Pre-construction × Research note
Waiting for a permit: why the average approval time is wrong
Ask how long a building department takes and most people quote the average for approved applications. That shortcut ignores every application still waiting, so it always makes approvals look faster than they are. The fix is a method medicine has used since 1958, shown here on a case study of 29,123 building filings.
- 47 days
- how much the approved-only median understated the real wait, in a case study of 29,123 filings (107 vs 154 days)[1]
- 1 in 4
- filings in the same case study were still waiting for approval, invisible to a simple average[1]
- 1958
- the year Kaplan and Meier published the method that counts still-waiting cases properly[2]
Key takeaways
- Averages of approved filings leave out every filing still waiting, which are exactly the slow ones. In a case study of 29,123 filings, the median moved from 107 to 154 days once they were counted.
- The bias gets worse for recent filings: for 2026 filings the naive median is 74 days against 182.
- Survival analysis (Kaplan–Meier, Cox, survival forests) uses “still waiting after 400 days” as information instead of throwing it away.
- Check calibration, not just ranking. A naive model ranked filings almost as well but promised a 73% chance of approval within 180 days when 59% happened.
- Permit time is not a side issue: Federal Reserve research finds construction productivity fell most in places with long permit times.
01Why approval time is a cost line
Until plans are approved, the construction start, the financing draw and the holding costs are all guesses. Every month of waiting is a month of interest, overheads and escalation, and regulation and approvals together are a material share of what a building costs: one home builders’ association estimated regulation at 23.8% of the average price of a new home in its market[3].
The variation between places is large, even between neighbouring cities. Research on housing approvals found that more than 80% of proposed multifamily developments in the jurisdictions studied needed a discretionary approval, and that the median time for similar projects ranged from about six months in one city to more than 25 months in the next[4].
And it shows up in productivity statistics. Federal Reserve economists found that single-family construction productivity declined most in areas with tighter supply constraints, especially locations with long permit times[5].
02The shortcut that everyone uses
The usual way to answer “how long does approval take?” is to take filings that were approved and average the days from filing to approval. It feels reasonable. It is biased, always in the same direction.
Statisticians call the still-waiting filings right-censored: we know the wait is at least 300 days, not what it will be. Dropping them makes the process look faster than it is. Giving them a made-up end date is worse. Kaplan and Meier solved this in 1958 with an estimator that uses each censored case for exactly as long as it was observed[2].
A filing still waiting after 400 days is information, not missing data.
03Case study: NYC Permit Approval Forecast
From my portfolio · project 02 of 15
Real public data (NYC Open Data)- Problem
- Approval dates for major building filings are guesses, and averages hide the slow cases.
- Decision supported
- When to plan the start, how much float to hold, how long to budget holding costs.
- Data
- 29,123 NYC DOB NOW filings, 2021–2026; a quarter still waiting.
- Method
- Kaplan–Meier, Cox, random survival forest and XGBoost AFT, tested on later filings.
- Result
- Approved-only median understates the wait by 47 days; probabilities within ~3.5 points of outcomes.
- Limits
- Ranking is modest (C-index ~0.69); drawing quality and examiner are not in the data.
04What 29,123 real filings show
I took New York City’s public DOB NOW filing data[6] for four major job types (New Building, two kinds of major alteration, and full demolition): 29,123 filings from 2021 to 2026, one row per job, with only the information known on the filing day[1]. A quarter were still open, and even among 2021 filings about one in eight had never been approved.
Figure
Median days from filing to plan approval, Kaplan–Meier (all filings)
Figure
Approved-only median vs survival median, all four job types
Other findings from the same data: Queens was fastest (median 130 days, against about 167–171 elsewhere), and after plan approval the first permit followed a median of about two months later. Approval is not the permit, and the permit is not the start on site.
05From a median to a forecast
A median by job type is a better rule of thumb. A forecast for one filing needs a model that learns from filing-day features and handles censoring properly. I compared five approaches on filings from July 2024 to June 2025 that the models never saw: Kaplan–Meier by job type, a naive regression trained only on approved filings, Cox proportional hazards, a random survival forest and an XGBoost accelerated-failure-time model[1].
The instructive result is the naive model. It ranked filings almost as well as the survival models, so a ranking metric alone would have hidden the problem. But its probabilities were over-optimistic: it predicted a 73% chance of approval within 180 days when 59% actually happened, because it never saw the filings that stall. The survival model’s probabilities landed within about 3.5 points of what happened, on average across ten groups.
06How strong is the evidence?
Not every finding in this note rests on the same kind of evidence. This is how I would weigh each one before acting on it.
| Finding | Evidence | Strength | Main caveat |
|---|---|---|---|
| Approved-only averages understate the wait | Statistical property of censored data; 29,123 NYC filings | Strong | Size of the gap varies by place and year |
| Survival models give calibrated probabilities | Out-of-time test on 5,708 NYC filings | Moderate | One city, one department, one period |
| Long permit times go with weaker productivity | Federal Reserve analysis of metro areas | Moderate | Association, not proof of cause |
| Regulation is about a quarter of a new home’s price | NAHB builder survey, 2021 | Indicative | Industry-association estimate |
07How to use this in planning
- Never average only the finished cases. For approvals, RFIs, change orders or submittals, count the open items as “at least this long”.
- Quote a probability, not a date. “60% chance of approval within six months” supports float and financing decisions; a single date hides the risk.
- Plan to the P80, not the median, when holding costs are high, and review the forecast when objections arrive.
- Separate the milestones. Plan approval, permit issue and site start are different events with different delays.
- Check calibration. Of the jobs your model gave a 70% chance, did about 70% happen? Ranking metrics alone won’t tell you.
NotesSources
- Iwale, A. (2026). NYC Permit Approval Forecast: survival analysis of DOB NOW filings, 2021–2026. GitHub.
- Kaplan, E. L. and Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282), 457–481.
- Emrath, P. (2021). Government regulation in the price of a new home: 2021. National Association of Home Builders.
- Terner Center for Housing Innovation, UC Berkeley. The Cost of Building Housing research series (framing paper).
- Garcia, D. and Molloy, R. (2023, revised 2025). Reexamining lackluster productivity growth in construction. Finance and Economics Discussion Series 2023-052, Federal Reserve Board.
- NYC Open Data. DOB NOW: Build – Job Application Filings (dataset w9ak-ipjd).
Figures are quoted from the sources above as published; where a source reports a range or a survey estimate, it is described that way. Results from my own projects say whether they use real public data or synthetic data.