Ensemble-Based Big-Data Analytics for Evaluating New Hampshire's Congressional Districting Fairness

Redistricting fairness has become a central concern in U.S. elections, with ensemble-based methods emerging as a leading tool. However, most existing studies focus on highdistrict and two-party voting settings. This paper conducts an ensemble-based, geospatial big data analysis to evaluate the political geography and fairness of congressional districting in New Hampshire, an analytically difficult two-district swing state with a large Independent-voting population, and highlights the unique challenges and solutions for similar scenarios. Our end-to-end pipeline, implemented in Python using custom code built on GeoPandas, GerryChain, NetworkX, MAUP, and NumPy, allows automated data preparation, graph modeling, MCMCbased ensemble generation, vote simulations, spatial visualizations, and ensemble-based diagnostics at scale. We construct a unified multi-election, precinct-level dataset, by preparing and reconciling election and geographic records, and perform a longitudinal analysis of vote-share distributions to characterize the state's evolving partisan landscape. We then develop graphbased models of precinct adjacency and apply the ReCom algorithm to generate thousands of valid plans under various constraints. We, finally, conduct various outlier analyses and assess outcome sensitivity to Independent voter behavior. Our results show that while conventional fairness measures provide inconsistent or inconclusive signals in this two-district setting, spatial diagnostics reveal signs of potential incumbent-favoring redistricting. Simulation experiments further demonstrate the pivotal role of Independent voters in this swing state. We also find that when imposing county-split constraints in ensemble generation, the enacted plan falls toward the high-split end of the distribution. Overall, our integrated methodology offers practical guidance for evaluating redistricting and political geography in New Hampshire and can be readily applied to other challenging and low-district settings, providing a scalable template for datadriven redistricting analysis.