Does Federal Disaster Assistance Follow the Damage?
Federal disaster assistance flows through one gate: the FEMA declaration. We joined 3,832 declarations against 1.09 million county-measured storm events carrying $93 billion in recorded damage. The gate opens for essentially everyone — 99.95% of measured damage sits in declared counties — but how often it opens has almost nothing to do with how hard a county is hit. From the 35th damage percentile to the 95th, the median county is declared exactly as often.
The declaration is a gate, not a scoreboard
A federal disaster declaration is not just a record — it is the legal gateway to help. Under the Stafford Act, the principal forms of federal disaster assistance exist only downstream of one: aid to individuals and households, public-infrastructure assistance, and hazard-mitigation grant funding, which is computed as a share of the assistance a declaration unlocks. A county’s declaration history is therefore, by construction, its federal assistance-eligibility history.
So we asked a checkable question: does that history track what measured instruments say actually happened? On one side, every declaration by FEMA (the Federal Emergency Management Agency) from fiscal 1996 through 2025 — 3,832 of them, naming 52,424 county-level designations. On the other, every hail, tornado, thunderstorm-wind, heavy-rain, and wildfire event the NOAA Storm Events record resolves to a county over the same thirty years: 1,091,622 events, $93.3 billion in recorded damage. The comparison of an administrative instrument to a measured record is deliberate — that gap is exactly what we set out to measure.
As a map, the record is nearly perfect
Our working hypothesis was an exposure gap — hard-hit counties the declaration record never noticed. The data killed it, and it took five rounds of forensic record-linkage to be sure. County names diverge between the two federal records in ways that manufacture false gaps: DE SOTO vs DESOTO, independent cities written three different ways, nested Census jargon like “Worcester (County) (in (P)MSA 1120,2600,9240), MA”, consolidated governments like Georgetown–Quitman County. Each repair shrank the apparent gap — from 2.17% of measured damage down to 0.05%, a factor of 43.
What survives isn’t a gap at all. Every remaining “never-declared” key is an Alaska forecast zone, a Puerto Rico region label, or a Pacific territory — bookkeeping units FEMA designates under other names — plus one literal typo (a Kansas county that doesn’t exist). We could not find a single verified county in the lower 48 with measured storm damage on the books and no declaration since 1996. If a county has suffered, the record knows.
As a ledger, it is flat
Median declaration count per measured-damage decile, 3,264 counties, 1996–2025. Decile 1 is the least-damaged tenth of counties; decile 10 the most.
A county at the 95th damage percentile has the same median declaration history as one at the 35th. Statistically, measured damage explains 5.0% of the variance in a county’s declaration count (Spearman ρ=0.148, N=3,264). Matching like to like — convective damage against Severe Storm and Tornado declarations only — lifts that to 6.7% (ρ=0.280). Every one of our twelve tests that reaches significance shows the same shape: a real coupling, and a weak one. The result is robust to excluding pandemic-era declarations (unchanged to four decimals), to dropping city–county name collisions, and to shifting the window.
Hail barely registers
Rank correlation (Spearman ρ) between a county’s measured damage from each peril and its declaration count. Bars marked n.s. are statistically indistinguishable from zero.
Hail carries 37.7% of all measured county-grain damage — $35.2 billion, second only to tornadoes — and its county distribution bears no detectable relation to the full declaration record; at this sample size any true effect is smaller than ρ=0.050. Even against convective declarations specifically, its coupling is one-third of tornado’s. A peril that destroys by accumulation across four hundred thousand events, rather than by single declarable catastrophes, barely writes itself into the assistance record.
Two records, two different countries
Left: where measured county-grain damage comes from. Right: what declarations are issued for, same thirty years.
The two lists barely share a vocabulary. Tornadoes are 39% of measured damage and 1.3% of declarations (tornado damage instead surfaces inside “Severe Storm”). And the drought case, which we documented in The Drought Relief Gap, is the extreme: in this same window drought carries $31.3 billion of zone-measured damage and produced three declarations.
Working as designed — and that’s the point
None of this says FEMA is hiding information or mis-computing anything. The declaration record is public, complete, and accurate about what it records: administrative acts. Declarations fire when single events cross statutory thresholds and a governor asks — they were never designed to be proportional to accumulated measured burden, and they aren’t. What the numbers establish is the size of the difference between the record the assistance system runs on and the record the instruments write: the first answers whether a county has suffered, and almost nothing about how much. Anyone using declaration history as a stand-in for damage history — in mitigation planning, in research — inherits that difference in full.
What this counts, and what it doesn’t
The measured side is convective-and-fire only: winter storms, heat, cold, and drought are recorded against forecast zones rather than counties and enter only our state-level checks, and floods have no adequate county-measured counterpart in our holdings — which is why our like-for-like test restricts the declaration side to matching incident types. Storm Events damage figures are National Weather Service estimates, not audited claims. Event counts partly track population and spotter density; we apply no population correction, and since population pushes both event reporting and declaration frequency up together, the weak correlations we report are if anything on the generous side. The 66 counties with declarations and no measured convective events are consistent with burden from the excluded perils, though we did not test their composition.
Reproducibility
Every number on this page is generated into the results.json file of the declared-vs-measured workspace by its analysis script, including the record-linkage repairs, all twelve correlation tests with Bonferroni correction, and every sensitivity rerun. Declarations are read from the TerraPulse fema dex (5,262 slots, FY 1953–2026); measured events from the county-grain peril dexes. Every result is a count or rank statistic over measured reports and recorded administrative acts. No model, forecast, or interpolated value is used anywhere.
TerraPulse Data Lab, 2026-09-04. Data: FEMA OpenFEMA Disaster Declarations and NOAA/NCEI Storm Events database (US government works, public domain). Statutory framework: Stafford Act, 42 U.S.C. § 5121 et seq.
Published paper
The full scientific paper, with methods, tables, and references.