Methodology & sources

Vintage and source

Every figure on this site is a real value read from the U.S. Census Bureau's American Community Survey (ACS) 2020–2024 5-year estimates — the table-based summary files on www2.census.gov, which need no API key and are public domain. A 5-year estimate pools five years of survey responses (2020 through 2024) so that even small counties have a publishable figure. Nothing here is scraped, modeled, or bought from a vendor.

The geography is the current 2024 county universe: 3,144 county-equivalents across the 50 states and the District of Columbia. (This reflects Connecticut's 2022 switch from 8 counties to 9 planning regions.) Puerto Rico and the island areas are excluded — their property-tax systems are not comparable, and an estimator would mislead.

The effective tax rate — a ratio of two medians, not a millage rate

The single most important definition on this site. The effective property-tax rate is the county's median real-estate taxes paid divided by its median owner-occupied home value:

effective rate (%)  =  median real-estate taxes (B25103)
                       ------------------------------------  ×  100
                          median home value (B25077)

This is a ratio of two independently-estimated medians — the typical tax bill against the typical home value — not a statutory (millage) rate a taxing authority sets, and not a figure you can read off a tax bill. The two medians come from different survey questions about different (though overlapping) sets of homes, so the ratio is an estimate of the typical burden, useful for comparing counties, not an official rate. This is exactly the figure a thin incumbent site was found publishing roughly 100× wrong; getting it right, per county, is the point of this project.

Margins of error

Because ACS figures are survey estimates, the Census Bureau publishes a margin of error (MOE) at 90% confidence with each one. Where a median is shown with a “± ” value (for example $4,489 ± $46), that is the 90% MOE: the true value is very likely within that band. A handful of counties (a few very high-value ones) have a median the Census Bureau published without a margin of error; there the estimate is shown alone, and no ± is invented.

Ranks and the estimator use the point estimates. Two counties whose bands overlap are not reliably different — treat small rank gaps as ties.

Three tax medians (with vs. without a mortgage)

Table B25103 reports median real-estate taxes three ways: for all owner-occupied units, for units with a mortgage, and for units without a mortgage. Mortgaged homes usually show a higher median — not because of a different tax rate, but because they skew toward higher-value, more recently-purchased homes. All three are shown on every county page so the figure is not mistaken for a rate difference.

Nearby counties — why distance, and why the official adjacency file is rejected

Each county page compares against its six nearest county-equivalents by centroid distance (great-circle, from the 2024 gazetteer), cross-state neighbours allowed and labelled. They are called “nearby,” never “bordering,” because distance is what we compute — we do not claim a shared border we have not verified.

The obvious source — the Census county_adjacency.txt reference file — is rejected as dangerously stale. It is a 2010-vintage file: measured against the 2024 gazetteer, 13 current geographies are entirely absent from it (including all nine Connecticut planning regions and several reorganized Alaska boroughs), and 25 FIPS it still lists no longer exist. Using it would ship a broken or empty “neighbours” table for the entire state of Connecticut. No newer adjacency file is published, so we compute nearness from current centroids instead — always current geography, uniform on every page, and zero third-party dependencies.

Suppressed values — never invented

For very small populations the Census Bureau suppresses a median, publishing a large negative sentinel (−666,666,666) instead of a number. This site never renders a sentinel, never coerces it to 0, and never averages it into a rank. A suppressed field is shown as “Not published by the Census Bureau for this county.” Nine counties (places like Kalawao, HI and Loving County, TX) lack a complete tax-and-value pair and so have no effective rate; none appears in the published launch set, but they still appear — honestly marked — in their state's county table. The build fails if any sentinel reaches a rendered field.

Ranks

Counties are ranked by effective rate both nationally and within their state, with 1 = highest. Ranking is among the 3,135 counties with a complete tax-and-value pair; the nine suppressed counties are left unranked, not ranked last. A county's rank by raw dollars paid can differ sharply from its rank by rate, because home values vary so much across the country — both perspectives are available.

Tax as a share of income

As a rough affordability gauge, each county page also shows median real-estate taxes as a percentage of median household income (B19013). It is a ratio of two medians like the effective rate, and it does not account for exemptions that lower taxes for seniors, veterans, or owner-occupants.

The estimator

The on-page estimator multiplies a home value you enter by the county's effective rate. It is framed everywhere as a median-based estimate, not a quote, an assessment, or a tax bill. It cannot know your assessed value, exemptions, assessment ratio, special districts, or homestead caps — all of which can move a real bill substantially. Use it to compare counties, not to budget a specific purchase.

Official-link policy

Property tax in the U.S. is assessed and collected locally. There is no verified, public dataset of all ~3,144 county assessor URLs, and this site will not fabricate or guess one. Each county page therefore links to its state tax authority — 51 links, each hand-verified — plus guidance to search for the local county assessor or appraisal district. (Three state links, for Alaska, Massachusetts, and New Hampshire, sit behind bot-blocking and could not be automatically status-checked; their identity was confirmed by other means.) If a verified per-county assessor dataset is ever sourced, this policy upgrades.

The five-year trend — non-overlapping vintages only

Each county page compares today's medians against the ACS 2015–2019 5-year estimates. The choice of vintage is forced by official guidance: overlapping 5-year period estimates must not be compared — consecutive vintages share up to four years of survey sample and are not statistically independent. The most recent 5-year period that does not overlap 2020–2024 is 2015–2019, so that is the comparison this site makes, and the only one it will ever show. The 2015–2019 medians are shown with their own margins of error.

A dollar change is only described as a rise or a fall when it passes the standard Census two-estimate significance test on the published 90% margins of error; otherwise the page says there is no statistically significant change. The change in the effective rate is tested the same way, with each vintage's rate MOE derived by the Census ratio formula. Because the two periods share no sample, the test's independence assumption holds.

County geography changed between the two periods, and pretending otherwise would fabricate trends. Connecticut's nine planning regions (which replaced its eight counties in 2022) get no comparison — the 2015–2019 figures describe different areas. Chugach and Copper River (created when Valdez-Cordova was split in 2019) have no 2015–2019 estimates. Kusilvak Census Area and Oglala Lakota County are renames of unchanged areas — verified present under their current FIPS codes in the 2015–2019 data — and are compared normally where the medians are published. A county whose 2015–2019 medians the Census Bureau did not publish, or published only as an open-ended (top- or bottom-coded) figure, gets an explanatory note instead of a number — never an invented one.

Update cadence

The ACS 5-year vintage releases once a year, each December. Between vintages the data is immutable, so there is no scheduled refetch — only a freshness check for a new vintage directory. The five-year trend comparison moves with it: when the 2021–2025 vintage lands, the non-overlapping comparison base becomes 2016–2020.