' key innovation is the novel application of parallel tempering with recombination techniques to statewide precinct graphs. This combination allows them to more effectively sample from distributions that prioritize specific redistricting criteria, like compactness, which were previously computationally intractable. They also employ a "heat bath" method, using samples from a simpler distribution to help the more complex chains mix faster.

Datasets

The primary dataset used is the precinct-level geographic and voting data for Georgia. This data is used to construct the redistricting graph and to simulate election outcomes under various scenarios. The authors analyze historical election data from 2016-2020 for various statewide races (Presidential, Senate, Governor, etc.) to evaluate the responsiveness of the enacted plan and their generated ensembles.

Key Insights and Ideas

  • Computational Bottleneck Solved: The paper demonstrates a significant advancement in overcoming the computational hurdles of generating redistricting ensembles that adhere to specific policy preferences.
  • Georgia's Unresponsive Map: The 2021 Georgia congressional plan stands out as being exceptionally "non-responsive." This means it consistently produces the same partisan outcome (9 Republicans, 5 Democrats) even when simulated statewide vote shares shift considerably.
  • Polarization as the Driver: The non-responsiveness isn't just random; it's driven by the way districts are drawn. The analysis points to increased polarization, where districts that could be competitive are instead made safely partisan, effectively insulating the outcome from voter will.
  • Packing and Cracking: Specific districts are identified as being "packed" with voters of one party (e.g., Democrats in districts 4, 5, and 13) or "cracked" (e.g., Republicans in district 9), which contributes to the overall non-responsiveness.
  • VRA Considerations Don't Explain It: The authors explore whether the Voting Rights Act (VRA) requirements for minority-electing districts could explain the non-responsiveness. Their analysis suggests that even when accounting for VRA-compliant districts, the enacted plan's behavior remains atypical, indicating that partisan considerations are the primary driver.

Key Results

  • The 2021 Georgia congressional plan is projected to reliably elect 9 Republicans and 5 Democrats across a wide range of simulated voting patterns.
  • Only 0.12% of the generated ensemble of non-partisan plans exhibited the same level of non-responsiveness as the enacted plan.
  • The analysis identified specific districts that are significantly more polarized in the enacted plan compared to the ensemble, with fewer competitive districts overall.
  • The developed parallel tempering method with recombination proved effective for statewide precinct graphs, a scale previously unaddressed by similar techniques.

Limitations and Future Work

  • Spanning Tree Bias: The authors acknowledge that their ensemble is still somewhat weighted towards plans with a higher number of associated spanning trees, suggesting further work is needed to fully overcome this limitation.
  • Spatial Analysis Tool: They mention a new tool for spatially matching districts between ensembles and enacted plans, which they plan to utilize in future studies.
  • General Assembly Plans: The authors have also applied their methods to Georgia's State House and Senate plans, finding similar issues of non-responsiveness and polarization, indicating the portability of their approach.
  • Enforcing VRA Districts: While they explored the impact of VRA districts, they note that further investigation into the effects of enforcing specific numbers of near-majority-minority districts could be a future direction.

Summary updated

Abstract

To audit political district maps for partisan gerrymandering, one may determine a baseline for the expected distribution of partisan outcomes by sampling an ensemble of maps. One approach to sampling is to use redistricting policy as a guide to precisely codify preferences between maps. Such preferences give rise to a probability distribution on the space of redistricting plans, and Metropolis-Hastings methods allow one to sample ensembles of maps from the specified distribution. Although these approaches have nice theoretical properties and have successfully detected gerrymandering in legal settings, sampling from commonly-used policy-driven distributions is often computationally difficult. As of yet, there is no algorithm that can be used off-the-shelf for checking maps under generic redistricting criteria. In this work, we mitigate the computational challenges in a Metropolized-sampling technique through a parallel tempering method combined with ReCom[11] and, for the first time, validate that such techniques are effective on these problems at the scale of statewide precinct graphs for more policy informed measures. We develop these improvements through the first case study of district plans in Georgia. Our analysis projects that any election in Georgia will reliably elect 9 Republicans and 5 Democrats under the enacted plan. This result is largely fixed even as public opinion shifts toward either party and the partisan outcome of the enacted plan does not respond to the will of the people. Only 0.12% of the ∼ 160K plans in our ensemble were similarly non-responsive.
BibTeX
@inproceedings{zhao2022mathematically,
  author = {Zhanzhan Zhao and Cyrus Hettle and Swati Gupta and Jonathan Christopher Mattingly and Dana Randall and Gregory Joseph Herschlag},
  title = {{Mathematically Quantifying Non-responsiveness of the 2021 Georgia Congressional Districting Plan}},
  booktitle = {EAAMO},
  year = {2022},
  url = {https://openreview.net/forum?id=_bI0fZb1MuS},
}

Loading related content...

Figure