Executive Summary
FEMA Individual and Households Program approval rates can separate sharply across counties facing the same disaster. Among 181 eligible county-event records from seven Texas disasters, the twenty largest gaps ranged from 28.1 to 47.4 percentage points above or below the rate for all other eligible counties in the event.
This was not a one-event curiosity. Thirteen counties appeared at least five points on the same side of their disaster remainder in two or more events. Liberty was above in four events, Fort Bend below in four, and Newton above in three. Eleven of the thirteen still retained that repeated direction in at least two events after a stricter comparison removed the largest other county from each event's remainder.
The pattern is useful because it narrows a broad monitoring problem to a specific review list. FEMA, Texas, and local applicant-support teams can use the queue to decide where to examine communications, case mix, inspection operations, and processing more closely. They should not use it to infer wrongdoing or automate a judgment. These aggregate data identify where to look; they do not reveal why the gaps occurred.
Business Question
Which Texas disaster-county records show the largest differences between the share of valid owner registrations approved for FEMA IHP assistance and the share approved among all other eligible counties in the same disaster? Which counties appear on the same side of that comparison in more than one event?
The decision owner is program and applicant-support leadership at FEMA, the state, and affected local jurisdictions. Their practical choice is where to spend limited review time—not whether a county or the program passed or failed.
The useful positive result is a short, transparent queue with repeated-event signals. The equally useful negative result would be that the queue dissolves under reasonable robustness checks or that county characteristics account for it mechanically. Neither occurred. The queue remains informative, but the analysis still cannot supply a causal explanation.
Data & Scope
The main source is FEMA's Housing Assistance Program Data — Owners. It contains aggregated, non-personally identifiable records for house owners. The denominator is valid owner registrations. The numerator is the number of those registrations approved for FEMA IHP assistance. This is not a renter dataset, a household count, a count of everyone affected, or a housing-repair-only approval measure.
The analysis uses a governed archive of the owners dataset captured on July 20, 2026, joined to FEMA disaster-declaration context. Seven Texas disasters pass the coverage rules:
| Disaster | Incident | Incident began | Eligible counties | Valid owner registrations | Approval rate |
|---|---|---|---|---|---|
| DR-1606 | Hurricane | September 23, 2005 | 22 | 270,037 | 58.2% |
| DR-1791 | Hurricane | September 7, 2008 | 22 | 424,220 | 15.6% |
| DR-4223 | Severe storm | May 4, 2015 | 8 | 17,122 | 39.1% |
| DR-4332 | Hurricane | August 23, 2017 | 34 | 441,317 | 47.2% |
| DR-4586 | Severe ice storm | February 11, 2021 | 52 | 184,616 | 15.4% |
| DR-4781 | Flood | April 26, 2024 | 23 | 123,304 | 38.5% |
| DR-4798 | Hurricane | July 5, 2024 | 20 | 576,011 | 70.9% |
Current county context comes from the 2020–2024 American Community Survey. Population density uses ACS population and land area from the Census Bureau's 2024 county Gazetteer. NFIP context uses FEMA's redacted policy records and is limited to the two 2024 flood-related events. All sources are aggregated; no individual registration or policy is linked.
Method
A county enters the analysis with at least 500 valid owner registrations. A disaster enters only if at least eight counties clear that floor. For every eligible county, we calculate its approval rate and compare it with the pooled rate among all other eligible counties in the same disaster. This keeps the incident and period fixed inside each comparison.
We rank the 181 resulting records by the absolute percentage-point difference and take exactly twenty. The queue is tested three ways: a 250-registration floor, a 1,000-registration floor, and removal of the county's largest ZIP followed by the original 500-registration floor. A record that falls below a floor is labeled ineligible in that view; it is not treated as evidence that the original effect reversed.
For repeated counties, an event must differ by at least five percentage points in the same direction. We also recalculate the comparison after removing the largest other eligible county, guarding against a single large county defining the remainder.
The association layer is deliberately secondary. It uses rank correlations inside each disaster, with 2,000 county-cluster bootstrap samples for descriptive intervals. It reports no p-values and cannot change the queue.
Findings
The top twenty are concentrated in three events: six records from DR-1606, ten from DR-4781, and four from DR-4798. That concentration is itself a reason to keep the result event-specific rather than generalize it to every Texas disaster.
Scroll chart horizontally to read all labels.
Full chart description
Twenty largest same-disaster FEMA IHP owner-registration approval-rate gaps by Texas county and disaster.
| Rank | Disaster | County | Valid owner registrations | County rate | Remainder rate | Gap | Largest ZIP share |
|---|---|---|---|---|---|---|---|
| 1 | DR-1606 | Brazoria | 5,074 | 11.7% | 59.1% | -47.4 pp | 20.2% |
| 2 | DR-1606 | Harris | 38,853 | 18.5% | 64.9% | -46.4 pp | 7.6% |
| 3 | DR-1606 | Fort Bend | 2,099 | 13.5% | 58.6% | -45.1 pp | 29.1% |
| 4 | DR-4798 | Waller | 1,450 | 30.8% | 71.0% | -40.3 pp | 35.7% |
| 5 | DR-4781 | Newton | 672 | 76.9% | 38.3% | +38.7 pp | 39.3% |
| 6 | DR-1606 | Galveston | 20,100 | 22.7% | 61.1% | -38.4 pp | 13.4% |
| 7 | DR-4781 | Tyler | 560 | 76.2% | 38.3% | +38.0 pp | 55.2% |
| 8 | DR-4781 | San Jacinto | 1,527 | 75.4% | 38.0% | +37.4 pp | 37.7% |
| 9 | DR-4781 | Henderson | 1,383 | 74.7% | 38.1% | +36.6 pp | 17.0% |
| 10 | DR-4798 | Trinity | 1,036 | 35.7% | 71.0% | -35.3 pp | 82.2% |
| 11 | DR-4781 | Jasper | 967 | 72.9% | 38.2% | +34.7 pp | 61.6% |
| 12 | DR-4781 | Anderson | 559 | 72.6% | 38.3% | +34.3 pp | 40.8% |
| 13 | DR-1606 | Jefferson | 68,357 | 83.4% | 49.7% | +33.7 pp | 12.9% |
| 14 | DR-4781 | Walker | 1,919 | 71.4% | 37.9% | +33.4 pp | 54.7% |
| 15 | DR-4781 | Polk | 1,982 | 70.3% | 37.9% | +32.4 pp | 67.7% |
| 16 | DR-4798 | Austin | 583 | 39.5% | 71.0% | -31.5 pp | 44.9% |
| 17 | DR-1606 | Orange | 28,059 | 85.7% | 55.0% | +30.7 pp | 32.4% |
| 18 | DR-4781 | Trinity | 749 | 68.0% | 38.3% | +29.7 pp | 91.3% |
| 19 | DR-4798 | Nacogdoches | 1,146 | 42.7% | 71.0% | -28.3 pp | 35.6% |
| 20 | DR-4781 | Hockley | 1,275 | 66.3% | 38.2% | +28.1 pp | 98.4% |
All twenty retain direction and at least a five-point difference at the 250-registration floor, and seventeen also keep a place in that variant's own top twenty. Fourteen remain eligible at the 1,000-registration floor; all fourteen retain the effect and their top-twenty place. Twelve remain eligible after their largest ZIP is removed; all twelve retain the effect and their top-twenty place. Eight fall below 500 registrations in that last view; Trinity and Hockley in DR-4781 deserve particular caution because one ZIP supplies more than 90% of their valid owner registrations.
The repeated-event view provides the stronger operational signal:
| County | Direction | Events | Disaster numbers | Largest-other-county guard |
|---|---|---|---|---|
| Liberty | Above | 4 | DR-1791, DR-4332, DR-4586, DR-4781 | 4 of 4 retained |
| Fort Bend | Below | 4 | DR-1606, DR-1791, DR-4332, DR-4586 | 3 of 4 retained |
| Newton | Above | 3 | DR-1606, DR-4332, DR-4781 | 3 of 3 retained |
| Collin | Below | 2 | DR-4586, DR-4781 | 2 of 2 retained |
| Hardin | Above | 2 | DR-1606, DR-4781 | 2 of 2 retained |
| Henderson | Above | 2 | DR-4586, DR-4781 | 2 of 2 retained |
| Jefferson | Above | 2 | DR-1606, DR-4332 | 2 of 2 retained |
| Matagorda | Above | 2 | DR-4332, DR-4586 | 2 of 2 retained |
| Nacogdoches | Below | 2 | DR-1606, DR-4798 | 2 of 2 retained |
| Shelby | Below | 2 | DR-1606, DR-4798 | 2 of 2 retained |
| Wharton | Above | 2 | DR-4332, DR-4586 | 2 of 2 retained |
| Dallas | Below | 2 | DR-4223, DR-4781 | 1 of 2 retained |
| Smith | Above | 2 | DR-4586, DR-4781 | 1 of 2 retained |
Association Analysis
The associations sharpen the next question but do not answer it. Within the same disaster, counties with higher current density, income, and home value tend to sit lower relative to their event remainder. The relationship with the size of the gap is smaller.
| Current county context | Signed-gap correlation | Absolute-gap correlation |
|---|---|---|
| Population density | -0.31 (-0.45 to -0.15) | -0.13 (-0.27 to 0.02) |
| Median household income | -0.27 (-0.40 to -0.12) | -0.18 (-0.31 to -0.06) |
| Median home value | -0.31 (-0.45 to -0.16) | -0.17 (-0.29 to -0.05) |
The queue's current medians point the same general way: 61.9 people per square mile versus 109.4 outside the queue; median household income of $62,700 versus $68,703; and median home value of $185,050 versus $209,300. This does not mean lower density or income caused an approval difference. It does mean those characteristics belong in any case-mix and operations review.
NFIP participation does not supply the missing explanation. Across 43 county-event records in the two 2024 flood-related disasters, the within-disaster correlations are 0.01 for direction and -0.02 for absolute size, with broad intervals around both. An apparent pooled relationship disappears once each disaster is considered on its own.
The unanswered question is now concrete: do the repeated county directions reflect differences in damage and case mix among owner registrants, outreach and document completion, inspection operations, or another event-specific process? The available aggregate data cannot distinguish among them.
Business Implications
Program teams can use the top twenty as a first review queue and the repeated-event table as a second priority layer. A useful review would bring together county staff and FEMA operations to examine, for each named event:
- registration and document-completion guidance;
- referral and inspection timing;
- the mix of housing and other-needs cases;
- damage and occupancy context; and
- whether one ZIP or locality dominates the aggregate record.
The result also argues for routine same-disaster monitoring. A statewide rate can hide large county separation, while a significance-based alert can be misleading because county heterogeneity is ordinary in these data. A fixed, effect-ranked queue is easier to explain and audit.
Nothing here should trigger an adverse action, change eligibility, label a county, or grade program performance automatically. Human review must add the operational evidence that the public aggregate data lack.
Limitations
The FEMA source covers owners, not renters or everyone who sought assistance. It contains aggregate counts, not registration files, denial reasons, eligibility decisions, inspection histories, communications, or applicant case mix. County is a recorded administrative label that may be overridden by the registrant. The data therefore cannot establish causation, error, fairness, discrimination, fraud, or program quality.
The seven qualifying disasters span 2005–2024. Same-disaster comparisons avoid pooling approval levels across program eras, but repeated county names remain separate event observations. The ACS context describes 2020–2024 and cannot be read backward as historical county conditions for older events. Population density is continuous; this report does not assign an official rural or urban category.
The top twenty are selected because they are the largest observed gaps, so their magnitudes are descriptive, not forecasts. Eight queue records fall below the 500-registration floor after removal of their largest ZIP. The queue is also concentrated in three disasters. Absence from the queue does not mean a county has no operational issue.
NFIP policy counts are independent county aggregates. They do not identify insurance held by FEMA registrants, private flood insurance, coverage adequacy, or claim status. The NFIP view is limited to 2024 because the governed policy archive begins in 2015 and current ACS housing units are not a defensible historical denominator for older events.
Sources
- FEMA, Housing Assistance Program Data — Owners, version 2.
- FEMA, OpenFEMA dataset metadata API for HousingAssistanceOwners.
- FEMA, OpenFEMA field metadata API for HousingAssistanceOwners.
- FEMA, FIMA NFIP Redacted Policies, version 2.
- U.S. Census Bureau, 2024 ACS five-year data.
- U.S. Census Bureau, 2024 Gazetteer files and Gazetteer record layouts.