Authors: Farhan Sadeek, Jalen Francis, Jayson Clark
Institution: The Ohio State University
Competition: IM²C 2025 (Intercollegiate Math Modeling Challenge)

This repository contains a complete computational framework for:

1. Objectively identifying Communities of Interest (COIs) using demographic, socioeconomic, and commuting data
2. Quantifying community fragmentation using information-theoretic metrics
3. Evaluating redistricting fairness through ensemble comparison against neutral maps

Applied to North Carolina's 2023 congressional map (SB 757), our analysis reveals systematic community splitting: 5 of 30 identified communities exhibit CSI scores > 0.91 (near-maximum fragmentation), with the enacted map splitting communities more severely than 89% of algorithmically-generated neutral alternatives.

.
├── README.md                      # This file
├── requirements.txt               # Python dependencies
├── script.py                      # Main COI identification & evaluation
├── gerrychain_analysis.py         # Ensemble comparison analysis
├── report.tex                     # Full LaTeX report source
├── data/                          # Data files (downloaded by scripts)
├── results/                       # Generated visualizations & results
│   ├── coi_map.png
│   ├── coi_splits_map.png
│   ├── ensemble_comparison.png
│   └── coi_results_complete.csv
└── .gitignore                     # Git ignore patterns

• Data Sources:
  • 2020 Decennial Census (P.L. 94-171)
  • American Community Survey 5-Year Estimates (2016-2020)
  • LEHD Origin-Destination Employment Statistics (LODES)
• Features (9 dimensions):
  • Demographics: Hispanic %, NH White %, NH Black %, NH Asian %
  • Socioeconomics: Median income, homeownership rate, bachelor's degree %
  • Commuting: Jobs in tract, workers in tract
• Algorithm: Spatial K'luster Analysis by Tree Edge Removal (SKATER)
  • Ensures spatial contiguity
  • Minimizes within-cluster variance
  • k = 30 communities

• Formula: Pielou's Evenness Index adapted to redistricting

  CSI = H / H_max = -Σ(p_i log₂ p_i) / log₂(N)
  

  where p_i = proportion of COI population in district i
• Interpretation:
  • CSI = 0: Community intact (not split)
  • CSI ≈ 1: Community maximally fragmented

• Algorithm: Recombination (ReCom) Markov chain
• Constraints: Population equality (±5%), district contiguity
• Sample Size: 100 neutral maps
• Result: SB 757 at 89th percentile of CSI distribution

• Python 3.10 or higher (required for GerryChain)
• ~2GB disk space for data files

1. Clone the repository:

   git clone https://github.com/jalenfran/immc_2025.git
   cd immc_2025

2. Create virtual environment:

   python3.12 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

python script.py

Runtime: 5-10 minutes
Outputs:

• coi_map.png - Map of 30 identified communities
• coi_splits_map.png - COIs overlaid with SB 757 districts
• coi_results_complete.csv - CSI scores for all 30 communities

python gerrychain_analysis.py

Runtime: 10-30 minutes (generates 10,000 MCMC steps)
Outputs:

• ensemble_comparison.png - Histogram comparing SB 757 to neutral maps
• ensemble_csi_scores.csv - CSI scores for all 100 ensemble maps
• Statistical significance results (percentile rank, z-score)

Note: Requires Python 3.10+ due to GerryChain type hint syntax.

All data is automatically downloaded by the scripts from public sources:

1. Geographic Boundaries:

   • TIGER/Line Shapefiles (2020)
   • NC SB 757 Districts (2023)

2. Demographic Data:

   • Decennial Census P.L. 94-171
   • American Community Survey (ACS) 5-Year Estimates

3. Commuting Data:

   • LEHD LODES Version 8.0

• Geospatial: geopandas, pyogrio, shapely
• Data Access: cenpy (Census API wrapper)
• Spatial Analysis: libpysal, spopt (SKATER implementation)
• Ensemble Generation: gerrychain (ReCom MCMC)
• Visualization: matplotlib

See requirements.txt for complete list with version constraints.

30 spatially-contiguous communities identified across North Carolina, ranging from 18,728 (rural) to 2,427,537 (metro) residents.

• SB 757 Mean CSI: 0.5050
• Ensemble Mean CSI: 0.4351 (σ = 0.0563)
• Percentile: 89% (SB 757 worse than 89% of neutral maps)
• Z-score: 1.24

All analysis is fully reproducible:

• Scripts automatically download data from public sources
• Random seed fixed where applicable (SKATER uses deterministic algorithm)
• Complete dependency versions specified in requirements.txt
• All code commented and documented

If you use this methodology in your work, please cite:

Sadeek, F., Francis, J., & Clark, J. (2025). Beyond the Line: A Data-Driven Framework 
for Quantifying Community and Evaluating Redistricting Fairness in North Carolina. 
IM²C 2025. https://github.com/jalenfran/immc_2025

MIT License - See LICENSE file for details.

For questions or collaboration:

• Farhan Sadeek: sadeek.1@osu.edu
• Jalen Francis: francis.628@osu.edu
• Jayson Clark: clark.4101@osu.edu

• Data: U.S. Census Bureau, LEHD Program
• Software: PySAL, GerryChain, GeoPandas communities
• Methodology: Assunção et al. (2006), Pielou (1966), DeFord et al. (2021)