How we count

A figure for what a house would sell for, built from recorded sales of similar houses nearby and, in some towns, the county’s own record of the house. It is an estimate, not an appraisal.

How it is made. The way an appraiser works: we take the most similar recent sales nearby, adjust each one for every difference by the amount this town’s own sales show the market pays (size, lot, age, type, and the date of sale), and give the most weight to the sales that needed the least adjusting. In each town and kind of house we combine that with our other measured methods in whichever mix tested best.

How we measure it. We price houses that have already sold, using only sales from before each one, then compare our figure with the real price. The method was chosen on sales before 2025 and scored on 2025 and 2026 sales it never saw. Half of the estimates land closer than the typical error below and half land further off. The range shown with each estimate is the one that held about 8 sales in 10 in testing; the table gives the share it actually held.

Why the sales listed under an estimate don’t average to it. Each sale is adjusted before it counts: to today’s prices and to this house’s size. Closer and more similar sales count for more. Part of the figure also comes from the house’s own record: its lot, its building size and, where the county assesses it, that assessment. The listing shows the sales that counted most; many more counted a little.

The seller’s description. Public records describe a typical house and never say what shape it was in. So a house sold as needing work (a flip, a listing with an after-repair value, or one the seller says needs renovation) shows no estimate: we could not measure one. A house described as renovated or new gets the same estimate as any other: we have no way to measure what that adds.

When prices are moving fast. Estimates built from past sales trail a market that turns quickly; in our tests the misses were widest in months when prices were falling.

Where there is no estimate. Wherever our typical error is above 30% (land, stores and rental buildings, and one-family houses in Williamsburg today), on rentals and apartments, and wherever there are too few recorded sales. We show nothing rather than a guess.

Every home sale recorded in public deed records: New York City Department of Finance for Brooklyn, New York State sales records for Rockland, Orange and Sullivan counties, and New Jersey property records for Ocean County. No buyer or seller names are shown.

Public records are published slowly. New York City runs about two months behind, New York State one to two, and New Jersey whenever the state posts its next file. The newest recorded sale we hold, by town:

• Boro ParkAugust 2026
• WilliamsburgAugust 2026
• FlatbushAugust 2026
• MonseyAugust 2026
• LakewoodJune 2026
• Kiryas JoelAugust 2026

Sales in a town, this year and last. We take the three whole months before the newest records and set them beside the same months a year earlier. When a new building is selling out on one street, it is left out of both years so the two compare like with like. Transfers under $50,000 are left out too: they are almost always family or nominal transfers, not sales. With fewer than 30 sales in either year we show this year only.

Asking-price comparisons come from live listings on Homes Row. They are what sellers are asking, not what anyone paid. A range is shown only when there are at least three listings to range over.

Walking times are worked out from the straight-line distance, with 30% added for the turns a real walk takes, at an easy pace of about 83 metres a minute. No route is measured, so a real walk can be a little shorter or longer. The shortest time we show is one minute.

Shuls, schools and shops come from OpenStreetMap and community directories. “Best known” means listed in the most directories. A place listed in error, or one you want removed: removals@homesrow.com.

Counted only from what happens on Homes Row: saved searches, saved homes and messages about homes. No outside data is used, and demand is shown as counts, never as names.