Published 30 August 2026 · Free to reproduce with attribution
Across 3,093 U.S. counties, which state a home sits in explains more than twice as much of what its owner reports paying to insure it as the modelled natural-hazard risk the home faces.
An analysis of 50,715,340 mortgaged households — 99.99% of every mortgaged owner-occupied home in the country — combining what homeowners report paying with FEMA’s own dollar-denominated loss model.
Interactive · all 3,093 counties
All 3,093 counties in the study. Type a county or state to see what its mortgaged homeowners report paying, and what counties facing the same modelled hazard pay. Comparison ranges use counties with at least 5,000 mortgaged households, the same basis as the 2.3× figure above, so a handful of very small counties cannot distort the spread.
Median annual premium
FEMA’s model covers natural hazards only. It says nothing about fire, theft, water damage or liability, which are what a homeowners policy mostly pays out on — so this is not a measure of how well insured you are overall.
The attribution link carries rel="nofollow" deliberately. Google treats links distributed inside a widget, where the host did not choose the anchor text, as a link scheme and asks that they be nofollowed. Embedding this passes no ranking signal to us and creates no risk for you. To link to the study editorially instead, use the citation.
Homeowners insurance is sold as a price on risk. Rate filings are justified to regulators on expected losses; consumers are told premiums rise because hazards rise. That is a testable claim, and until recently it was awkward to test, because the public numbers on all sides were modelled rather than measured.
Two federal datasets now make it straightforward. The Census Bureau’s American Community Survey asks mortgaged homeowners what they actually pay — not a quoted rate for a hypothetical house, but a reported bill, weighted to the population. FEMA’s National Risk Index publishes, for every county, the expected annual dollar loss to buildings from each of eighteen natural hazards.
Put them side by side and the question becomes arithmetic. We did that for every county in the United States. The short answer: risk matters, but geography matters more than twice as much.
Each dot is one county: FEMA’s modelled expected annual building loss from the perils a standard policy covers, against the median premium its mortgaged homeowners report paying. The relationship is real and positive — but loose. Hazard risk accounts for 20% of the variation between counties.
Now replace risk with nothing at all except a set of state indicator variables — no hazard data, no home values, nothing about the house. That model explains 44%. Weighting by households, the gap widens: 65% for state against 30% for risk.
Knowing which state a house is in tells you more about its insurance bill than every hazard model ever built for its county.
Sorting counties into twenty equal bands of hazard risk and comparing the cheapest and dearest county inside each band isolates price differences risk cannot explain. The median gap is 2.3×. Every pair below faces materially the same modelled exposure.
Within a single state’s regulatory regime, risk ought to sort prices cleanly. Mostly it does not. The median within-state correlation is +0.218, and in 11 of 43 states it runs backwards: the counties facing more modelled hazard pay less than the counties facing less.
There is a plainer reason risk and price come apart, and it is not really about pricing at all. Of all the building loss FEMA expects American homes to suffer each year, 76.5% comes from perils a standard HO-3 policy does not cover — flooding above all, then earthquake.
These are not edge cases or fine print. They are the two hazards most capable of destroying a house outright, and the two a homeowner is least likely to know they are carrying alone.
Flood at least has a separate market. Whether anyone uses it is a different question, and one we answered separately: across 2,304 counties, the median covers just 15% of the homes inside its mapped flood zones. See Risk Without Cover.
That exposure sits disproportionately in old housing. In the 158 counties FEMA scores at or above 95 for earthquake risk, 22,030,850 homes — 57% of the housing stock — were built before 1980, and so predate the substantially strengthened seismic provisions of the 1976 Uniform Building Code. Standard homeowners policies exclude earthquake in every one of them.
Building age is a rough proxy here rather than a verdict on any individual house. US codes have carried seismic requirements since 1927, entered their modern form when the 1961 code adopted the Structural Engineers Association of California recommendations, and were strengthened again in 1976 and repeatedly since. Adoption also varied by jurisdiction. What the figure establishes is the scale of the exposure, not that 22 million specific homes are unsafe.
It predates the current crisis. The ACS 5-year file spans 2019–2023 and is centred near 2021. It does not capture the premium surges of 2023–2025 in Florida, California and Louisiana. Read it as the baseline those increases departed from, not as today’s market.
It covers mortgaged owners only. Renters are absent, as are the 39% of American owners who own outright and face no lender requirement to insure at all.
Premiums are self-reported, and county estimates carry real uncertainty. Households may report escrowed amounts or misremember. The ACS margin of error on this table has a median of about 9% of the estimate at county level, and exceeds 20% in the smallest tenth of counties — which is why the findings here rest on patterns across thousands of counties and on household-weighted figures, never on any single county’s number.
Risk is a model. The National Risk Index is FEMA’s estimate, not observed insurer experience, and county resolution is coarse relative to how wildfire and flood risk actually vary within a county.
Natural hazard is only part of what a policy covers. Homeowners insurance also pays for house fires, burst pipes, theft and liability, which no public county-level dataset measures. The gap between price and modelled hazard is therefore not, by itself, evidence that any premium is too high or too low.
Correlation, not conduct. Nothing here identifies why any state’s prices sit where they do. Rate regulation, residual markets, reinsurance costs, litigation environments and market concentration all plausibly contribute, and this analysis distinguishes none of them.
Household-weighted median annual premium and covered-peril loss rate for all 51 states and DC, ordered from the most expensive homeowners insurance rates by state to the least. Figures are ACS 5-year 2023 estimates, centred near 2021, so they are the baseline the 2023–2025 increases departed from rather than today’s market.
| State | Counties | Median premium | Covered loss / $1k |
|---|---|---|---|
| FL | 67 | $2,012 | $1.17 |
| LA | 64 | $1,877 | $1.86 |
| OK | 76 | $1,807 | $0.62 |
| TX | 237 | $1,767 | $0.49 |
| CO | 63 | $1,717 | $0.59 |
| KS | 105 | $1,652 | $0.40 |
| NE | 80 | $1,604 | $0.64 |
| MS | 81 | $1,496 | $0.81 |
| RI | 5 | $1,495 | $0.12 |
| MN | 87 | $1,493 | $0.28 |
| CT | 9 | $1,448 | $0.26 |
| MA | 14 | $1,443 | $0.17 |
| MO | 115 | $1,419 | $0.28 |
| AL | 67 | $1,385 | $0.62 |
| GA | 159 | $1,358 | $0.33 |
| SD | 63 | $1,354 | $0.44 |
| ND | 51 | $1,351 | $0.56 |
| WY | 23 | $1,348 | $0.27 |
| MT | 52 | $1,346 | $0.29 |
| AR | 75 | $1,314 | $0.46 |
| NY | 62 | $1,280 | $0.09 |
| SC | 46 | $1,279 | $1.15 |
| TN | 95 | $1,255 | $0.18 |
| NJ | 21 | $1,246 | $0.23 |
| CA | 58 | $1,239 | $0.24 |
| KY | 120 | $1,235 | $0.26 |
| HI | 4 | $1,222 | $0.29 |
| MD | 24 | $1,204 | $0.14 |
| AK | 27 | $1,194 | $0.14 |
| IA | 99 | $1,189 | $0.37 |
| IL | 102 | $1,187 | $0.17 |
| NC | 100 | $1,179 | $0.60 |
| IN | 92 | $1,155 | $0.20 |
| VA | 133 | $1,138 | $0.17 |
| WA | 39 | $1,113 | $0.09 |
| NM | 32 | $1,099 | $0.21 |
| NH | 10 | $1,048 | $0.19 |
| MI | 83 | $1,044 | $0.19 |
| OH | 88 | $1,041 | $0.16 |
| DC | 1 | $1,041 | $0.10 |
| PA | 67 | $996 | $0.10 |
| VT | 14 | $969 | $0.08 |
| AZ | 15 | $954 | $0.17 |
| WI | 72 | $950 | $0.19 |
| WV | 55 | $949 | $0.09 |
| DE | 3 | $937 | $0.19 |
| ME | 16 | $914 | $0.34 |
| OR | 36 | $912 | $0.13 |
| ID | 43 | $903 | $0.30 |
| NV | 15 | $896 | $0.23 |
| UT | 28 | $855 | $0.33 |
Florida, at a household-weighted median of $2,012 a year, followed by Louisiana ($1,877), Oklahoma ($1,807), Texas ($1,767) and Colorado ($1,717). These are ACS 5-year 2023 figures centred near 2021 and do not reflect the 2023–2025 increases.
Utah, at $855 a year, then Nevada ($896), Idaho ($903), Oregon ($912) and Maine ($914). The spread between the most and least expensive state is about 2.4×.
The household-weighted median across 3,093 counties is $1,317 a year for mortgaged owner-occupied homes. That is what households report paying in the American Community Survey, not a quoted rate for a model home.
Only weakly. Across 3,093 counties, FEMA’s modelled hazard risk explains 20% of the variation in what people pay, while state of residence alone explains 44% — more than twice as much. Counties facing the same modelled hazard differ in price by a median of 2.3×.
Flood is excluded from standard HO-3 policies and requires separate NFIP or private cover. It is not a small exclusion: inland flooding alone accounts for 59.2% of FEMA’s modelled annual building loss in the United States, and earthquake — also excluded — a further 16.1%. In total 76.5% of modelled building loss falls outside the policy most homeowners buy.
In the 158 counties FEMA scores at or above 95 for earthquake risk, 22,030,850 homes — 57% of the housing stock there — were built before 1980, predating the substantially strengthened seismic provisions of the 1976 Uniform Building Code. US codes had carried seismic requirements since 1927 and modernised them in 1961, and adoption varied by jurisdiction, so age is a proxy for seismic design rather than proof any particular house is unsafe. Standard policies exclude earthquake in all of them.
The findings and underlying figures are free to reproduce with attribution. Both source datasets are public, and the method above is sufficient to rebuild every number on this page.
Every figure on this page, and the state table behind the charts: 51 rows with the county count, median premium and modelled risk index for each.
risk-without-price-by-state.csv · the full figures as JSON
There is no county-level file for this study. The analysis runs on 3,093 county observations, but they are held as anonymous risk-and-premium pairs rather than a named table, so a county breakdown would have to be reconstructed rather than published. The state file and the JSON together contain every number quoted above.
Every exhibit on this page as a high-contrast image, free to reuse with credit. The PNG drops straight into a document or a slide; the SVG stays sharp at any size.
Exhibit 1 (PNG) · SVG
Exhibit 2 (PNG) · SVG
Exhibit 3 (PNG) · SVG
Exhibit 4 (PNG) · SVG
Exhibit 5 (PNG) · SVG
Across 3,093 US counties, modelled hazard explains part of what homeowners pay for insurance, but two counties facing the same modelled risk can differ by more than double, according to an analysis of Census and FEMA National Risk Index data by BeforeRegret.
Risk Without Price, Before Regret, 30 August 2026. Data source: Census ACS · FEMA National Risk Index. https://www.beforeregret.com/research/risk-without-price/
Free to use on any site. It sizes itself to its content and carries its own credit line.
