Datadory notebook
Characteristics of New Housing Data: What the Census Counts in a New Home
Characteristics of new housing data comes from the Census Bureau's annual Characteristics of New Housing program: roughly 90 Excel tables plus Survey of Construction microdata covering square footage, bedrooms, bathrooms, wall material, sales prices and financing in new privately-owned US homes, with summary tables from 1973, microdata from 1999, and free public-domain access throughout.
1,744 datasets. Pick your catch.
Where does characteristics of new housing data come from?
One federal program owns it. US Census Characteristics of New Housing is the Census Bureau's annual profile of what is actually being built, derived from the Survey of Construction — the household- and builder-level sample survey authorized under Title 13, with voluntary responses and partial funding from HUD. Every published number on new-home size and features traces back to that single survey.
What do the roughly 90 Excel tables each year measure?
Each annual cycle refreshes about 90 Excel summary tables of 100–150 KB apiece. The subject list reads like a spec sheet for an American house: age-restricted communities, air conditioning, bathrooms, bedrooms, beds-by-baths combinations, construction method, contract price and contract price per square foot, exterior wall materials by type, financing, fireplaces, stories, floors, foundation, foyer, framing, and heating system with fuel type.
Every characteristic ships in four counting variants, and choosing among them is the first decision any analysis makes. Files named _cust.xls cover all completed units; sold files cut to houses that sold; mfu_* covers multifamily units and mfb_* multifamily buildings. A floor-plan study wants the completed-units file, while an upgrade-pricing study wants sold*, because the populations differ.
Schema is stable and shallow: Year rows, a total-houses-in-thousands column and a percent-distribution column. The comparison table below maps the four file families plus the microdata onto the questions they answer.
How far back can home-size and bedroom trends be traced?
Summary series run deep, and the bedrooms table is the anchor, starting in 1973. In that first year 1,197 thousand new single-family houses were completed: 769 thousand — about 64% — had three bedrooms, 148 thousand (12.4%) had two or fewer, and 280 thousand had four or more. By 1980 the completed total had fallen to 957 thousand, with 603 thousand three-bedroom houses holding essentially the same share.
Two supporting layers extend the record. A Historical Tables collection covers 2003–2017 for anyone rebuilding older comparisons, and a Summary of Changes documents file changes since 2005. Presentation layers — Annual Highlights and an interactive Single-Family Infographic — sit on top for readers rather than pipelines.
Record-level depth is shorter but different in kind: one SOC microdata file per year from 1999 to 2025, carrying roughly 45 documented variables including SQFS square footage, BEDR bedrooms, FULB and HAFB bathroom counts, WAL1 and WAL2 wall materials, and a WEIGHT column for reproducing national estimates. Tabulated history reaches furthest back; microdata answers questions the tables never published.
Does the series go below the national level?
Partly, and knowing the ceiling prevents bad joins. Most characteristics are also published at Census region level — Northeast, Midwest, South, West — but the summary tables publish no geography below region. Metro-level work requires dropping to the microdata, whose DIV and METRO fields identify the Census division and metropolitan area of each sampled unit.
That constraint is the flip side of the Building Permits Survey, the slice's quality-9 record, whose counts descend to state, CBSA, county and place — with Island Areas added in the 2025 annual release. Pairing the two moves from authorized units by jurisdiction to the physical features of what got built, each source covering the half the other lacks; the vs US Census Building Permits Survey comparison works through the trade in full.
Disclosure control shapes the microdata rather than the tables: files contain no address- or builder-identifying information, modified values are flagged, and files from 2017 onward carry CBDRB disclosure-authorization numbers such as CBDRB-FY26-0249 for the 2025 file. One access quirk survives: SAS-format microdata is available only by phone request at (301) 763-5160.
Who uses characteristics of new housing data, and for what?
All eight Datadory personas carry this dataset, and the relevance split shows what it is good for. Three tie at the maximum score of 3: data scientists analyze square footage, bedroom and materials trends from the 1973-forward tables plus microdata; market researchers profile new-construction product mix by region for building-products studies; journalists, academics and students cite an official annual series reaching back to 1973.
Five more ride along at relevance 2 or 1: sales teams targeting builders with feature-mix signals, competitive-intel teams tracking feature adoption, e-commerce operators aligning assortment to home sizes in growth markets, developers mining microdata for custom cuts, and investors gauging home-size trends behind building-products equity theses. The workflows land on the data scientists homebuilding and market researchers homebuilding pages.
| File family | What it counts | Grain | History | Best-fit question |
|---|---|---|---|---|
| All completed units (*_cust.xls) | Every finished new privately-owned residential unit, by characteristic | Annual; national plus most characteristics at Census region level | Bedrooms series from 1973 to latest year; Historical Tables cover 2003–2017 | Floor-plan and feature mix of what actually got built |
| Sold units (sold*) | Only new houses that sold, with sale-linked characteristics such as contract price and financing | Annual; national and regional counts and percent distributions | Parallel annual series alongside the completed-units files | Upgrade and pricing behavior among actual buyers |
| Multifamily units (mfu_*) | Individual units inside multifamily buildings | Annual; national and regional | Published alongside the single-family tables each cycle | Apartment and condo product comparisons |
| Multifamily buildings (mfb_*) | Whole buildings rather than individual units | Annual; national and regional | Published alongside the unit-level files each cycle | Building-scale questions such as stories and construction method |
| SOC microdata (socNN) | Record-level sampled units: STRT/COMP/SALE dates, SLPR and CONPR prices, SQFS, BEDR, FULB/HAFB, WAL1/WAL2, FINC, BASE, GAR, DIV/METRO, WEIGHT | One row per sampled unit; Census division and metro identifiers, no address or builder identification | 1999–2025, one file per year; 2025 released July 1, 2026 | Custom cross-tabs Census never tabulated |
Pick up where this leaves off
Every one of these ships with sample rows before you commit to anything.
US Census Survey of Construction (SOC) Program Hub
US Census Characteristics of New Housing
NAHB Housing Economics - HMI, Starts & Market Data
EIA Residential Energy Consumption Survey (RECS) 2020 - Appliances Tables (HC3.x)
DOEID
American Housing Survey (AHS) National Microdata
FHFA House Price Index (HPI)
Want rows instead of a pitch? Name the datasets.
API, files, or your warehouse. Daily, weekly, or hourly.
Get a sampleQuestions worth asking
What is the difference between the summary tables and the SOC microdata?
Same survey, two products. The summary tables publish about 90 pretabulated distributions per year — Year, total houses in thousands and percent distribution — with series such as bedrooms running from 1973. The microdata are the underlying records, 1999 to 2025, with roughly 45 variables including SQFS, BEDR, WAL1/WAL2 and WEIGHT, for cuts Census did not publish.
Does characteristics of new housing data include existing homes?
No. Coverage is new, privately-owned residential structures captured by the Survey of Construction, so resales are excluded entirely. To profile the whole occupied stock, pair the series with the American Housing Survey National Microdata, biennial in odd years since 1973, whose 2023 household file carries 55,669 records.