Research Summary: Towards stratified sampling for redistricting plans

Recent research introduces a novel approach for stratified sampling in redistricting plans, addressing challenges in evaluating rare events and sampling complex target measures within high-dimensional combinatorial phase spaces. The methodology involves constructing and diagnosing candidate strata by clustering districts into 'letters' and forming 'words' at a plan level, facilitating soft assignment of plans to strata and estimation of stratum masses and overlap.

Why it matters

This research is strategically important as it enhances the precision and reliability of complex sampling processes, particularly in high-dimensional spaces like those encountered in planning and resource allocation. Improving the ability to evaluate rare events and sample complex target measures can lead to more robust and equitable outcomes, reducing potential biases in critical decision-making processes.

Key insights

• Algorithmic advancements have accelerated the sampling of redistricting ensembles (balanced graph partitions).
• Evaluating rare events and sampling complex target measures remains a significant challenge due to high-dimensional and combinatorial phase spaces.
• The research proposes a prerequisite for stratified sampling by constructing and diagnosing candidate strata.
• Districts are clustered into representative 'letters' which are then used to form plan-level 'words' based on observed plans.
• A partition of unity over these 'words' enables soft assignment of plans to strata.
• The method allows for the estimation of stratum masses and an overlap-induced flux matrix.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.18741