America’s Oldest Housing Stock, City by City

What median year structure built actually is

Table B25035 publishes one number for a place: the year that divides its housing units into two equal halves, half built before it and half after. The universe is all housing units, owned, rented, occupied and vacant. The Census Bureau notes that median age of housing is simply the survey year minus that value, which is the convention this page uses: 2023 minus the median year built.

It is computed on a fixed ten-interval distribution running from “Built 2020 or later” down to “Built 1939 or earlier”, so it is an interpolated figure rather than a raw one. Two consequences follow, and both matter more than the ranking itself.

The first is that the estimate carries a margin of error, and at 90 per cent confidence those errors are one or two years for cities of this size, small, but large enough that a one-year gap between two cities is not a difference. The second is the floor, which is the subject of its own section below.

HyreRoof analysis: the median is a middle-of-the-distribution statistic and it is silent about the tails. A city with an old core and thirty years of aggressive suburban annexation can post a median in the late 1980s while still holding tens of thousands of pre-war houses. Atlanta does exactly that. So this study publishes the median and the pre-1940 share and the pre-1980 share for every city, because the three answer different questions.

The forty oldest of the hundred largest cities

The top of that chart is the industrial Northeast and Midwest, and it is close to a clean sweep: Buffalo, St. Louis, Rochester, Cleveland, Pittsburgh, Detroit, Baltimore, Philadelphia, Cincinnati, Milwaukee, Toledo. San Francisco and Boston are the two coastal exceptions, and they are there for the same underlying reason, a housing stock that was substantially complete before 1950 and a physical envelope that has left little room to add to it since.

The bottom is the Sun Belt and it is just as consistent: Irvine (2001), Henderson (2000), Cape Coral (1999), Chandler (1997), Raleigh (1996), Las Vegas and Charlotte and Durham (1994). These are places where more than half of every home standing today was built after the first Gulf War.

HyreRoof analysis: the useful way to read the spread is not as a league table of decay. Detroit’s median of 1947 and Irvine’s of 2001 are both statements about when construction happened, and the fifty-four years between them describe two completely different roofing markets.

One is dominated by repeat replacement on small, steep, complicated roofs over dimensional lumber decking; the other by first and second replacements on large, simple roofs over plywood or OSB, much of it still under an original manufacturer warranty.

Where the median lies to you

HyreRoof calculation: the two measures agree closely: the correlation between median year built and pre-1940 share across the hundred cities is r = −0.936. That is exactly what you would expect and it is not the interesting part. The interesting part is the residual.

Atlanta ranks 82nd of 100 on median year built (1987) and 42nd on pre-1940 share (12.1 per cent). Fort Worth ranks 90th on the median and 61st on the share; Durham 93rd and 68th.

These are cities with a genuinely old core and an enormous volume of post-1990 construction around it, and the median sits in the new material because that is where the mass is. Reading Atlanta as a young-housing city because its median is 1987 loses thirty-one thousand pre-war houses.

The error runs the other way too. Fort Lauderdale ranks 41st on median year built (1972) but 87th on pre-1940 share (1.6 per cent); Anaheim 42nd and 83rd; Anchorage 65th and 91st. These places have a middle-aged stock with almost no pre-war tail at all, post-war built, and built more or less all at once.

HyreRoof analysis: for anything to do with roofs, the tail usually matters more than the middle. Pre-war construction is where you find board sheathing rather than panel decking, slate and clay tile rather than asphalt, low or absent attic ventilation, and the accumulated layers that HUD’s inspection guide warns about. A single median cannot carry any of that, which is why the table above publishes three columns instead of one.

What an old housing stock does mean for roofing

Since the honest answer to “does an old city mean old roofs” is no, it is worth being precise about what an old city does mean. Four things, all of which follow from the construction rather than from the covering’s age.

The material mix is different. HUD’s inspection guide describes slate, clay tile and asbestos cement shingles as “extremely durable and, if of high quality and properly maintained, may last the life of the structure,” and metal as lasting “50 years or more if properly painted or otherwise maintained.” A pre-war stock contains materially more of these than a 1990s stock does, and they do not sit on a twenty-year cycle at all.

HyreRoof analysis: that cuts against the intuition the ranking invites. Some of the oldest housing in the country carries the least frequently replaced roofs in the country.

The layer count is different. The same guide states that “no more than two layers of asphalt shingles should normally be in place at any one time”, and puts the service life of a second layer laid over a first at about fifteen years against about twenty for the first.

Older housing has had more opportunities to accumulate layers, and a house at the two-layer limit faces a tear-off rather than a roof-over on its next job: a different scope and a different price.

The deck is different. Pre-war roofs are commonly framed and sheathed in dimensional lumber rather than panels, and wood shingle roofs of the period were often laid on spaced sheathing. Neither is a defect. Both change what a tear-off finds underneath and what the re-cover has to be fastened to.

The regulator is different. Old stock and old cities go together, and older cities are more likely to run their own licensing, permitting and inspection regimes on top of whatever the state does, or in place of it. Which regime you are in changes what a credential proves; we have set out the six ways American states regulate roofing and why the published counts of “licensed states” disagree with each other.

Method

Source. Every housing figure was read from the Census Bureau’s ACS 2023 5-year table-based Summary File: the pipe-delimited releases acsdt5y2023-b25035.dat, acsdt5y2023-b25034.dat, acsdt5y2023-b25036.dat and acsdt5y2023-b25037.dat, joined to the geography file Geos20235YR.txt on GEO_ID. Nothing was taken from a secondary summary, a real-estate listing site or another “oldest cities” article.

Universe. Summary level 160, incorporated places, in the fifty states and the District of Columbia and Puerto Rico. Census-designated places, zonas urbanas and comunidades were excluded because they are statistical areas rather than municipal governments; Urban Honolulu CDP, Arlington CDP in Virginia, and Paradise, Spring Valley, Enterprise and Sunrise Manor CDPs in Nevada would otherwise have entered the hundred on size.

Selection. The hundred places with the most total housing units (B25034_001E), which is the only variable used to select. That threshold is 95,116 housing units. Selecting on housing units rather than population is deliberate: the question is about buildings, and a place with many small households would otherwise be under-weighted relative to the roofs it contains.

Ranking. Ascending on B25035_001E, which is the single measured variable. Where two cities share a median year the tie is broken for display by the pre-1940 share, and the table says so. No composite score is used anywhere on this page. A twenty-factor “oldest housing” index would be unfalsifiable and this desk does not build them.

HyreRoof calculation, median age. Median age of housing = 2023 minus median year structure built, which is the arithmetic the Census Bureau’s own subject definitions describe. For the three floor cities the result is written as “85+ yr” because 1938 is a sentinel and the true median is 1939 or earlier.

HyreRoof calculation, cohort shares. Pre-1940 share = B25034_011E ÷ B25034_001E. Pre-1980 share = (B25034_007E + B25034_008E + B25034_009E + B25034_010E + B25034_011E) ÷ B25034_001E. Both are ratios of estimates from the same table; their margins of error are not propagated here and the Limitations section says so.

HyreRoof calculation: the correlation. Pearson r between median year built and pre-1940 share across the hundred cities is −0.936, computed on the published values including the three at the floor. Including them biases the coefficient toward zero, because 1938 understates the true medians; the real relationship is at least this strong.

Verification of the sentinels. The 1938 floor and the 2021 ceiling were not assumed from documentation. Across all 543,542 rows of the B25035 release, 34,417 carry the value 1938 and 84 carry 2021, and exactly 34,501 rows carry the −333333333 margin-of-error annotation. The counts match to the row.