Methodology

The survey

The American Housing Survey is sponsored by the U.S. Department of Housing and Urban Development and conducted by the U.S. Census Bureau. In 2023 it asked homeowners about the home improvement jobs of the two years before their interview: the kind of job, its cost, who did most of the work, the year and the main source of the money.

We use the 2023 national and metropolitan public use files. Data source and how to check our numbers

Which jobs are counted

• Owner-occupied homes. The survey asks only owners about home improvement jobs.
• Jobs with a cost above $0.
• Costs the owner reported. The public file marks every value the Census Bureau changed. When an owner did not give a cost, the Bureau filled one in from similar households. Of the 44,689 job records in the national file, 6,794 have a cost that was filled in or edited. We leave them out. For the 16 job types on this site, that removes 18% of the professionally done jobs and leaves 18,698.
• A known answer to "who did the work". The same rule applies to the 144 records where the Bureau filled in whether the household did the work.

A job enters a group, such as a home size or a type of heating system, only if the Census Bureau did not change that value for the home. Lot size is the one exception: for 94% of the jobs in our lot size tables the Bureau filled it in, and the public file does not say which. Those tables say so.

What a cost includes

The survey asks: "How much did the entire … job cost, including your costs and any amount covered by insurance, not counting your time?" A cost here is therefore the full cost of the job, including any part an insurer paid, without the household's own time. The Census Bureau rounds the costs in the public file and caps the very largest ones.

Median, middle half, low end and high end

• Median: half of the jobs cost less and half cost more.
• Middle half: the range from the 25th to the 75th percentile. One job in four cost less than the lower figure and one in four cost more than the upper figure.
• "1 in 10 cost less" and "1 in 10 cost more": the 10th and 90th percentiles.

Jobs are weighted with the survey's household weights. Percentiles are not interpolated: each one is the cost of one job. Calculated this way on the costs as reported, our medians match the Census Bureau's own verification table in all 35 rows.

We show medians, not averages: a few very expensive jobs pull an average well above what most jobs cost.

The 30-job rule

A figure is shown only for a group with at least 30 jobs. Thirty jobs still give only a rough figure, so every table and every calculator result shows its job count.

How the calculator combines your choices

The calculator reads its result from one group of real jobs. A group combines at most two of your choices; more would leave too few jobs in most groups.

• With one choice, the result is that group.
• With two choices, the result is the group that matches both, if it has at least 30 jobs. If it does not, the result uses one of the two and says so.
• With three or more choices, the result combines the two that matter most for that job, and each choice is also listed on its own. The order differs by job: for a roof replacement it is home size first, then location.

If no group of at least 30 jobs matches, the calculator says so. It does not model a result.

The quote check

The calculator places your quote among the jobs in the group you chose and reports the share that cost less. It works from the group's percentiles in steps of five, so the answer is worded with "about"; below the 5th or above the 95th percentile it is "lower than at least 95%" or "higher than at least 95%". A quote above the median is not necessarily too high: the survey does not record the size of a job or the materials.

Location

The public files identify four regions, nine Census divisions and 20 metro areas (15 in the national file, 5 in the metropolitan file), not states. When you pick a state, the calculator uses that state's Census division and says so; if the division has too few jobs, the region.

Home size, home value and other features of the home

These describe the home, not the job. Home size is the owner's estimate of the square footage of the whole home, with a basement and without garages, unfinished attics and open porches. Home value is the owner's estimate of what the property would sell for; the Census Bureau multiplies each value in the public file by a random factor that averages 1, which is why we use four wide groups. Stories include a basement and a finished attic.

Years and today's prices

The jobs were done in 2021, 2022 and 2023. The figures are in 2026 prices: each reported cost is multiplied by a factor for the year of the job.

The index is the price index for residential improvements (home improvements) of the U.S. Bureau of Economic Analysis, the BEA (National Income and Product Accounts, Table 5.4.4, line 46; 2017 = 100). The BEA builds it as the average of three series: the price index for new one-family houses under construction, the producer price index for home maintenance and repair, and the employment cost index for the construction industry. Its latest value is 158.793, for the second quarter of 2026; the factor for a year is that value divided by that year's average.

• It is not a new price survey. A figure in 2026 prices is what owners reported, carried forward by the change in the index.
• One index for every kind of job. The price of a single kind of job may have moved more or less than the index.
• Year, not month. The survey records the year of a job, so the average of that year is used. 71 of the 18,698 jobs have no year on record and get the 2022 factor (the middle year).
• Rounding. Each cost in 2026 prices is rounded to the nearest $100; below $1,000 to the nearest $10, and below $100 to the nearest dollar.
• The reported costs stay visible. Each job page has a table by year with the costs as reported.
• Updates. The BEA publishes the index every quarter and revises earlier values. The figures here use its file released September 30, 2026.

Does the index move the way reported costs did?

The survey was also taken in 2019 and 2021. For the 16 job types on this site, we compared how reported costs changed from one survey to the next with the change in the index over the years in which those jobs were done.

Over the whole period the two are a few percentage points apart; between consecutive surveys they differ more.

Reported costs: the typical cost of each job (geometric mean, survey weights), with the change averaged over the 16 job types. Jobs compared: 19,170 in the 2019 survey, 20,709 in the 2021 survey, 18,698 in the 2023 survey.

Federal estimates by type of unit

The survey does not record the type of unit installed. The pages for that equipment (HVAC replacement, furnace replacement, water heater replacement) show estimates from the U.S. Energy Information Administration in a separate table: its report "Updated Buildings Sector Appliance and Equipment Costs and Efficiencies" (Appendix A, released March 23, 2023) gives the total installed cost of a typical unit and of a high-efficiency unit of each type.

• They are estimates, not survey records. Every table and calculator result that shows one says so, and they never enter a survey figure.
• One unit of a given size, the size the agency chose as typical; the table shows it. Where the report gives a low and a high figure, both are shown.
• Prices. The agency's figures are in 2022 dollars. We multiply them by 1.14, the rise of the index above from 2022 to the second quarter of 2026, and round them like the survey costs.
• Real projects come first. Where a city's permit records cover the same type of unit, the page shows them before the estimate and says where the two differ; the calculator shows the records for such a type and the estimate for the others.

Federal laboratory estimates by type of insulation

The survey does not record the type of insulation or the area. The insulation page shows estimates from the National Laboratory of the Rockies, a national laboratory of the U.S. Department of Energy, in a separate section: its "National Residential Efficiency Measures Database" (version 4.0.0, released April 3, 2024) gives a low, a middle and a high cost per square foot for each type, rising with the R-value.

• They are estimates, not survey records. The laboratory bases them on the sources it lists for each type (retail and distributor websites, RS Means and Homewyse). They never enter a survey figure.
• What is shown. The cost of installing insulation in an existing home (the database's retrofit cost) for 4 parts of a home (attic floor, exterior walls, floor over a crawlspace and basement walls), at selected R-values within the range the database covers, calculated with the database's own equation.
• Which types. A type is shown only if the laboratory's own note reports a medium or large sample and "high source diversity" for it. That is the case for 13 types; 5 are left out.
• Prices. The laboratory's figures are in 2023 dollars. We multiply them by 1.13, the rise of the index above from 2023 to the second quarter of 2026. Costs per square foot are shown to the cent.
• Totals. The calculator multiplies the three costs per square foot by the area you enter (50 to 10,000 square feet) and rounds the totals like the survey costs.

The laboratory's data is free to use, copy, modify and distribute for any purpose, provided its notice stays with copies of the data and DOE/NLR/ALLIANCE is credited. Those names may not be used to endorse a product. The laboratory has not reviewed or endorsed this site. The full notice is on the data source page.

County and city permit records

The survey does not record the material or the size of a job. Some counties and cities publish their building permits, and a permit can carry the kind of work, a size and a dollar value. On 5 pages a separate section shows the permit records of one place by material, by size or by type of unit.

• They are one county's or city's records, not survey records. That place has its own building code and prices. Every table and calculator result that shows them names the place, and they never enter a survey figure.
• Which records. Permits issued in the last two years. What is left out (permits without a value, repairs, permits that legalize work already done, work in newly built homes) is listed under "About these records" on each page.
• The value is the dollar figure written on the permit, not adjusted for price changes and rounded like the other costs.
• Sizes are the square footage or the length on the permit; the cost per square foot or per linear foot is the value divided by that size.
• The kind of work comes from a field on the permit or, where it appears only in the description, from that text by fixed rules.
• The test a place has to pass. A place is used only while the median of its records is within 25% of the median that homeowners in the same metro area reported for the same kind of job (2019, 2021 and 2023 surveys combined, in 2026 prices). The table shows the result for every place on the site.

The City of Seattle makes no claims as to the completeness, accuracy, timeliness or content of this data; it is used at one's own risk. Montgomery County makes this data available on an "as is" and "as available" basis and makes no representations or warranties as to its completeness, accuracy or timeliness. Miami-Dade County provides this data for use "as is" and does not guarantee that it is accurate, correct or complete. The City of Chicago makes this data available on an "as is" basis and makes no warranty as to its content, accuracy, timeliness or completeness. None of these counties and cities has reviewed or endorsed this site. The notices their terms of use require are on the data source page.

Do-it-yourself jobs

A job counts as do-it-yourself when someone in the household did most of the work. Its cost leaves out the household's own time, so it is not comparable with a contractor's price.

What the data cannot show

• The size of a job, the materials, the brand or the model. The permit records and the federal estimates described above fill part of that gap on some pages.
• Kinds of work the survey puts in one category, such as windows and doors, or flooring, carpet and drywall.
• States, counties or cities other than the listed metro areas.
• Jobs the survey does not ask about as their own category, such as gutters, garage doors, concrete slabs or painting.

Checking our work

Each time our data is built, the job counts, medians and means of the Census Bureau's verification table are recalculated from the file first. If one row does not match, nothing is published. All 35 of 35 rows match; the data source page shows them.