Does Federal Disaster Assistance Follow the Damage?

Author: Claude (TerraPulse Lab) Status: Draft — hed/dek SIGNED OFF (Mike, 2026-09-04); Frame A (statutory motivation) applied Hed: Does Federal Disaster Assistance Follow the Damage? Dek: Eligibility flows through the declaration record by statute. Thirty years of measured storms show that record captures whether a county was struck — and almost none of how much. Created: 2026-09-04 Rotation: paper type 7 (county emergency managers, R family)

Hypothesis

Does a county's FEMA declaration history track its measured severe-weather burden? Two testable readings: declarations follow burden (counts rise with measured damage), or declarations are a threshold instrument (broad coverage, weak proportionality). H0: county declaration count is unrelated to measured burden (Spearman ρ = 0).

Data Sources

• fema eventdex — 5,262 declarations FY 1953–2026; window FY 1996–2025 = 3,832 declarations, 54,475 designated areas (98.2% county-equivalent, 3,723 distinct counties).
• Peril dexes at county grain (NCEI CZ_TYPE C), 1996–2025: hail 419,106 · wind 564,553 · heavy_rain 31,751 · wildfire 1,749 · tor 74,463 segments = 1,091,622 rows, $93.3B parsed damage (K/M/B suffixes handled).
• Zone-grain (state-level only): winter, cold, heat, drought_report — 478,385 rows, $52.5B. Flood: no adequate measured counterpart held.

Methodology

Name-normalized county×state join between NCEI cz_names and FEMA designated_areas. Five divergence classes found and repaired (spacing, "(C)" suffix, paren-free cities, nested MSA parentheticals, consolidated governments) — the apparent never-declared damage share fell 2.17% → 0.05% across the repairs (pre-repair figure archived in results.json). Non-county designation units (Maine townships, reservations, school districts) excluded from the panel per Sasha review. Outer-join panel of 3,330 keys (3,189 both · 75 measured-only · 66 declared-only); tests on the 3,264 with measured events. Pearson (log10(1+x)) + Spearman, 12 tests, Bonferroni 0.0042; analytic detection floor |ρ| > 0.050. Sensitivity: damage- vs count-based burden; convective-only declarations; ex-Biological (unchanged to 4 decimals); city-merge exclusion (ρ 0.148→0.143); window 2000-2025 (ρ=0.133).

Findings

1. Coverage null (the pitch inverted): 99.95% of measured damage lies in declared counties. Every zero-declaration key is an AK/PR/territory zone label or a linkage artifact; zero verified CONUS counties with measured convective damage lack a declaration.
2. Weak proportionality: burden explains ≤6.7% of variance in declaration count (best: events vs convective declarations, ρ=0.280, p=7e-60). All 9 tests survive Bonferroni; all effects small.
3. Hail nearly invisible: 37.7% of measured damage, ρ=-0.027 vs all-type declarations (n.s., below the 0.050 floor); ρ=0.091 vs convective declarations — detectable, one-third of tornado's coupling. Figure 1: median declarations per damage decile = 11, 14, then 15-16 flat across the upper eight deciles.
4. Composition mismatch: measured tor 39%/hail 38%/wind 14% vs declarations Fire 41%/Severe Storm 26%/Hurricane 10%; drought = 3 in-window declarations against $31.3B state-grain damage (cross-ref drought-relief-gap).

All numbers: data/results.json (generated by scripts/analyze.py).

References

• workspace drought-relief-gap (type 2) — the drought-specific mismatch
• workspace crop-loss-by-peril (type 1) — crop damage shares
• FEMA OpenFEMA Disaster Declarations Summaries; NOAA NCEI Storm Events