Original research · Demand
Homes Entering the Roof Replacement Window
A cohort model over the Census Bureau’s year-structure-built table, with its assumption published, run at three values, and tested against a federal survey that counted the actual jobs.
Written by HyreRoof Research Primary-source research and fact checking
The finding
Read this before you read the model
No federal dataset records the age of a roof. The American Community Survey asks when the building was first constructed: the Census Bureau’s own definition is explicit that year structure built refers to construction, “not when it was remodeled, added to, or converted”, and never asks about the covering.
So housing age is a proxy for roof age, and it is a weak one: a roof is replaced several times in a building’s life, and the roof on a 1955 house is on average much younger than the house.
That is not a reason to abandon the question. It is a reason to model it out loud. Everything below turns a cohort of buildings into an estimate of replacement events using one assumption (the average replacement cycle) which is stated, published as a formula, and run at three values rather than presented as a single confident number. Where the model and a measured federal count disagree, the page says so and works out why.
If you want the housing-age data on its own, without the model, that is the companion study: America’s oldest housing stock, city by city. If you want an estimate for one specific roof rather than for a national stock, the roof life calculator is the right instrument and this page is not.
The shape of the American housing stock
There are 142,332,876 housing units in the United States (ACS 2023 5-year, table B25034, margin of error ±9,434). They are not evenly spread across the decades, and the unevenness is the whole reason this question is worth asking.
The single largest cohort is the 1970s at 20,484,570 units, homes that were between 44 and 53 years old in 2023. The pre-1940 cohort is the second largest at 16,917,912. HyreRoof analysis: that second figure is worth pausing on, because it inverts the usual assumption about American housing. There is more pre-war housing standing today (16,917,912 units) than there is housing from the entire 2010s (12,736,038). The stock is old, and it is old at the tail rather than merely in the middle.
The newest cohort, built 2020 or later, is 1,644,980 units, 1.2% of the stock. Almost every roof in America is on a building that predates the pandemic, and three quarters of them are on a building that predates the euro.
Every cohort, with its margin of error
| Cohort | Housing units | MOE | Share of stock | Age in 2023 |
|---|---|---|---|---|
| Built 2020 or later | 1,644,980 | ±11,011 | 1.2% | 0–3 years |
| Built 2010 to 2019 | 12,736,038 | ±30,455 | 8.9% | 4–13 years |
| Built 2000 to 2009 | 19,324,640 | ±35,744 | 13.6% | 14–23 years |
| Built 1990 to 1999 | 18,211,985 | ±33,159 | 12.8% | 24–33 years |
| Built 1980 to 1989 | 18,543,944 | ±33,251 | 13.0% | 34–43 years |
| Built 1970 to 1979 | 20,484,570 | ±35,137 | 14.4% | 44–53 years |
| Built 1960 to 1969 | 14,254,921 | ±31,903 | 10.0% | 54–63 years |
| Built 1950 to 1959 | 13,784,571 | ±27,895 | 9.7% | 64–73 years |
| Built 1940 to 1949 | 6,429,315 | ±18,285 | 4.5% | 74–83 years |
| Built 1939 or earlier | 16,917,912 | ±34,312 | 11.9% | 84+ years |
| Total housing units | 142,332,876 | ±9,434 | 100% | , |
US Census Bureau, ACS 2023 5-year estimates, table B25034 (Year Structure Built), universe: housing units. Retrieved 3 September 2026.
The margins are large in absolute terms and small in relative terms: the biggest, on the 1970s cohort, is ±35,137 against an estimate of 20,484,570, or 0.17 per cent. At national level these cohorts are precise enough to model on. At county level they are not, which is why this study stops at metropolitan areas.
The model, and the assumption it rests on
The arithmetic is deliberately simple, because a simple model can be argued with and a complicated one cannot. Take a replacement cycle of L years. Every home that has reached age L has been through at least one replacement, and in a steady state it generates one replacement every L years. So:
Annual replacements ≈ (housing units aged L or more) ÷ L
Two inputs. The numerator comes straight out of B25034, with one interpolation: the cohorts are decade bins, so reaching a threshold inside a decade requires assuming units are spread evenly across that decade’s years. The denominator is the assumption, and it is the whole ballgame.
Why three values rather than one. Federal sources do not agree on a single figure and it would be dishonest to pick one. HUD’s Residential Rehabilitation Inspection Guide states that asphalt shingles have “a service life of about 20 years for the first layer and about 15 years for a second layer added over the first layer, depending on their weight, quality, and exposure.” HUD’s Durability by Design gives roofing a range of 15 to 30 years.
HUD’s Rehab Guide Volume 3: Roofs records that the Insurance Institute for Property Loss Prevention “assumes that the effective life of an average asphalt shingle is 17 years.” And the same guide notes that shingles are sold on warranty durations of 20, 25, 30 or 40 years while warning that “there is no direct relationship between base mat thickness, shingle weight, performance, and warranty.”
So the model runs at 20, 25 and 30 years. Twenty is HUD’s first-layer asphalt figure. Thirty is the top of HUD’s roofing range. Twenty-five sits between them as a central case. Every table on this page publishes all three columns, and no sentence anywhere presents one of them as the answer.
The same country, three assumptions
One dataset, one formula, three service-life values. The spread between them, 2,561,186 replacements a year, is larger than the entire annual output of most state roofing markets, and it comes from nothing but the choice of assumption.
116,357,074 units aged 20+ (81.7% of stock) · 40.9 per 1,000 units HUD: asphalt shingles have “a service life of about 20 years for the first layer”
106,806,020 units aged 25+ (75.0% of stock) · 30.0 per 1,000 units Midpoint of the 15–30 year roofing range in HUD’s Durability by Design
97,700,027 units aged 30+ (68.6% of stock) · 22.9 per 1,000 units Top of that same 15–30 year range
HyreRoof calculation, from ACS 2023 5-year table B25034. These are not Census figures and the Census Bureau has not published anything of this kind. The formula and the interpolation are in Method so the numbers can be reproduced or rejected.
The reality check: a federal survey that counted the jobs
A model of roof replacements is worth very little without something measured to test it against, and one exists. The American Housing Survey, sponsored by HUD and conducted by the Census Bureau, asks owners what major improvements were done to the home in the last two years, and defines its roofing category precisely: “This is meant to capture the replacement of the entire roof or at least most of it.
Anything less, such as the repair of a hole or leak, or the replacement of a small section, would be considered maintenance.” That is a re-roof, as a roofer would define one.
The 2023 survey recorded 8,300,000 roofing projects and $93.5 billion of roofing expenditure across the two years to 2023, with a median project cost of $10,000. Roofing was the largest single improvement category by total spending, ahead of kitchen remodels at $83.1 billion and HVAC at $82.4 billion.
HyreRoof calculation. 8,300,000 projects over two years is 4,150,000 a year. AHS covers owner-occupied homes: it counts 86,900,000 homeowners, of whom 51,600,000 made improvements. Dividing gives a crude average interval of 20.9 years between roof jobs per owner-occupied home. Running the same interval against the ACS count of owner-occupied units (82,892,037, table B25036) gives 20.0 years.
The age-adjusted answer is shorter still. Not every owner-occupied home is old enough to have been re-roofed, so the honest calculation solves for the cycle length L at which our own model, run over the owner-occupied cohorts in B25036, produces 4,150,000 events a year. The answer is L ≈ 17.1 years. At a 20-year cycle the model predicts 3,365,271 owner-occupied replacements a year; at 25 years, 2,445,572. Both are below what the survey measured.
HyreRoof analysis. Three independent numbers land in the same place: HUD’s 20-year first-layer service life, the 17-year effective life the same agency attributes to the Insurance Institute for Property Loss Prevention, and a 17.1-year interval implied by a completely separate federal survey counting actual jobs.
The convergence is the strongest evidence on this page that the low end of the assumption range is closer to American practice than the high end, and it is the reason the 30-year column, which is the one most often quoted in marketing, should be treated as an upper bound rather than a central estimate.
Four reasons the measured interval runs short
A roof-over resets a shorter clock
The AHS definition explicitly allows it: “The roof may replace or be installed over the old roofing materials.” HUD puts the service life of a second layer laid over a first at about fifteen years against about twenty for the first, so a stock with roof-overs in it cycles faster than a stock without them, and hits the two-layer limit, after which a tear-off is unavoidable.
Storms do not wait for the service life
The same survey recorded 663,000 occupied housing units with tornado or hurricane repairs and 129,000 with fire repairs. Hail and wind claims replace roofs on a schedule set by weather rather than by wear, and in an insurance-driven market that pulls the average interval down without any roof having failed.
Projects are counted, not homes
AHS counts improvement projects reported by a household over two years. A home that had a roof replaced twice in that window: a repair after a storm, then a full replacement, could contribute more than one. Our arithmetic treats each project as one replacement event, which biases the implied interval short.
The two universes are not identical
AHS 2023 is a single-year estimate counting 86,900,000 homeowners; ACS 2023 5-year is a five-year average counting 82,892,037 owner-occupied units. The gap is about 4.8 per cent and it runs in the direction that makes our implied interval slightly too short. It does not change the conclusion; it is one of the reasons we say “about 17 years” rather than quoting a decimal.
Every state, at all three assumptions
| State | Housing units | Aged 20+ | Aged 25+ | Aged 30+ | Annual at 25 yr | Per 1,000 units |
|---|---|---|---|---|---|---|
| Rhode Island | 484,615 | 448,908 (92.6%) | 433,666 (89.5%) | 416,599 (86.0%) | 17,347 | 35.8 |
| New York | 8,539,536 | 7,762,067 (90.9%) | 7,495,769 (87.8%) | 7,239,315 (84.8%) | 299,831 | 35.1 |
| Connecticut | 1,536,049 | 1,398,095 (91.0%) | 1,343,061 (87.4%) | 1,284,753 (83.6%) | 53,722 | 35.0 |
| Massachusetts | 3,014,657 | 2,692,405 (89.3%) | 2,584,845 (85.7%) | 2,472,085 (82.0%) | 103,394 | 34.3 |
| Pennsylvania | 5,779,663 | 5,183,582 (89.7%) | 4,945,502 (85.6%) | 4,685,816 (81.1%) | 197,820 | 34.2 |
| New Jersey | 3,775,842 | 3,328,369 (88.1%) | 3,156,715 (83.6%) | 2,985,831 (79.1%) | 126,269 | 33.4 |
| Michigan | 4,599,683 | 4,079,810 (88.7%) | 3,835,391 (83.4%) | 3,549,172 (77.2%) | 153,416 | 33.4 |
| Ohio | 5,271,573 | 4,654,975 (88.3%) | 4,393,516 (83.3%) | 4,093,889 (77.7%) | 175,741 | 33.3 |
| Puerto Rico | 1,575,105 | 1,412,304 (89.7%) | 1,307,949 (83.0%) | 1,188,312 (75.4%) | 52,318 | 33.2 |
| Illinois | 5,443,501 | 4,809,707 (88.4%) | 4,512,134 (82.9%) | 4,222,021 (77.6%) | 180,485 | 33.2 |
| Vermont | 337,072 | 292,990 (86.9%) | 275,318 (81.7%) | 257,737 (76.5%) | 11,013 | 32.7 |
| California | 14,532,683 | 12,554,581 (86.4%) | 11,758,724 (80.9%) | 11,008,895 (75.8%) | 470,349 | 32.4 |
| New Hampshire | 644,253 | 555,196 (86.2%) | 518,915 (80.5%) | 486,063 (75.4%) | 20,757 | 32.2 |
| West Virginia | 859,653 | 741,999 (86.3%) | 692,378 (80.5%) | 636,662 (74.1%) | 27,695 | 32.2 |
| Maine | 746,552 | 636,559 (85.3%) | 592,225 (79.3%) | 551,112 (73.8%) | 23,689 | 31.7 |
| Kansas | 1,285,221 | 1,089,779 (84.8%) | 1,016,229 (79.1%) | 939,318 (73.1%) | 40,649 | 31.6 |
| Wisconsin | 2,750,750 | 2,342,850 (85.2%) | 2,172,022 (79.0%) | 1,994,430 (72.5%) | 86,881 | 31.6 |
| Maryland | 2,545,532 | 2,153,608 (84.6%) | 2,003,904 (78.7%) | 1,832,531 (72.0%) | 80,156 | 31.5 |
| Iowa | 1,427,175 | 1,195,140 (83.7%) | 1,119,768 (78.5%) | 1,048,655 (73.5%) | 44,791 | 31.4 |
| Hawaii | 564,905 | 477,410 (84.5%) | 441,875 (78.2%) | 404,542 (71.6%) | 17,675 | 31.3 |
| Indiana | 2,953,344 | 2,484,281 (84.1%) | 2,299,893 (77.9%) | 2,094,923 (70.9%) | 91,996 | 31.1 |
| District of Columbia | 356,101 | 288,973 (81.1%) | 276,889 (77.8%) | 271,501 (76.2%) | 11,076 | 31.1 |
| Nebraska | 855,631 | 710,069 (83.0%) | 661,399 (77.3%) | 615,378 (71.9%) | 26,456 | 30.9 |
| Missouri | 2,809,501 | 2,352,159 (83.7%) | 2,169,137 (77.2%) | 1,982,568 (70.6%) | 86,765 | 30.9 |
| Minnesota | 2,519,538 | 2,089,675 (82.9%) | 1,923,607 (76.3%) | 1,763,447 (70.0%) | 76,944 | 30.5 |
| Wyoming | 275,131 | 223,696 (81.3%) | 205,306 (74.6%) | 189,435 (68.9%) | 8,212 | 29.8 |
| New Mexico | 949,524 | 780,538 (82.2%) | 707,167 (74.5%) | 629,364 (66.3%) | 28,287 | 29.8 |
| Kentucky | 2,010,655 | 1,642,263 (81.7%) | 1,494,419 (74.3%) | 1,335,492 (66.4%) | 59,777 | 29.7 |
| Oregon | 1,838,631 | 1,494,903 (81.3%) | 1,363,656 (74.2%) | 1,221,379 (66.4%) | 54,546 | 29.7 |
| Oklahoma | 1,763,036 | 1,414,683 (80.2%) | 1,303,122 (73.9%) | 1,208,374 (68.5%) | 52,125 | 29.6 |
| Alaska | 327,610 | 266,538 (81.4%) | 241,543 (73.7%) | 219,630 (67.0%) | 9,662 | 29.5 |
| Virginia | 3,654,784 | 2,947,900 (80.7%) | 2,677,786 (73.3%) | 2,411,032 (66.0%) | 107,111 | 29.3 |
| Montana | 522,939 | 413,630 (79.1%) | 377,407 (72.2%) | 344,235 (65.8%) | 15,096 | 28.9 |
| Louisiana | 2,094,002 | 1,655,884 (79.1%) | 1,510,466 (72.1%) | 1,386,635 (66.2%) | 60,419 | 28.9 |
| Alabama | 2,316,192 | 1,829,268 (79.0%) | 1,647,260 (71.1%) | 1,457,554 (62.9%) | 65,890 | 28.4 |
| South Dakota | 402,364 | 313,008 (77.8%) | 285,476 (70.9%) | 261,810 (65.1%) | 11,419 | 28.4 |
| Washington | 3,262,667 | 2,549,603 (78.1%) | 2,303,689 (70.6%) | 2,054,739 (63.0%) | 92,148 | 28.2 |
| Mississippi | 1,332,811 | 1,050,771 (78.8%) | 940,635 (70.6%) | 834,255 (62.6%) | 37,625 | 28.2 |
| Arkansas | 1,382,664 | 1,076,841 (77.9%) | 966,021 (69.9%) | 851,741 (61.6%) | 38,641 | 27.9 |
| Tennessee | 3,095,472 | 2,394,172 (77.3%) | 2,147,888 (69.4%) | 1,897,126 (61.3%) | 85,916 | 27.8 |
| North Dakota | 374,866 | 276,292 (73.7%) | 255,149 (68.1%) | 236,952 (63.2%) | 10,206 | 27.2 |
| Florida | 10,082,356 | 7,746,772 (76.8%) | 6,861,904 (68.1%) | 6,092,033 (60.4%) | 274,476 | 27.2 |
| Colorado | 2,545,124 | 1,933,428 (76.0%) | 1,724,902 (67.8%) | 1,531,940 (60.2%) | 68,996 | 27.1 |
| Delaware | 457,958 | 345,122 (75.4%) | 306,394 (66.9%) | 275,224 (60.1%) | 12,256 | 26.8 |
| North Carolina | 4,815,195 | 3,575,618 (74.3%) | 3,133,196 (65.1%) | 2,709,014 (56.3%) | 125,328 | 26.0 |
| Georgia | 4,483,873 | 3,369,174 (75.1%) | 2,911,218 (64.9%) | 2,492,041 (55.6%) | 116,449 | 26.0 |
| Idaho | 776,683 | 564,603 (72.7%) | 494,297 (63.6%) | 433,316 (55.8%) | 19,772 | 25.5 |
| South Carolina | 2,401,638 | 1,745,287 (72.7%) | 1,528,189 (63.6%) | 1,323,910 (55.1%) | 61,128 | 25.5 |
| Arizona | 3,142,443 | 2,322,999 (73.9%) | 1,980,826 (63.0%) | 1,694,786 (53.9%) | 79,233 | 25.2 |
| Utah | 1,193,082 | 842,113 (70.6%) | 736,101 (61.7%) | 644,081 (54.0%) | 29,444 | 24.7 |
| Texas | 11,890,808 | 8,332,217 (70.1%) | 7,320,703 (61.6%) | 6,523,452 (54.9%) | 292,828 | 24.6 |
| Nevada | 1,307,338 | 926,564 (70.9%) | 758,414 (58.0%) | 603,231 (46.1%) | 30,337 | 23.2 |
Housing units from US Census Bureau, ACS 2023 5-year table B25034. The aged-20+, aged-25+ and aged-30+ columns and both right-hand columns are HyreRoof calculations from that table, described in Method. Ordered by the share of stock aged 25 or more. Retrieved 3 September 2026.
The spread on the last column is 35.8 per 1,000 units a year in Rhode Island against 23.2 in Nevada: a 1.5-to-1 range in modelled replacement intensity, driven entirely by when each state built its housing. New York, Connecticut, Massachusetts and Pennsylvania sit at the top with Rhode Island; Texas, Utah, Arizona and South Carolina at the bottom with Nevada.
The 46 largest metropolitan areas
The ladder separates two kinds of roofing market almost cleanly. At the top, Providence–Warwick at 89.0 per cent, Pittsburgh 87.7, Cleveland 87.6, Los Angeles–Long Beach–Anaheim 87.1, New York–Newark–Jersey City 87.0: nearly nine in ten homes are past a 25-year cycle. These are replacement markets, and they have been for a long time.
At the bottom, Austin–Round Rock–San Marcos at 47.6 per cent, Raleigh–Cary 50.8, Las Vegas–Henderson–North Las Vegas 54.5, Charlotte–Concord–Gastonia 57.8, Houston–Pasadena–The Woodlands 58.7, between two fifths and half the stock has never reached the threshold at all.
HyreRoof analysis: the share and the volume tell opposite stories and both are true. Austin has the lowest share of any large metro and still holds 476,172 units past 25 years. Greater New York has a similar share to Cleveland and eight times the units. A contractor reads the volume column; an analyst reads the share column; and a page that publishes only one of them is misleading somebody.
The metro table
| Metropolitan area | Median yr built | Housing units | Aged 25+ | Share aged 25+ | Annual at 25 yr |
|---|---|---|---|---|---|
| New York-Newark-Jersey City, NY-NJ | 1959 | 7,991,914 | 6,950,394 | 87.0% | 278,016 |
| Los Angeles-Long Beach-Anaheim, CA | 1969 | 4,762,557 | 4,147,322 | 87.1% | 165,893 |
| Chicago-Naperville-Elgin, IL-IN | 1970 | 3,886,535 | 3,201,407 | 82.4% | 128,056 |
| Philadelphia-Camden-Wilmington, PA-NJ-DE-MD | 1966 | 2,608,248 | 2,222,347 | 85.2% | 88,894 |
| Miami-Fort Lauderdale-West Palm Beach, FL | 1982 | 2,662,397 | 2,061,943 | 77.4% | 82,478 |
| Dallas-Fort Worth-Arlington, TX | 1992 | 3,030,663 | 1,831,120 | 60.4% | 73,245 |
| Washington-Arlington-Alexandria, DC-VA-MD-WV | 1983 | 2,483,270 | 1,807,870 | 72.8% | 72,315 |
| Boston-Cambridge-Newton, MA-NH | 1963 | 2,046,121 | 1,717,562 | 83.9% | 68,702 |
| Houston-Pasadena-The Woodlands, TX | 1992 | 2,839,496 | 1,667,110 | 58.7% | 66,684 |
| Detroit-Warren-Dearborn, MI | 1968 | 1,914,787 | 1,641,544 | 85.7% | 65,662 |
| San Francisco-Oakland-Fremont, CA | 1967 | 1,866,271 | 1,590,031 | 85.2% | 63,601 |
| Atlanta-Sandy Springs-Roswell, GA | 1993 | 2,462,125 | 1,503,968 | 61.1% | 60,159 |
| Phoenix-Mesa-Chandler, AZ | 1993 | 2,030,723 | 1,225,291 | 60.3% | 49,012 |
| Seattle-Tacoma-Bellevue, WA | 1985 | 1,681,971 | 1,179,246 | 70.1% | 47,170 |
| Minneapolis-St. Paul-Bloomington, MN-WI | 1980 | 1,530,451 | 1,154,381 | 75.4% | 46,175 |
| Riverside-San Bernardino-Ontario, CA | 1986 | 1,598,577 | 1,118,990 | 70.0% | 44,760 |
| Tampa-St. Petersburg-Clearwater, FL | 1985 | 1,492,509 | 1,066,615 | 71.5% | 42,665 |
| Pittsburgh, PA | 1960 | 1,169,819 | 1,026,114 | 87.7% | 41,045 |
| St. Louis, MO-IL | 1973 | 1,267,353 | 1,018,603 | 80.4% | 40,744 |
| San Diego-Chula Vista-Carlsbad, CA | 1980 | 1,240,607 | 987,157 | 79.6% | 39,486 |
| Baltimore-Columbia-Towson, MD | 1975 | 1,195,538 | 971,697 | 81.3% | 38,868 |
| Cleveland, OH | 1962 | 1,016,658 | 890,446 | 87.6% | 35,618 |
| Denver-Aurora-Centennial, CO | 1986 | 1,268,353 | 863,318 | 68.1% | 34,533 |
| San Juan-Bayamón-Caguas, PR | 1977 | 989,904 | 825,327 | 83.4% | 33,013 |
| Cincinnati, OH-KY-IN | 1975 | 961,797 | 761,142 | 79.1% | 30,446 |
| Portland-Vancouver-Hillsboro, OR-WA | 1983 | 1,050,449 | 754,457 | 71.8% | 30,178 |
| Kansas City, MO-KS | 1979 | 950,404 | 714,602 | 75.2% | 28,584 |
| Sacramento-Roseville-Folsom, CA | 1983 | 945,650 | 695,329 | 73.5% | 27,813 |
| Columbus, OH | 1981 | 915,425 | 673,202 | 73.5% | 26,928 |
| Orlando-Kissimmee-Sanford, FL | 1994 | 1,120,478 | 664,504 | 59.3% | 26,580 |
| Charlotte-Concord-Gastonia, NC-SC | 1994 | 1,139,944 | 658,374 | 57.8% | 26,335 |
| Providence-Warwick, RI-MA | 1962 | 728,781 | 648,374 | 89.0% | 25,935 |
| Indianapolis-Carmel-Greenwood, IN | 1982 | 898,849 | 645,426 | 71.8% | 25,817 |
| San Antonio-New Braunfels, TX | 1992 | 1,037,336 | 611,248 | 58.9% | 24,450 |
| Milwaukee-Waukesha, WI | 1966 | 695,586 | 588,561 | 84.6% | 23,542 |
| San Jose-Sunnyvale-Santa Clara, CA | 1975 | 715,146 | 572,923 | 80.1% | 22,917 |
| Virginia Beach-Chesapeake-Norfolk, VA-NC | 1983 | 758,459 | 571,825 | 75.4% | 22,873 |
| Nashville-Davidson--Murfreesboro--Franklin, TN | 1993 | 871,937 | 519,657 | 59.6% | 20,786 |
| Las Vegas-Henderson-North Las Vegas, NV | 1997 | 935,960 | 510,164 | 54.5% | 20,407 |
| Austin-Round Rock-San Marcos, TX | 2000 | 1,001,295 | 476,172 | 47.6% | 19,047 |
| Louisville/Jefferson County, KY-IN | 1977 | 597,784 | 452,403 | 75.7% | 18,096 |
| Jacksonville, FL | 1991 | 713,833 | 440,580 | 61.7% | 17,623 |
| Memphis, TN-MS-AR | 1982 | 576,520 | 432,156 | 75.0% | 17,286 |
| Oklahoma City, OK | 1983 | 615,515 | 426,930 | 69.4% | 17,077 |
| Richmond, VA | 1984 | 561,343 | 408,270 | 72.7% | 16,331 |
| Raleigh-Cary, NC | 1999 | 601,103 | 305,242 | 50.8% | 12,210 |
Metropolitan statistical areas with 500,000 or more housing units, ordered by the modelled number of units aged 25 or more. Housing units and median year built from ACS 2023 5-year tables B25034 and B25035; the remaining columns are HyreRoof calculations. Retrieved 3 September 2026.
These 46 areas hold 73,430,441 housing units, 51.6% of the national stock, and account for 2,208,054 of the 4,272,241 modelled annual replacements at a 25-year cycle.
Who owns the roofs in the window
Owner-occupied
82,892,037 occupied units (ACS 2023 5-year, table B25036). At a 25-year cycle, 61,139,295 of them are past their first covering, 73.8% of owner-occupied stock, and 2,445,572 modelled replacements a year.
This is the universe the American Housing Survey measures, and it is where the 17-year implied interval comes from. It is also where the decision sits: an owner-occupier chooses when the roof gets replaced, and the choice is normally made once or twice in a lifetime with no prior experience of making it.
Renter-occupied and vacant
44,590,828 renter-occupied units, plus 14,850,011 vacant units, together 41.8% of all housing units.
HyreRoof analysis: no federal survey measures re-roofing on this part of the stock the way AHS measures it on owner-occupied homes, so the model runs there without a reality check. Treat every national figure on this page as better evidenced for the owner-occupied two thirds than for the rest, and treat a vacant unit as a roof that still fails but is not a customer.
What breaks this model
- Not every roof is asphalt
HUD’s inspection guide describes slate, clay tile and asbestos cement as materials that “may last the life of the structure”, and metal as lasting “50 years or more if properly painted or otherwise maintained.” Nothing in ACS identifies roofing material, so a single cycle length is applied to a mixed stock. In an area with a lot of tile or slate the model overstates; in a three-tab asphalt tract it may understate.
- The steady state is an approximation
Dividing by L assumes replacements are spread evenly across the stock. They are not: the cohorts are lumpy, and a decade with 20.5 million units in it sends a bigger pulse through the trade than a decade with 6.4 million. The model gives a long-run average, not a forecast for a particular year.
- Within-decade interpolation is an assumption
B25034 publishes decade bins. Reaching a threshold inside a decade requires assuming units are evenly distributed within it, which they are not, construction is cyclical. The effect is largest on the 20-year column, which cuts 40 per cent of the way through the 2000s cohort, a decade with a boom at the start and a collapse at the end.
- Climate is not in it
The same shingle does not last the same time in Phoenix, Buffalo and Miami. Hail frequency, UV load, freeze-thaw cycling and hurricane exposure all move the real interval, and a single national L cannot carry any of them.
- Demolition and replacement are invisible
B25034 counts units standing at the time of interview. A demolished house leaves the numerator and denominator at once, and the model has no way to see it happen.
- Multi-unit buildings are one roof and many units
The denominator is housing units, not buildings or roofs. A 200-unit block has one roof and contributes 200 units. That inflates the modelled event count in dense metros, greater New York most of all, relative to a market of detached houses.
Using this without over-claiming
- 1 Quote a range, never a point
The three columns are the finding. “Between 3,256,668 and 5,817,854 US roof replacements a year, depending on the assumed cycle length” is defensible. A single number is not.
- 2 Prefer the share column for comparing places
Share of stock aged 25 or more is a rate and comparable across areas of very different size. The absolute count is a market size, which is a different question.
- 3 Cite AHS when you need a measured number
8.3 million roofing projects and $93.5 billion over two years are counted, not modelled. If the claim can be made with those instead, make it with those.
- 4 Do not apply any of it to one house
A national cycle length says nothing about your roof. The roof life calculator and the repair or replace calculator work from material, age and observed condition instead.
- 5 Separate wear from weather
A hail-driven replacement is not the model’s replacement. If your market is storm-driven, the interval is set by claims rather than by service life, and the hail claim estimator is the more relevant tool.
- 6 Check who is allowed to do the work before you cost it
What a roofing credential proves varies by state, six structurally different regimes, set out in our licensing study.
Method
Research question. How many US housing units are old enough that their original roof covering would have been replaced at least once, and how many replacement events does that imply per year, under explicit and varied assumptions about the replacement cycle?
Sources. Housing counts from the Census Bureau’s ACS 2023 5-year table-based Summary File: acsdt5y2023-b25034.dat (year structure built), acsdt5y2023-b25035.dat (median year built), acsdt5y2023-b25036.dat (tenure by year structure built), joined to Geos20235YR.txt on GEO_ID. Measured replacement volume from the 2023 American Housing Survey home-improvement release. Service-life anchors from three HUD publications, named individually in Sources.
Geography. Summary level 010 for the United States, 040 for states and the District of Columbia and Puerto Rico, 310 for metropolitan statistical areas. Micropolitan areas are excluded. The metro cut is 500,000 or more housing units, which yields 46 areas holding 51.6% of the national stock.
HyreRoof calculation, units aged L or more. Let c be the ten B25034 cohorts and the reference year be 2023, the terminal year of the 5-year period. The cut year is 2023 − L. Every cohort entirely at or before the cut is added whole; the cohort containing the cut contributes the fraction (cut − decade start + 1) ÷ (decade length); cohorts after the cut contribute nothing; and “Built 1939 or earlier” is always included.
At L = 20 the cut is 2003, so 4/10 of the 2000–2009 cohort is included. At L = 25 the cut is 1998, so 9/10 of the 1990–1999 cohort. At L = 30 the cut is 1993, so 4/10 of the same cohort. The function that does this is exported from this page’s data module as unitsAged so the arithmetic and the published numbers cannot drift apart.
HyreRoof calculation, annual replacements. Units aged L or more, divided by L. At national level: 116,357,074 ÷ 20 = 5,817,854; 106,806,020 ÷ 25 = 4,272,241; 97,700,027 ÷ 30 = 3,256,668. The per-1,000 column is that result divided by total housing units and multiplied by 1,000.
HyreRoof calculation: the AHS-implied interval. The 2023 AHS reports 8,300,000 roofing projects over two years, 4,150,000 a year, on an owner-occupied universe. We solved for the L at which unitsAged over the B25036 owner-occupied cohorts, divided by L, equals 4,150,000. Searching L in hundredths of a year between 5 and 40 gives L = 17.1, at which the model produces 4,149,139 events a year against the 4,150,000 target. The crude interval, owner-occupied units divided by annual projects, with no age adjustment, is 20.0 years on the ACS denominator and 20.9 years on the AHS one.
HyreRoof calculation, average project cost. $93.5 billion ÷ 8,300,000 projects = $11,265 mean spend per roofing project. AHS separately publishes a median of $10,000. The mean sits above the median, as it should for a right-skewed cost distribution; both are two-year figures in nominal dollars and neither is a quote.
Reproducibility check. Summing our per-state aged-25+ figures across the fifty states and the District of Columbia gives 106,806,016 against the independently computed national figure of 106,806,020: a difference of four units, entirely from rounding each state before adding. Where a check like that fails, the model is wrong; here it does not.
What we did not do. No weighting, no composite index, no adjustment for climate, material mix, storm frequency or roof pitch, and no attempt to project forward. Each of those would improve the model and each would make it unfalsifiable, which is a worse trade than it sounds.
Limitations
- Housing age is a proxy for roof age
The load-bearing limitation, repeated here because it governs everything above. No ACS or AHS table records the installation date of a roof. This page estimates a population of replacement events; it does not observe one.
- The cycle length is an assumption, not a measurement
Every headline number moves by more than 40 per cent across the range we publish. Anyone quoting a single figure from this page without its assumption has taken the least reliable part and dropped the caveat.
- Margins of error are not propagated
The cohort estimates carry published margins of error and the aged-L+ figures are linear combinations of them, so they inherit sampling error we have not carried through. At national and state level that error is small relative to the assumption uncertainty; at metro level it is larger.
- Not every county, contrary to the brief
B25034 publishes for all 3,144 counties and we hold the file. We stopped at 46 metros of 500,000+ housing units because on small counties the cohort margins of error exceed the differences between them, and a 3,144-row table of interpolations would look far more authoritative than it is.
- AHS and ACS are different surveys
AHS 2023 is a single-year survey of 86,900,000 homeowners; ACS 2023 5-year is a pooled 2019–2023 average with 82,892,037 owner-occupied units. Comparisons between them are indicative, not exact, and the AHS home-improvement estimates come from the Table Creator and internal use file with the Census disclosure-avoidance protections applied.
- Nominal dollars, two-year window
The $93.5 billion and the $10,000 median are nominal spending over the two years to 2023 and are not adjusted for inflation. What happened to roofing prices over that period is a separate study: the roofing inflation index.
- The service-life anchors are old documents
HUD’s Residential Rehabilitation Inspection Guide is dated February 2000 and was prepared for HUD’s Office of Policy Development and Research by the National Institute of Building Sciences; the Rehab Guide volume carries a disclaimer that its conclusions are the contractor’s rather than HUD’s. They are the best federal statements we could retrieve on roof service life, and they are not current market research.
- Two retrieval routes failed and were replaced, not worked around
On 3 September 2026 the Census data API at api.census.gov returned HTTP 302 to a missing-key page for an unauthenticated request, so every ACS figure here comes from the public table-based Summary File instead: the same estimates from the same programme by a route needing no credential. On the same date the mobile edition of HUD’s Residential Rehabilitation Inspection Guide returned an empty body and the PDF edition was used in its place. The AHS home-improvement figures come from the published Census infographic rather than from the AHS Table Creator, which is behind the same keyed API.
- This is not a forecast
A steady-state average is not a projection of next year. The lumpiness of the cohorts means real annual volume oscillates around the average, and no part of this page attempts to say when the peaks fall.
Questions
How many US homes need a new roof?
How many roofs are actually replaced in the US each year?
How long does a roof last, according to the federal government?
Why does this page use three assumptions instead of one number?
What is the average interval between roof replacements in America?
Which metro area has the largest share of homes past a 25-year roof cycle?
Which state has the most homes in the replacement window?
How much do Americans spend on roofing each year?
Does a home built in 1970 need a roof now?
Why is the model’s prediction lower than what the survey measured?
Does this include commercial and rental buildings?
Why stop at 46 metros instead of covering every county?
Can I use these figures in a report or a news story?
Written and audited by
HyreRoof Research
Primary-source research, data analysis and fact checking
We are a research desk, not a sales floor. We read the statute, the licensing board’s own pages, the code section or the federal dataset ourselves, and we publish the figure with the document it came from and the date we retrieved it. Where a number cannot be traced to a primary source, we publish the shorter page and say what we could not verify. On our first study that rule removed a Minnesota exam statistic and left two states blank. Those gaps are on the page, not in a file somewhere.
- 36
- primary sources read and cited
- 16
- federal and state government domains
- 36
- citations carrying a retrieval date
- 3
- researched pages published
How this desk works
- Primary sources only. Statutes from the legislature’s own publishing system, licensing rules from the board that issues the licence, datasets from the agency that collected them. Never a directory, an aggregator or another guide.
- Three states, not two. A requirement is recorded as verified present, verified absent, or not verified. Most comparisons collapse the third into the second, which turns an unchecked cell into a factual claim.
- Retrieval dates on everything. Regulation changes. A citation without the date it was read is not a citation.
- Failures are published. When a source blocks automated retrieval we record the failure and leave the row empty, rather than filling it from a secondary summary.
- Authorship is organisational. Research is attributed to this desk, never to an invented expert. Outside commentary, where used, is attributed to named and verifiable people.
Data as of 3 September 2026. Authorship on this site is organisational: the analysis belongs to the desk rather than to a named individual, and we do not publish credentials we do not hold. Our editorial policy sets out how we source, date and correct what we publish.
Sources & retrieval dates
- US Census Bureau, ACS 2023 5-year Summary File, table B25034 , Year structure built, ten cohorts with margins of error. File acsdt5y2023-b25034.dat Retrieved 3 September 2026.
- US Census Bureau, ACS 2023 5-year Summary File, table B25036 , Tenure by year structure built. File acsdt5y2023-b25036.dat Retrieved 3 September 2026.
- US Census Bureau, ACS 2023 5-year Summary File, table B25035 , Median year structure built. File acsdt5y2023-b25035.dat Retrieved 3 September 2026.
- US Census Bureau, ACS 2023 5-year geography file , Geos20235YR.txt, summary levels and GEO_ID to name crosswalk Retrieved 3 September 2026.
- US Census Bureau, 2023 ACS Subject Definitions , “Year Structure Built” (p. 52): year built refers to when the building was first constructed, not when it was remodeled, added to, or converted Retrieved 3 September 2026.
- US Census Bureau and HUD, 2023 American Housing Survey, Home Improvements , 8.3 million roofing projects, $93.5 billion roofing expenditure, $10,000 median roofing project, $827 billion total improvement spend, 2021–2023 Retrieved 3 September 2026.
- US Census Bureau and HUD, American Housing Survey for the United States: 2023, Definitions , Appendix A: “Roofing… is meant to capture the replacement of the entire roof or at least most of it… The roof may replace or be installed over the old roofing materials” Retrieved 3 September 2026.
- US Census Bureau, American Housing Survey programme page , Survey sponsorship, scope and access to the AHS Table Creator Retrieved 3 September 2026.
- US Department of Housing and Urban Development, Office of Policy Development and Research , Residential Rehabilitation Inspection Guide, February 2000, prepared by the National Institute of Building Sciences, §2.5: asphalt shingle service life about 20 years first layer and 15 years second, the two-layer limit, wood shingles 25–30 years, metal 50+ years, slate and tile potentially the life of the structure Retrieved 3 September 2026.
- US Department of Housing and Urban Development, Rehab Guide Volume 3: Roofs , Shingles classified by warranty duration of 20, 25, 30 or 40 years with no direct relationship to performance; the Insurance Institute for Property Loss Prevention’s 17-year effective-life assumption Retrieved 3 September 2026.
- US Department of Housing and Urban Development, Durability by Design , Appendix B, Estimated Life Expectancy and Homeowner Maintenance Chart: roofing 15–30 years, condensed from the NAHB Life Expectancy Survey (1997) Retrieved 3 September 2026.
- US Census Bureau, Notes on ACS Estimate and Annotation Values , Meaning of the −333333333 annotation used where a median falls in an open-ended interval Retrieved 3 September 2026.
- US Census Bureau, ACS 2023 table shells , Line labels and universes for B25034, B25035 and B25036 Retrieved 3 September 2026.
A national model; two tools that answer for one roof
A national cycle length cannot date your covering. Both tools work from material, age and what you can actually see, and neither asks for your details.
HyreRoof does not perform roofing work. Every modelled figure here is our own arithmetic over public federal data and is labelled as such. The Census Bureau and HUD have published nothing of the kind and should not be cited for it. If a figure is wrong, or a newer ACS or AHS vintage supersedes it, tell us; our editorial policy covers corrections.