The State-by-State Home Construction Table Doesn't Exist — Here's What Does

A source-anchored look at how long it takes to build a house in America, and why the map you want can't be drawn yet.

Data year 2025, with a 2019 pre-pandemic baseline. All Census figures are means; Census publishes no medians for construction duration.

The finding that shapes everything else

There is no measured, state-level construction-duration statistic in the United States. The Census Bureau's Survey of Construction is the only systematic government measure of how long it takes to build a house, and its methodology page says so directly: "Statistics from the SOC are tabulated only for the United States and four Census Regions. The SOC does not have a large enough sample size to make state or local area estimates."

The sample is roughly 900 permit-issuing places drawn from 169 sampling units, out of a universe of about 19,900. No custom tabulation or records request can extract precision that isn't in the sample.

So the accompanying dataset does the only honest thing: it anchors every state to its Census region, labels that figure MODELED, and applies no state-level adjustment to construction time, because no published evidence supports one. Within a region it shows zero variation in build time. Zero is what the evidence supports.

Two related corrections. Census publishes duration for the four regions, not the nine divisions — circulating division figures are microdata tabulations, and NAHB's version measures authorization-to-completion, a different quantity. And Census publishes means only; every median below is either from a non-Census source or interpolated here from Census's five-bucket distribution.

What the measured anchor actually says

Single-family, mean months, 2025 (2019 in parentheses):

The first two columns are published by Census. The third is their sum, computed here — Census publishes no permit-to-completion figure, and the two series are tabulated on different universes (units authorized versus units completed), so treat it as derived.

Multifamily runs roughly twice as long: 16.5 months nationally in 2025, from 15.4 in 2019, peaking at 17.1 in 2023.

The mean overstates the typical house. Interpolating Census's own distribution, the median single-family build is 5.9 months against a 7.4-month mean — a long right tail pulls the average up. For the permit-to-start gap the median is 1 month in every region, against means of 0.9 to 1.5.

The pandemic has mostly, but not entirely, unwound. Construction time rose 1.6 months from 2019 to the 2023 peak and has given back 1.2 of it. The permit-to-start gap rose from 1.1 to 1.5 months and has given back 0.1. Meanwhile single-family units authorized but not started stood at 144,000 in June 2026 against 88,000 in June 2019 — up 64% (Census New Residential Construction, Table 2a; a seasonally adjusted end-of-period stock, not an annual rate). The thing that hasn't normalized is the gap between getting permission and breaking ground.

1. Fastest and slowest — and why the ranking is a trick

Here is the requested top and bottom ten, and here is why you should distrust it.

Sorted on permit-to-completion, the ten slowest are the nine Northeast states (Connecticut, Maine, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Vermont) at 13.0 months, plus any one of the thirteen Western states at 9.6. The ten fastest are any ten of the seventeen Southern jurisdictions, all at 8.1 months.

Every state in that list has its region's number. A "ranking of 51 states" is a ranking of four numbers. Sorting it produces a table that looks precise and carries no state-level information at all.

What the drivers look like when you can measure them:

• Product mix is real and measurable — but it explains less than you'd think. Custom homes (contractor-built plus owner-built) are 33.8% of Northeast starts versus 14.9% in the South. Reweighting the Northeast to the South's mix moves it from 11.3 to 10.5 months — about 0.8 months, or roughly 15% of the 5.2-month gap between them (assumptions in the appendix). The remaining 85% is that the Northeast is slower at every product type: 9.9 months for a spec house against 5.5 in the South.
• Size is a genuine confound and is unavailable by state. No source publishes new-home square footage by state; the ACS never collects it at all. At region level Northeast homes are larger — 2,459 sq ft median against 1,969 in the Midwest — so part of the Northeast penalty is product, not process.
• Climate spans a range the data cannot see. Heating degree days run from 0 in Hawaii and 552 in Florida to 11,164 in Alaska — yet Alaska, Hawaii and California share one anchor. Nobody publishes a months-per-degree-day elasticity, and fitting one across four regional observations would move numbers without informing them. Labor tightness varies as widely — construction is 8.6% of nonfarm employment in Idaho, 3.7% in New Jersey — with no published link to duration either.

2. Permitting versus construction: where the real variation lives

This is where states genuinely differ, and where the data is thinnest.

The Wharton index's Approval Delay measure — reported in months — runs 3.7 months in the least-regulated quartile of municipalities, 5.0 on average, and 8.4 in the most-regulated, with a tail of communities reporting 18 to 24 months. Set that beside a national construction time of 7.4 months: in the most regulated quartile, getting permission takes longer than building the house.

The sharpest case is San Francisco, where California's own housing agency found an average of 523 days to entitle a project and a further 605 days to issue a building permit — about 37 months of approvals against 8.3 months of construction. Soltas and Gruber (2026) find Los Angeles County takes 4.2 years from permit submission to occupancy for a standardized 30-unit building, roughly twice Raleigh or Fort Worth, with permitting about 40% of the total.

But only about a third of states publish anything. Of 51 jurisdictions:

• 14 have a published permit duration — Arizona, California, Hawaii, Idaho, Illinois, Indiana, Michigan, New York, Oklahoma, Tennessee, Texas, Virginia, Washington, Wisconsin — and several of those are stale (Knoxville's is from 2004, Indiana's from 2014), commercial-only, or state-agency rather than local residential. Two more, Colorado and DC, publish permit-timing information that isn't a duration at all: a queue-position date and a percent-meeting-threshold compliance rate.
• 17 have a statewide building-permit clock — though three are conditional: Arizona sets no day count and merely requires each municipality to adopt its own, Georgia's applies only on the private-provider track, and Montana's applies only when a single-family checklist is attached. A further 20 states have a land-use clock only, 12 have neither, and 2 could not be established from retrievable statute text.
• Exactly one state requires anyone to report whether the clock is met. Washington's RCW 36.70B.080, as amended in 2023, compels covered jurisdictions to report their average days from submittal to decision. Everywhere else, the deadline exists and the compliance data does not.

Two cautions. Florida's much-cited permit-posting requirement, read against the statutes, requires posting a policy and per-application status plus a fee-and-volume report — no time metric. And Arizona scores high on Wharton yet Phoenix publishes a measured 30-day residential plan review: the index captures entitlement stringency, not plan-review throughput. States are not consistently bad at both.

3. The owner-built gap, and why it isn't about size

Nationally in 2025, single-family start-to-completion by purpose:

An owner-built house takes 2.1× as long as a spec house — and it is the smallest of the three. That kills the obvious explanation. Contractor-built homes are slower partly because they're bigger; owner-built homes are slower despite being smaller, which points at financing, part-time labor, and self-management rather than at the product.

The gap widens westward. In the West, owner-built runs 15.6 months against 6.4 for spec — 2.4× — at a median 1,686 square feet versus 2,224. A house a quarter smaller takes nearly two and a half times as long. In the South the same comparison is 10.8 versus 5.5 months.

At state level this split is carried down from the region, and its practical meaning is a mix effect: a New England state, where a third of starts are custom, has a genuinely slower average product than a Texas suburb even if every builder worked at identical speed.

4. How much of the spread is real?

Blunt answer: for construction time and the permit-to-start gap, essentially none of it is real. It is regional assignment. Two states in the same region are identical by construction; two in different regions differ by the difference between four published numbers.

For permitting the spread is real — but it is a jurisdiction-level spread, and it is large within states. San Francisco against Oakland is 23 months against 7 for the same CEQA mechanism. Los Angeles ADUs take 260 days inside the Coastal Zone and 147 outside. Seattle's Large Multifamily runs 374 days against Middle Housing's 117. A state average would conceal more than it revealed.

And the measured figures are not comparable to each other. Phoenix reports calendar days, Boise business days. Tulsa, Boise and Phoenix measure the first review cycle; Chicago, Fairfax and Honolulu measure elapsed time end to end — a tenfold difference in what's counted. Chicago's "new construction" mixes residential and commercial. The City of Fairfax — an independent Virginia city, not Fairfax County — issued eight new residential permits in its window. So the dataset refuses to compute a total elapsed time wherever the permit figure is first-review-only or scoped to a different product: only 32 of 255 rows carry a P3 number.

Which is why the commercial rankings that already exist are worthless. The most complete-looking one ranks 17 states and concedes in its own methodology that its figures are cities' published guidelines — advertised targets, not observed durations. No credible fifty-state ranking of build time or permit time exists anywhere. That vacuum is the finding.

Appendix: methodology

Anchor. Census SOC files were downloaded and parsed directly rather than read from a summary — a summarizing pass mis-read the built-for-sale column as the total, which is how the 2022 US figure gets reported as 7.6 months when the correct total is 8.3.

State construction figures. Region mean, carried down, no adjustment. An adjustment requires a published quantitative link between a state covariate and construction duration. Permit turnaround and WRLURI both measure approvals, not building. Climate has no published elasticity. Size is unavailable by state. Rather than manufacture one, the dataset states the anchor and its limits.

Mix decomposition. Two assumptions underlie the 15% figure in section 1: the national 12.6 : 5.2 contractor-to-owner split is applied within each region's custom share, and 2024 mix weights are applied to 2025 durations. Both the reconstructed base (11.29 for the Northeast) and the gap (5.19 months) are computed on the same reconstructed footing, so the share is not inflated by mixing reconstructed and published bases.

Medians. From Census's five-bucket distribution, with class boundaries at 0.5/3.5/6.5/9.5/12.5, median = L + ((0.5N − F)/f) × w. Worked: US single-family 2025 is 17/42/20/9/12, cumulative 17/59/79/88/100, median class 4–6 months, so 3.5 + ((50 − 17)/42) × 3 = 5.86 months. Where the median falls in the open-ended 13+ class — as it does for multifamily everywhere — it is reported as not estimable rather than extrapolated.

Permitting. Never modeled. Measured where published, with scope, statistic type, day convention and period recorded. Targets and statutory ceilings are recorded as such and never converted into duration estimates. Otherwise: no published data.

What would replace the models with measurements.

1. Not a Census custom tabulation. The custom-tabulation program doesn't cover the construction surveys, and the sample can't meet publication standards at state level regardless. The SOC public-use microdata carries authorization, start and completion dates but its only geographic variable is division; the metro variable was discontinued in 2017.
2. The Federal Statistical Research Data Center gives qualified researchers internal microdata with finer geography via the Standard Application Process — but sample size, not disclosure policy, is the binding constraint.
3. The realistic path is administrative permit data, and two states already compel it. California's Housing Element Annual Progress Report requires every city and county to report per-project application dates (Table A) and entitlement, permit and occupancy dates (Table A2) under Gov. Code § 65400; joining them yields per-jurisdiction durations. Washington's SSB 5290 requires covered jurisdictions to report average submittal-to-decision days to Commerce. Extending an APR-style per-application date requirement to the other 48 states is the single reform that would make this study's modeled layer unnecessary.
4. Municipal open data replicates this from below, today. Chicago publishes a per-permit processing-time field — the medians here were computed from it directly. Austin, Dallas, Houston, Kansas City, Louisville, Nashville, Minneapolis, New Orleans and Buffalo publish comparable datasets. A sweep of the largest 200 permit-issuing jurisdictions, normalized on application-to-issuance in calendar days for new single-family, would produce the first defensible cross-state comparison of permitting.
5. Targeted requests would close the highest-value gaps blocked only by dashboard rendering: Raleigh NC and Montgomery County MD (turnaround dashboards already computed, not machine-readable), Denver CPD, Portland OR (median and mean split into staff time versus applicant time — the best-designed metric found anywhere in this study), and Washington Commerce's consolidated statutory reports.

Conflicts presented rather than resolved. Census's current vintage puts 2024 single-family start-to-completion at 7.7 months; NAHB, analyzing the same survey in September 2025, published 7.6 — a revision between vintages. NAHB's headline "9.1 months to build a house" measures authorization-to-completion and must not be spliced into Census's start-to-completion series. Gyourko & Molloy restate the 2005 Wharton approval-delay figure as 10.2 months where the source table reads 10.3. The widely cited "6.5 months average permit delay in Washington State" is not an independent measurement — it is derived from the 2018 Wharton survey, so citing both double-counts one source.

Reliability. Census RSEs on the 2025 distribution run 3–13% for single-family totals and up to 52% for multifamily cells; region-by-purpose cells are noisier still, with some Northeast owner-built RSEs reaching 80%. RSEs capture sampling error only. The Wharton state table is from the 2005 wave — the 2018 wave publishes no state table — and several states rest on tiny samples: Hawaii on one municipality, Delaware five, Montana six.

Sources: US Census Bureau Survey of Construction and Building Permits Survey; California HCD; Washington Dept. of Commerce; NAHB; Gyourko, Saiz & Summers (2008); Gyourko, Hartley & Krimmel (2021); Soltas & Gruber (2026); Terner Center; Pew Charitable Trusts; UC Berkeley CLEE; municipal open data portals and building-department publications. Full citations with URLs are in the Sources tab of the accompanying workbook.