Property Tax Disparity — Assessed Value vs Sale Price

Measures whether lower-value homes are assessed at a higher share of what they actually sell for than higher-value homes — the property tax disparity — using the IAAO sales-ratio method over official county and state open data.

The "market value" side of every ratio is a recorded sale price from the jurisdiction's own records. No automated valuation model, no Zestimate, no commercial valuation service. That matters twice: nothing here depends on another company's terms of service, and a recorded sale price is evidence an assessment appeal or a newsroom can actually stand behind. An AVM estimate is not.

What you get

Per county, an IAAO ratio study:

Measure	What it means	IAAO standard
Median ratio	Level of assessment	0.90 – 1.10
COD	Horizontal equity — are similar homes treated alike?	5–15 homogeneous, 5–20 heterogeneous
PRD	Vertical equity — > 1.03 is regressive	0.98 – 1.03
PRB	Vertical equity by regression, robust to outliers	−0.05 – +0.05

Plus sale-price quintiles with the median ratio in each (the disparity, shown directly), pass/fail on every threshold, and a plain-language verdict.

Set output to parcels to get every individual sale with its own ratio.

Coverage — stated honestly

There are about 3,144 county-equivalents in the US, and property tax is levied locally in all 50 states. This Actor covers 252 jurisdictions across 4 states:

Source	Level	Jurisdictions	Notes
maryland	parcel	24	All 23 counties + Baltimore City, SDAT
connecticut	parcel	169	All towns, 1.14M sales 2001–2023
new-york	jurisdiction	58	State's own municipality equity survey
cook-county-il	parcel	1	Chicago; assessment + sales joined on PIN

Why not all 3,144? Because the data does not exist in machine-readable form for most of them. A sales-ratio study needs two things published together: assessed values and actual sale prices. A sweep of the Socrata open-data network across 12 query terms found only 15 domains publishing both, and most were a single city. The majority of US counties publish assessments behind a parcel-lookup web form with no bulk export and no sale prices at all, and many states treat sale prices as non-public entirely (non-disclosure states such as TX, UT and MO).

Coverage grows by adding verified entries to src/registry.py. Every entry there was queried live before being listed — a promising dataset name in a catalog is not evidence it carries a sale price.

Why a registry and not a loop

Three things differ per jurisdiction, and getting any of them wrong produces a confidently wrong number:

1. Level of assessment. Cook County assesses class-2 residential at 10% of market value, Maryland at 100%, Connecticut at 70%. Comparing a raw assessed value to a sale price without dividing by the LOA is off by an order of magnitude.
2. The arm's-length screen. Ratio studies are only valid over open-market sales. Cook ships boolean filter flags; Maryland encodes it as English prose in how_conveyed (902k of its transfers are explicitly non-arm's-length); Connecticut ships neither. Leave quitclaims and $1 family transfers in and you manufacture fake regressivity at the low end.
3. Shape. Cook needs a PIN join across two datasets; MD and CT carry assessment and sale price on the same row; NY publishes no parcel sale price at all, so it returns the state's own municipality-level equity survey instead.

Methodological caveats

These are surfaced rather than buried, because a buyer who misreads them publishes something wrong:

• Connecticut median ratios run low (~0.6–0.8). CT towns revalue on a 5-year cycle, so an assessment can be 4 years stale against a current sale. In a rising market that depresses the median for reasons that are not inequity. COD, PRD and PRB are computed within a town and remain valid; read the median as a revaluation-lag signal.
• Cook County's county-wide COD runs high (~29). Cook is one county spanning 38 townships with very different markets, and mixing them inflates dispersion. Township is carried on every parcel record so you can stratify.
• Minimum sample. Counties with fewer than 30 arm's-length sales in the year are skipped, not reported thin. IAAO sets 30 as the floor.
• PRD and PRB can disagree. PRD is a ratio of two means that a few extreme sales can swing; PRB fits the trend across the distribution. When they disagree the verdict says so and defers to PRB rather than picking the louder number.

Network routing

Portals disagree about which IPs they trust, and not consistently with each other: Cook County serves a datacenter IP fine but 403s the residential proxy, while Maryland does the exact opposite. Pinning either route breaks whichever source disagrees. Every request therefore tries direct first and falls back to the proxy on 403/429. Supply a free socrataAppToken to lift the anonymous rate limit on large runs.

Sources

All public, all official:

• Cook County Assessor — datacatalog.cookcountyil.gov (uzyt-m557, wvhk-k5uv)
• Maryland SDAT / MD Planning — opendata.maryland.gov (ed4q-f8tm)
• Connecticut OPM — data.ct.gov (5mzw-sjtu)
• NY Dept. of Taxation and Finance — data.ny.gov (4sut-q3dt)

Schemas verified live 2026-08-08.

Who buys this

Assessment appeal firms and property tax consultants, civil-rights and housing policy researchers, investigative newsrooms, county boards reviewing their own assessor, and law firms building equal-protection claims on assessment uniformity.