Replication Package for "Measuring Flood-Insurance Payments and Recorded Damage: A Settlement-Basis Audit of Hurricane Harvey Claims"

This repository contains the replication materials for the study “Measuring Flood-Insurance Payments and Recorded Damage: A Settlement-Basis Audit of Hurricane Harvey Claims.” The study examines how settlement-basis differences affect comparisons between recorded building damage and flood-insurance payments in administrative claims data. The replication workflow reconstructs the analytical samples, classifies payment-to-damage states separately for actual-cash-value (ACV) and replacement-cost-value (RCV) settlements, evaluates denominator sensitivity and dollar differences, audits ACV-basis records with payments above recorded ACV damage, and documents the consequences of outcome-dependent sample restrictions. It also includes limited coverage-and-deductible screening-proxy sensitivity analyses and retrospective model-validation diagnostics. The package uses a preserved and checksum-verified OpenFEMA API snapshot of the NFIP Redacted Claims dataset (NfipClaims, version 3), filtered to Hurricane Harvey claims. The snapshot was extracted on 31 July 2026 and is accompanied by the API request manifest, official field metadata, an analysis data dictionary, and SHA-256 checksums. Because the OpenFEMA dataset is dynamically updated, the preserved snapshot—not the current live API response—is the reproducible research input. The package contains: the preserved Hurricane Harvey claims snapshot and source manifests; data-cleaning, sample-construction, descriptive-analysis, sensitivity-analysis, and model-validation code; machine-readable tables and diagnostic outputs; final figures and their source files; the anonymous manuscript and Online Resource; numerical, reference, formula, visual, and release-quality audits; a file manifest and SHA-256 checksum list. The complete workflow can be run from the package root with: python code/run_all.py The verified release environment used CPython 3.13.5 with random seed 20260801. Successful execution writes overall_status=PASS to logs/overall_status.txt after completing the analytical and release checks. This repository supports an administrative measurement audit. It does not provide a formal recalculation of claim entitlement, a legal determination of correct payment, an estimate of household recovery, a fairness or welfare analysis, or a causal estimate of housing-market effects. The included coverage-and-deductible formulas are limited screening proxies and cannot reproduce all policy provisions, exclusions, supplemental claims, recoveries, adjustment histories, or final claim adjudication.