Methodology

The Index is computed from a research warehouse of recorded residential transactions in Manhattan, Brooklyn and Queens. The underlying records are public: property transfers recorded with the New York City Department of Finance and its Automated City Register Information System (ACRIS), alongside the Department of City Planning’s Primary Land Use Tax Lot Output (PLUTO) file for parcel geography and building attributes. Those records are the City’s, not ours. What The Roebling Index contributes is the work on top of them: assembling the transactions into buildings, resolving identity across lots and addresses, screening out the transfers that are not market sales, and computing the series under the fixed rules set out below. Figures should be attributed to The Roebling Index; the source records should be attributed to the City.

A building does not need a published profile on this site to be counted: the Index runs on the full warehouse universe, not on the subset of buildings we have written about. The two counts are different on purpose and should never be substituted for one another.

The current release carries 373,468 index-eligible transactions across the full published history, 2004 to the source-data cutoff, over all three boroughs. That is a corpus size, not the sample behind any single published figure. A one-year change rests only on the sales recorded in the two years it compares, which is a far smaller number and is printed next to that change on the Index itself. Of that total, 253,455 fall in Manhattan below 96th Street; the remainder are Harlem, Washington Heights & Inwood, Roosevelt Island, Brooklyn and Queens.

Series begin in 2004. Each published scope is a neighborhood or a parent market with a fixed boundary; a granular neighborhood is an exclusive subset of its parent, and parents are computed directly from all of their own transactions rather than by averaging their sub-markets.

The governing principle is that provenance is not the same as metric eligibility. Whether a transfer was a genuine market transaction is a separate question from whether it belongs in a particular price series, and the Index keeps the two fields apart permanently.

The two metrics use different universes, and this matters when comparing them. Condominium price per square foot is computed on market-price observations, which include sponsor first sales — a new development’s first closings are real market-clearing prices and excluding them would misrepresent the segment. Cooperative price per room is computed on the secondary market only, which excludes sponsor sales. Condo and co-op series are therefore not like-for-like on new-development exposure, and a condo-versus-co-op comparison should be read with that asymmetry in mind.

The Index and the per-building sales pages use the same eligibility logic and are subject to automated consistency checks.

Removed from both metrics:

• Nominal transfers and intra-family conveyances.
• Commercial, retail, parking and storage units, and whole-building bulk transfers — a residential guard applied because a warehouse property class of “condo” does not by itself exclude them.
• Internal bulk transfers. Same-day clusters are classified, not blanket-excluded: a new-development sell-out (distinct units at distinct, market-scaled prices) is kept, while an internal transfer (uniform or opaque units and prices, or an unverifiable lone row) is dropped. Blanket exclusion would silently discard the first-closing day of every new development.
• Rows with unusable numeric fields.
• Possible late-sponsor sales without affirmative sponsor evidence. A recent build year alone is not evidence.
• For condominium $/sf only: trades whose square footage cannot be affirmatively validated against a repeat sale or a consistent apartment line, within ±15%.
• Duplicate records, where deduplication requires unit agreement. Two same-day, same-price sales in different units are kept, because they are two sales.

For each scope, year and metric, the Index takes the median of eligible observations. Medians are used rather than means because the top of this market is thin and heavy enough to move an average on a single trade.

A year with fewer than 10 eligible sales is suppressed, not smoothed and not interpolated. A gap in a series means the sample was too thin to publish, not that nothing traded.

Every series is published in both nominal dollars and real dollars adjusted to 2025 using the Consumer Price Index. Nominal and real can point in opposite directions over a decade, and the real figure is the one that answers whether an owner gained purchasing power.

Every long-run change ends at a complete year: full-year 2025 for nearly every series, or the series’ latest complete year where 2025 lacks a qualifying sample (see below). The warehouse carries 2026 transactions and the artifact carries a 2026 point, but a partial year is roughly half a sample and moves percentages in both directions, so it is never the endpoint of a long-run change. Separately labeled year-to-date figures, such as the homepage’s current-year change, compare the available months of 2026 with the same months of the year before. Each figure states its endpoint and comparison period.

Changes are quoted against three fixed windows: one year, 2022 as the rate-shock reference, and 2016 as the decade anchor. 2016 is the standing anchor for every series in the Index, fixed in the generator and applied uniformly — not selected per claim. Where a scope has no observation in an anchor year, the nearest adjacent year is used, searched one year forward then one year back.

A series whose most recent qualifying year is more than two years before the base year is dropped rather than quoted. A neighborhood whose data stops in 2019 does not get to publish a “current” median.

A scope that clears the sample rules publishes plainly. A scope that clears them but carries a structural qualification publishes with a caveat, and the caveat travels with the figure. The two in current use are: smaller sample than the other published series, and mostly a small-multifamily and row-house market; thin apartment sample. The second matters most in Brooklyn, where a neighborhood can carry an enormous recorded sale history that is largely houses rather than apartments.

A scope that does not clear the rules is suppressed entirely. It has no figure rather than a soft one.

Rent figures follow their own rules, set out here. They are computed from closed leases recorded in New York brokerage listing systems and accessed through Compass, not from public records: New York does not record leases publicly. Coverage runs from 2004 through September 2026; published figures start in 2015.

Eligible leases. A lease counts only when the record supports it. Held out: rents outside a plausible range, a closing rent far from the final asking rent, a lease followed by the same apartment back on the market within 90 days, a listing up longer than 150 days, free-rent concessions where the net rent cannot be determined, and apartments with no bedroom count.

Change by year. A repeat-rent index: each apartment is compared with its own previous lease, 10 to 60 months earlier, in a weighted repeat-sales regression, so the change is like for like rather than a shift in which apartments happened to rent. The current year is compared with the same months of the year before, never with a full year. A market publishes as solid with 500 or more lease-to-lease comparisons, good with 100 to 499, and directional with 30 to 99; below 30 it is not published.

Levels. Condo and rental buildings are measured by median rent per square foot per year, from leases whose square footage falls inside a plausible range for the bedroom count. About half of leases report square footage, and those leases rent somewhat higher than the rest, so the figure describes the apartments that report it. Co-ops are measured as the sales figures are, by median monthly rent per room, from sublet leases; for the same apartments, the rental and sales room counts agree within one room 95% of the time. A cell resting on fewer than 30 leases is marked, one under 10 is not shown, and a level’s change from the prior year is shown only at 100 leases or more.

Geography. On a neighborhood page, rents come from the same buildings as the sales figures: a building the sales series has placed counts where the sales series counts it, and a building that has only ever rented is placed by its listing neighborhood. Corridor pages are built from a list of buildings rather than a boundary, so their rents are for the named rental market, and the page says where its buildings reach beyond it.

Building pages. A building shows its own closed rents when it has at least three closed leases in the last 36 complete months: the lead figure (per room for a co-op, per square foot otherwise) and median monthly rent for each apartment size with at least two leases. Figures are aggregates; no unit or unit rent is published.

Limitations. The data covers apartments rented through brokerage listings, which leans toward market-rate units. More than half of eligible leases sit in buildings registered as containing at least one rent-stabilized apartment; the registration does not say which apartments, and in a test across twelve large markets, removing those buildings moved the year-over-year figures by a median of 0.3 points or less. The latest months fill in as late-reported leases arrive, so the current year can move.

These are open and disclosed rather than resolved. They bound what the Index can currently support.

• Supertall square footage. Floor-numbered, single-sale units in supertall towers cannot be validated against a repeat sale or an apartment line, so some high-floor resales are held out of the $/sf series. The very top of the market is understated as a result.
• Co-op transfer coverage. Cooperative transfers are recorded differently from deeded sales, and coverage varies by building. Affected building pages disclose this rather than blending sources to cover the gap.
• Medians carry mix. A neighborhood median reflects what happened to trade that year. It is not a like-for-like measure of the same apartment over time, and it is not an appraisal.
• Boundaries are a choice. Neighborhood lines are fixed and applied by parcel, but reasonable people draw some of them differently. A building near an edge may be filed where a broker would not file it.

Cite as: The Roebling Index, The Roebling Team at Compass, with the release month and the window you are quoting — for example, “up 12% since 2016, per The Roebling Index (September 2026 release), measured to full-year 2025.” Naming the endpoint and the anchor matters: the same series produces different numbers against different anchors, which is why both are fixed and published here.

The summary table is downloadable as CSV, and the full Index is at /roebling-index. Journalists can find bios, a headshot and current research angles on the press page.

The Roebling Index is published by Corey Cohen, a Licensed Associate Real Estate Broker and Principal of The Roebling Team at Compass. The research is used in the brokerage practice, and the brokerage practice funds the research. Sales figures are computed from public records and rent figures from closed-lease records, by fixed rules that do not vary by listing, client or engagement, and the rules are published above so that the output can be checked independently of who produced it.

Index figures are indicative market research. They are not an appraisal, not investment advice, and not a valuation of any specific apartment.