1. What the study measures

The study has four parts. Each is published on its own, and none is folded into another’s numbers.

The Roof Age Index is an estimate, not a measurement. No public record tracks when every roof was replaced. We estimate roof age from how often homeowners report replacing roofs and from the age of the housing stock. Every roof-age figure is labeled “estimated” and published with its estimated range.

2. Headline results

Estimated roof age

* Western North Carolina’s range is extended to account for roofs replaced after Hurricane Helene (section 4.7).

Storm and climate

3. Where the study covers

Markets

Western North Carolina follows the 23-county definition used by Western Carolina University’s Regional Outlook Report. The Charlotte region is the Charlotte-Concord-Gastonia metropolitan area. Knoxville & East Tennessee is the Knoxville metropolitan area plus Sevier County. Metropolitan areas follow the U.S. Office of Management and Budget’s July 2023 delineations.

Which homes

The study covers owner-occupied, single-family detached homes. It excludes renter-occupied homes, apartments and other multifamily buildings, townhomes, manufactured homes and commercial buildings. The replacement rates used in the model are measured on the same group of homes.

ZIP codes

ZIP-level figures use Census ZIP Code Tabulation Areas (ZCTAs). These closely approximate, but don’t exactly match, U.S. Postal Service ZIP codes. Each ZCTA is assigned to the county holding most of its land area; about a quarter of ZCTAs extend into a second county. We estimated 338 ZCTAs and publish 313. ZIPs with fewer than 50 qualifying homes are withheld, following Census practice.

4. The Roof Age Index

4.1 Definitions

Why 20 years. The National Association of Home Builders / Bank of America Home Equity Study of Life Expectancy of Home Components (February 2007) lists asphalt roofing at 20 years. The study notes that roof life “depends on local weather conditions, proper building and design, material quality, and adequate maintenance.” Results at 18, 22, 25 and 28 years are in section 4.6.

4.2 The approach in one paragraph

Every year, some share of homes get a new roof. If we know that annual replacement rate, and we know how old each home is, we can estimate how likely its current roof is to be 20 years old or older. A 60-year-old house has almost certainly had its roof replaced at least once, but it may not have been replaced in the last 20 years. A 15-year-old house cannot have a 20-year-old roof. We take the replacement rate from the Census Bureau’s American Housing Survey and the age of the housing stock from the American Community Survey.

4.3 Step 1: how often roofs are replaced

Source. The American Housing Survey (AHS), national public use files for 2017, 2019, 2021 and 2023. The AHS asks homeowners about improvement projects completed in the last two years. We counted a home as having a roof replacement when it reported a completed project coded “added or replaced roofing.” In 2017–2021 the survey worded this as “added or replaced roof over entire home”; the 2023 survey shortened it to “added or replaced roofing.”

Group. Owner-occupied, one-unit detached homes, by Census division. North Carolina and South Carolina are in the South Atlantic division; Tennessee is in the East South Central division.

Estimator. The survey records completion as “this year,” “last year” or “two years ago,” not as a date. Our point estimate uses only the middle category, the calendar year before each survey. That year falls fully inside the two-year window for every respondent, whenever they were interviewed. Our upper estimate uses the whole two-year window divided by two. Because people tend to remember older projects as more recent than they were, which pulls them into the two-year window, the upper estimate runs high.

Pooling. Each division’s four survey years are combined, with more precise years weighted more heavily.

Annual roof replacement rate, 2017–2023 surveys combined:

The range runs from the lower 95% confidence bound of the point estimate to the upper estimate. Sample counts are unweighted survey records summed across the four survey years.

By survey year (annual rate, % ± standard error):

Standard errors use the AHS’s 160 replicate weights and the Census Bureau’s published formula: variance = (4/160) × the sum of squared differences between each replicate estimate and the full-sample estimate. The AHS national sample follows the same homes across survey years, so the combined standard errors are reported conservatively: each division uses the standard error of its most precise single survey year.

No trend is claimed. Rates in the 2023 survey were higher than earlier years, but the 2023 question wording also changed. Without the 2023 survey, no upward trend is statistically significant in any division. The rate is treated as stable, and year-to-year comparisons are made only between surveys that asked the same question.

4.4 Step 2: the age of the housing stock

Source. American Community Survey (ACS) 2020–2024 five-year estimates, for every county and ZCTA in the study:

• Table B25036 (tenure by year structure built): owner-occupied homes by decade built
• Table B25032 (tenure by units in structure): the share of owner-occupied homes that are single-family detached

The ACS reports year built in ranges (“built 2000 to 2009”), not exact years. We assume homes are spread evenly across the years within each range. This handles the 20-year cutoff better than assigning every home the middle year of its range. The open-ended “1939 or earlier” range is treated as 1900–1939. Home ages are calculated as of 2026.

4.5 Step 3: from replacement rate to roof age

Assume a home faces the same chance of a roof replacement every year, the replacement rate h. For a home that is A years old, the chance that its roof is at least k years old is:

P(roof age ≥ k) = (1 − h)^k     if the home is at least k years old

                = 0             if the home is younger than k

The formula works because a roof is k or more years old only if it wasn’t replaced in any of the last k years. A county’s estimated share of homes with a 20+ year roof is therefore:

(1 − h)^20 × (share of owner-occupied single-family detached homes built at least 20 years ago)

Markets and counties are combined by weighting each by its number of homes.

Cost benchmark. The median cost of a completed roof job reported in the 2023 AHS, for the same group of homes: $11,000 in the South Atlantic division (Western North Carolina and Charlotte) and $8,900 in the East South Central division (Knoxville). These medians include partial and do-it-yourself jobs, so they understate a full professional replacement. The estimated replacement values are conservative for that reason.

4.6 Uncertainty and sensitivity

The estimated range combines two sources of uncertainty:

1. The replacement rate, run at the low and high ends of its range (section 4.3).
2. The shape of roof aging. The formula in 4.5 assumes a roof is as likely to be replaced in its fifth year as its fortieth. Real roofs wear out. So we also ran a wear-out model: a renewal model with a Weibull lifetime of shape 3, set so the long-run replacement rate matches the survey rate exactly.

The published range is the lowest and highest result across these runs.

Why we lead with the 20-year figure. The share of homes with a 20+ year roof moves about 6–10 points between the two aging models. The share with a 25+ year roof nearly halves, and a 40+ year figure is not credible under the simpler model: it implies about one roof in ten is over forty years old. So the 20-year figure is the headline. The 25-year figure appears only with its range. No 40-year figure is published.

Estimated share of homes with a roof at or beyond each age (point replacement rate):

Estimated share with a 20+ year roof under each assumption:

Estimated median roof age: Knoxville & East Tennessee 13–19 years; Western North Carolina 13–17 years; Charlotte region 12–15 years.

4.7 Hurricane Helene (Western North Carolina)

The replacement rates come from surveys conducted through 2023. Hurricane Helene struck in September 2024, so the model cannot see roofs replaced because of the storm. Any such replacement lowers the true share of old roofs.

No public source counts roofs replaced after Helene. We set an upper limit on the effect from two sources:

• FEMA individual assistance registrations (disasters DR-4827 in North Carolina and DR-4829 in South Carolina): FEMA-verified structural damage to owner-occupied houses, excluding flood damage.
• The North Carolina Department of Insurance Helene data call (claims as of September 12, 2025): 58,165 homeowners claims with a payment statewide.

We distributed the insurance claims across counties in proportion to FEMA’s non-flood damage reports, and counted every paid claim as a roof replacement. That deliberately overstates the effect. By that upper limit, Western North Carolina’s share of old roofs could be as much as 6.4 points lower market-wide, and up to about 12 points lower in the hardest-hit counties.

How it is applied. Headline estimates are unchanged, but the low end of every Western North Carolina estimated range is lowered by that county’s or ZIP’s upper-limit effect. This is why Western North Carolina’s range is 23–43%. The Charlotte region’s effect is under half a point, so its figures are unchanged. None of the Knoxville-area counties were designated for FEMA individual assistance.

5. Hail exposure

What is measured

Radar hail days a year: the average number of days a year that NOAA’s NEXRAD weather radar detected hail 1 inch or larger over a typical location in the county, 2000–2025. We also report the same measure for hail 1.75 inches (golf ball size) or larger.

Method

1. Source: NOAA’s Severe Weather Data Inventory, NEXRAD Level-III hail detections, 2000–2025 (313,241 detections in the study area).
2. Size: detections with an estimated maximum hail size of 1.00 inch or more. That is the National Weather Service’s severe hail threshold since January 5, 2010, applied to every year so the whole period uses one definition.
3. Days: storm days run noon to noon (Universal Time), so an evening storm isn’t split across two dates.
4. Location: the study area is divided into a grid of cells about 7 × 6 miles. A cell has a hail day if any qualifying detection falls inside it that day, so several radars scanning the same storm count once. A county’s figure is the average across the grid cells within it, so large counties don’t score higher just for being large.
5. Index: a county or market’s value ÷ the average across all three markets, weighted by area, × 100.

Why radar instead of storm reports

NOAA’s Storm Events Database records hail reports: what someone saw and reported. Across the 44 counties, reports track population closely. Hail reports per square mile rise with housing density (correlation +0.75), while reports per 1,000 homes fall with it (−0.76). Report-based figures largely measure how many people are around to see hail. Radar detects hail whether or not anyone is there.

Radar-based county values are stable: the full 2000–2025 figures and the 2016–2025 figures correlate at 0.94. Radar and storm reports rank the three markets in the same order. Storm reports are cited only as county records (“NOAA has logged N severe hail reports in this county”), never to rank.

6. Roof Stress Index

What is measured

Average across the three markets: 38 days at 90°F or hotter and 76 freeze-thaw days. Average sunlight differs by less than 3% across the three markets.

This is a descriptive comparison of climate stress. It is not an estimate of roof life and is not used in the Roof Age Index.

Method

• Source: NOAA National Centers for Environmental Information, U.S. Climate Normals 1991–2020, the official 30-year climate baseline for U.S. weather stations. 84 stations with the needed records fall inside the study’s counties.
• Days at 90°F or hotter: the normal number of days a year with a high of 90°F or more.
• Freeze-thaw days: days when the low falls to 32°F or below and the high rises above 32°F. This equals the normal number of days with a low of 32°F or below, minus the normal number of days with a high of 32°F or below.
• County figures: the median across the county’s own stations within 200 meters of the elevation where most residents live (the Census 2020 county population center). A county with no station in that range uses the nearest comparable station, weighing both distance and elevation. Mountain microclimates vary sharply: stations at similar elevations in one county can differ by 40 freeze-thaw days a year. Taking the median keeps any one sheltered or urban station from setting a county’s figure. 23 counties rest on a single station, and four use a station in a neighboring county.
• Market figures: averaged across counties, weighted by number of homes.

Why station records instead of modeled weather data

We also retrieved 30 years of daily modeled weather data (Open-Meteo, based on global reanalysis) for every county. It ranks places correctly but undercounts extremes. At three NOAA airport stations it showed 35, 20 and 3 days a year of 90°F or hotter, against 49, 37 and 9 in NOAA’s official records (Charlotte, Knoxville and Asheville). Published figures therefore use station records; the modeled data is used only for sunlight.

7. Composite Risk Index

What it shows

The Composite Risk Index shows where aging roofs and frequent hail coincide. It ranks the study’s 44 counties against each other. It is relative to these counties and is not a forecast of damage. Heat and freeze-thaw are shown alongside but are not part of the ranking: they run in opposite directions across these counties, so “more climate stress” has no single meaning.

Method

1. Two measures. Each county’s estimated share of homes with a 20+ year roof, and its radar hail days a year.
2. Thirds. The 44 counties are split into thirds on each measure: 1 point for the lowest third, 3 for the highest. Cutoffs in the base case: roof age 37.1% and 39.5%; hail 0.89 and 1.59 days a year.
3. Tiers. Points are added: 6 Highest, 5 Elevated, 4 Moderate, 2–3 Lower. No weights are applied.
4. Tested under 12 sets of assumptions. County roof-age estimates cluster in a narrow band (27–42%), so a county’s tier can depend on assumptions. Every county was re-tiered under 12 sets: the base case; eight combinations of low, point and high replacement rates for the two Census divisions; the wear-out aging model; the Hurricane Helene adjustment; and hail from 2016–2025 only. The thirds are recalculated each time.
5. Published tier. The tier each county lands in most often, with the number of assumption sets that agree. When 6 or fewer agree, or two tiers tie, the county is published with no consistent tier.

8. How figures are rounded and labeled

• “Estimated” appears on every roof-age figure, table header and chart label.
• Estimated ranges are shown alongside headline estimates.
• Percentages are rounded to whole numbers. Estimated home counts and dollar values are rounded to two significant figures (e.g. 110,000 homes, $1.2 billion). Counts of all owner-occupied homes, which come directly from the Census, are rounded to three. Halves round up.
• Exact values appear in the downloadable data.
• Withheld geographies: ZIPs with fewer than 50 qualifying homes.

9. Limitations

1. Modeled, not measured. Roof age is estimated from survey-based replacement rates and the age of homes, not from records of individual roofs.
2. Regional replacement rates. Each Census division’s rate is applied to every county and ZIP in it. Local differences in replacement behavior aren’t captured.
3. “Added or replaced.” The survey counts roofing added to a new addition as well as replacements, so the replacement rate may run slightly high.
4. Recent buyers. Homeowners who moved in during the survey window report only work done since they moved in, so some replacements by previous owners are missed. This partly offsets limitation 3.
5. Aging model. The simpler aging model overstates very old roofs. That is why 20 years is the headline threshold and the range includes a wear-out model.
6. Housing data vintage. The ACS 2020–2024 estimates predate homes built since 2024. Newer homes would slightly lower the share of old roofs.
7. Year built is reported in ranges and is spread evenly within each range.
8. Hurricane Helene. The storm’s effect is bounded, not measured, and the upper limit deliberately overstates it (section 4.7).
9. ZIP codes. ZCTAs approximate ZIP codes, and about a quarter extend into a second county.
10. Radar hail is an estimate of hail size inside the storm, not a ground report. Hail can melt before reaching the ground. The Blue Ridge partly blocks radar beams, so the highest mountain counties may be undercounted. Year-to-year hail totals are affected by radar upgrades and are not compared.
11. Climate station records represent specific locations. Single-station counties are less certain, and the normals cover 1991–2020.
12. The Composite Risk Index is relative to the 44 counties studied. A “Lower” tier means lower than the other counties here, not low risk.
13. No damage forecast. Nothing in this study estimates insurance claims, roof failures or how long a particular roof will last.

10. Data sources

All data were retrieved September 14, 2026.

License

The data, tables and charts in the Roof Age Index are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). You may share and adapt them for any purpose, including commercially, as long as you credit Benton Roofing, Roof Age Index 2026, link to https://bentonroofing.com/roof-age-index/, and indicate any changes you made. The Benton Roofing name and logo are not covered by this license.

Suggested citation

Benton Roofing, Roof Age Index 2026, https://bentonroofing.com/roof-age-index/.

Corrections

We will correct errors promptly and note each correction on this page with its date. To report an error or ask about the method, contact contact@bentonroofinginc.com.

Updates

The study is designed to be updated annually as new Census and NOAA data are released. Changes in survey wording or methods between editions will be noted here. Year-to-year changes will be reported only where the underlying data are comparable.