Written and maintained by CASRAI Editorial Board
Last updated
A housing survey is a structured data collection about homes and the households in them. In the United States the term most often means the American Housing Survey (AHS), sponsored by the Department of Housing and Urban Development (HUD) and conducted by the U.S. Census Bureau. This page explains how the AHS is designed, what it measures, how its samples changed across redesigns, and how to analyse its microdata correctly. It does not cover home inspections or property (land) surveys, which share the name but not the method.
Definition: what is the American Housing Survey?
The AHS is a longitudinal survey of housing units. HUD sponsors it and the Census Bureau conducts it. It began in 1973 and has been fielded every two years, in odd-numbered years, since the early 1980s. It covers the United States as a whole and selected large metropolitan areas, and it asks about the physical condition of homes and neighborhoods, the costs of financing and maintaining a home, and the characteristics of the people living there (Census Bureau, AHS About page, read 2026-11-08).
Who uses it and why
The Census Bureau lists HUD’s biennial Worst Case Housing Needs report to Congress, housing program design, funding allocation, other federal agencies and academic researchers among the users of the data. For research administrators and methodologists, the AHS is a standard example of a federal secondary data source: you do not field it yourself, you analyse released files. See secondary data analysis for the general workflow.
Sample design and redesigns
The AHS follows the same housing units over time, so the sample is periodically redrawn. The Census Bureau’s history page gives these figures (read 2026-11-08):
| Year | What changed | Sample size reported by Census |
|---|---|---|
| 1973 | First survey, called the Annual Housing Survey | 60,000 housing units |
| 1985 | Redesign based on 1980 census data | Base sample of about 47,000 housing units |
| 2015 | New national sample drawn from the Census Master Address File, with oversamples of selected metropolitan areas and HUD-assisted units | About 85,000 national units; about 115,000 in total with oversamples |
| 2025 | New sample drawn | Size not stated on the history page |
The survey was annual through 1981, then biennial. Paper questionnaires ended in 1997, and since 2011 the questionnaire has a core of recurring questions plus rotating supplements. Collection runs from May to September of odd-numbered years, and the Bureau reports that it is typically able to include no more than 30 metropolitan areas in a given year because of budget. A planned move to continuous monthly collection has been delayed to January 2026 or later.
Why this matters for comparisons: a new sample means a new set of housing units, so check the Census methodology notes before treating a difference across a redesign boundary as real change. Metro estimates exist only for the metros included in that year.
What the AHS measures
Core topics recur each cycle: housing inventory, vacancies, owners and renters, physical condition, equipment breakdowns, occupants, neighborhood quality, mortgages and costs, fuel, improvements, assisted housing, recent movers and home values. Rotating modules add current themes; the 2023 cycle included housing insecurity, healthy homes, power outages, heat risk and disaster-related moves (Census Bureau, AHS history page, read 2026-11-08).
Release schedule and how to access the data
Public use files (PUFs) and summary tables are released approximately 12 months after collection. HUD User announced the 2023 national and metro PUF and summary tables on 2024-09-25. Two routes exist: the Census Table Creator builds custom tables for the United States, selected states and metros for 2011 and later, while the PUF supports your own microdata analysis. If you need only a published statistic, use Table Creator; if you need a regression or a subgroup not in the tables, use the PUF.
Formula with every symbol explained
Survey microdata carry a weight per record so that estimates represent the housing-unit population rather than just the sampled units. A weighted mean is
x̄w = Σ(wi × xi) / Σ wi, where xi is the value for record i (for example monthly housing cost), wi is that record’s weight, and the sums run over all records.
- x̄w: the weighted mean, the estimate for the housing-unit population.
- xi: the value of the variable for record i, such as monthly housing cost.
- wi: the weight on record i, roughly the number of housing units that record represents.
- Σ: sum over all records i in the analysis file (or subgroup).
Worked example: why weights matter in microdata
The numbers below are a toy illustration, not AHS data. Three records have monthly costs of 800, 1,200 and 1,500 and weights of 1,000, 2,000 and 500.
- Numerator: 800 × 1,000 + 1,200 × 2,000 + 1,500 × 500 = 800,000 + 2,400,000 + 750,000 = 3,950,000
- Denominator: 1,000 + 2,000 + 500 = 3,500
- Weighted mean: 3,950,000 / 3,500 = 1,128.57
- Unweighted mean for comparison: (800 + 1,200 + 1,500) / 3 = 1,166.67
The two differ because the record with cost 1,200 represents more housing units. Take the actual weight variable name, and the guidance on variance estimation, from the codebook and methodology documentation released with the file you use.
Interpretation
Use this table as a reading guide for AHS estimates. The Census Bureau documentation released with each file governs the exact methods.
| Situation | How to read it |
|---|---|
| Weighted vs unweighted figure | A weighted estimate describes the housing-unit population; an unweighted one describes only the sampled units. They differ when records represent different numbers of units, as in the toy example (1,128.57 vs 1,166.67). |
| Comparison across a redesign (for example 2015, when a new national sample was drawn, or 2025) | The sampled units changed, so check the Census methodology notes before reading a difference as real change. |
| Metro estimate | Exists only for metros included in that survey year (no more than 30 in a given year), so a metro trend needs the metro in every year compared. |
| Subgroup or small area | Take the variance-estimation guidance from the file documentation rather than treating the estimate as exact. |
Assumptions and common mistakes
- Ignoring the weights, which describes the sample instead of the population.
- Comparing metro figures across years without checking that the metro was in both samples.
- Treating a redesign boundary as if the sample were continuous.
- Using the AHS for what it does not cover. It is a survey of housing units; for rent benchmarks see fair market rents, and for area-level indexes see the social vulnerability index and area deprivation index.
How to compute a weighted mean in Excel, R or Python
Replace the column names with those in the codebook for your file.
Excel: =SUMPRODUCT(B2:B4, C2:C4) / SUM(C2:C4) with values in column B and weights in column C.
R: weighted.mean(x = df$cost, w = df$weight)
Python: import numpy as np; np.average(df["cost"], weights=df["weight"])
On the toy data above, all three return 1,128.57.
How to report it
Report the survey, the year, the file type (public use file or Table Creator), and whether estimates are weighted. The general APA 7 pattern for a dataset reference is: Author. (Year). Title of dataset [Data set]. Publisher. URL. For the AHS, the author is the U.S. Census Bureau, and HUD is the sponsor. Take the exact file title, release year and URL from the Census page for the file you used, and check whether that page gives its own suggested citation, which takes precedence.
Pattern: U.S. Census Bureau. (release year). American Housing Survey [survey year] public use file [Data set]. U.S. Department of Commerce. URL of the file page
In text: (U.S. Census Bureau, release year). In the methods section also state the weight variable used, as named in the codebook, and the software (for example R, with weighted.mean).
Frequently asked questions
What is a housing survey?
A survey that collects data on housing units and their occupants. In U.S. research and policy it usually refers to the American Housing Survey run by the Census Bureau for HUD.
How often is the AHS conducted?
Every two years, in odd-numbered years, with collection between May and September.
Where does the sample come from?
Since 2015 the national sample has been drawn from the Census Bureau’s Master Address File; a new sample was drawn in 2025.
How is it different from a home survey?
A home inspection or property survey assesses one building or parcel for a buyer or lender. The AHS is a statistical survey of the housing stock. For the planning context in which such data are used, see what is urban planning.
Sources
- U.S. Census Bureau. American Housing Survey: About. https://www.census.gov/programs-surveys/ahs/about.html (read 2026-11-08)
- U.S. Census Bureau. AHS Introduction and History. https://www.census.gov/programs-surveys/ahs/about/ahs-introduction-history.html (read 2026-11-08)
- HUD User. 2023 AHS release announcement, 2024-09-25. https://www.huduser.gov/portal/elist/2024-Sept-25.html
