Replication Package for "Equity-Aware Interpretable Policy Rules for Personalized Hurricane Risk Communication"
Authors/Creators
Description
This repository contains the supplementary replication package for the manuscript “Equity-Aware Interpretable Policy Rules for Personalized Hurricane Risk Communication.” The package provides the computational materials needed to reproduce the analysis pipeline, policy-learning results, manuscript tables, and publication-quality figures associated with the study.
The manuscript develops a Fairness-constrained Interpretable Distillation Tree (FID-Tree) framework for personalized hurricane risk communication. The framework combines multi-arm causal forests, policy learning, demographic parity auditing, and shallow interpretable decision trees to derive transparent and equity-aware message assignment rules.
The raw survey microdata are not redistributed in this Zenodo archive. Instead, the analysis uses the publicly available DesignSafe-CI dataset:
Retchless, D., & Ross, A. (2022). Hurricane Risk Perceptions of U.S. Gulf Coast Residents in 2017. DesignSafe-CI. https://doi.org/10.17603/ds2-s7wv-pn32
Users should download survey_results.csv from the DesignSafe-CI repository and place it in data/survey_results.csv or in the project root before running the notebook. The supplementary README explicitly notes that the trial microdata file is not redistributed when subject to data-use restrictions and provides the expected workflow for placing the file and running the notebook.
The archive includes the Jupyter notebook, supporting scripts, dependency file, and instructions needed to regenerate the reported outputs. A successful replication produces manuscript figures, tables, robustness checks, fairness diagnostics, and the consolidated results workbook. The required Python dependencies include packages such as numpy, pandas, scikit-learn, statsmodels, econml, fairlearn, matplotlib, and seaborn.
Files
Additional details
Dates
- Submitted
-
2026-05-26