Whether permitting is slow can only be answered by counting the applications still waiting

The permitting argument in the United States is conducted almost entirely over one number, and that number is a count of permits issued. A count of authorisations is published for each state, month by month. It cannot measure delay. A count of decisions leaving a system says nothing about how long anything waited inside it, and that limitation is structural rather than a matter of insufficient detail. Finer geography does not repair it. A longer history does not repair it.

My position is that the question of whether American permitting is slow is, on the published record, unanswerable, and that the missing quantity is specific enough to name. It is the stock of applications pending, published with the age of that stock. Until an agency publishes it, the claim that permitting is the binding constraint and the claim that it is not are both unfalsifiable. Anyone asserting either with confidence is asserting something they have not checked, because the check does not exist.

Throughput is not duration

The relation between the three quantities is an identity rather than a finding. The average time an item spends in a queue equals the average number of items in that queue divided by the rate at which items arrive. Any two of those terms fix the third. Housing units authorised by permit is a departure rate and not an arrival rate, and the two run together only in a queue whose backlog is neither growing nor shrinking. Whether this one is stable is unknown for the same reason the wait is unknown, because the backlog is not published. The record therefore supplies one term of the identity on an assumption it cannot check, and a departure rate carries no information about waiting on its own.

Take the 2025 figure as it stands, an annual rate of 1431 thousand units, and suppose for the sake of the arithmetic that applications arrive at that same rate. Multiplied by a tenth it gives 143.1 thousand, and a pending stock of that size would imply an average wait of a tenth of a year. Multiplied by a half it gives 715.5 thousand, and a stock of that size would imply half a year. Both are consistent with the identical published figure. Neither of those stocks is a measurement, the supposition behind them is not one either, and the arithmetic is shown here so that nobody mistakes any of it for a measurement. The published series does not distinguish between them and never will, whatever happens to its frequency or its sample.

The count that cannot be slow

Housing units authorised by permit reads, at an annual rate, 1479 thousand units in 2020, 1734 in 2021, 1685 in 2022, 1516 in 2023, 1474 in 2024 and 1431 in 2025. From the 2021 peak to the latest reading the annual rate is 303 thousand units lower. Two accounts fit that fall without contradiction. In the first, fewer people applied, and the system issued fewer permissions because fewer were sought. In the second, the same people applied and fewer got through. The series is identical in both worlds, which is the whole difficulty.

Going down to the states does not break the tie. The state file carries new private housing units authorised for all fifty-one jurisdictions, month by month, from the Census Bureau. In the observation dated July 2026, Texas authorised 18,391 units against 16,550 in the prior month, a rise of 1,841. California authorised 8,285 against 10,013, a fall of 1,728. Two large states moved in opposite directions in the same month, and neither figure carries a single day of elapsed time. It is the same instrument. There is simply more of it.

A median over finished decisions would not settle it either

The obvious reform is to publish a median time to decision, and it would be an improvement on nothing. It would not be sufficient, and the reason is the part of this subject that gets skipped. A statistic computed over decisions issued is computed over the applications that finished. The applications that have not finished are excluded from it by construction, and those are precisely the slow ones. An agency whose worst files never close would report an excellent median for exactly as long as they stayed open.

That is why the pending stock is the load-bearing number rather than the decorative one. A backlog, reported with the age of everything in it, cannot be improved by declining to decide. It is the one permitting statistic that gets worse when an agency performs worse. In my judgement that is also why it is not published. I cannot check a motive against a data file and I am not pretending to. I am saying which of the two numbers an organisation would choose if it were choosing for itself, and the record is consistent with that choice having been made.

The four numbers

What an agency would have to publish is short enough to state in a sentence each. Applications received in the period. Decisions issued in the period, which is the only one of the four that the record in front of me carries. The elapsed time from acceptance of a complete application to a decision, given as a distribution rather than a mean, because the tail is the entire subject and an average buries it. And the count of applications still pending at the close of the period, broken out by how long each has been pending. Four series, one frequency, one published definition of the events that start and stop the clock. With all four the identity closes and the argument ends. With any three of them it does not.

The objection I take most seriously

American permitting is not one process. It is local jurisdictions, state-level regimes and a set of federal consents that apply to different projects under different statutes, and there is no single queue whose length could be reported. A national figure assembled across that variety would be an average of incommensurable things, and a reader would be worse informed for having seen it. That objection is correct about the aggregate. It is wrong about the requirement.

Nothing in the identity requires a national total. It holds wherever arrivals, departures and a stock can be defined, which is to say inside each programme taken separately. A state environmental agency reporting its own received, issued, elapsed and pending figures is reporting something internally consistent, and two programmes reporting on the same definitions are comparable even where no sum of them would mean anything. The collection machinery is not the obstacle either. The Census already publishes an authorisation count for every state every month, and the state file carries a value for all fifty-one jurisdictions with no gaps at all. What is missing is not the channel. It is the question being sent down it.

Where the record is emptiest

Housing at least has a count. Environmental permitting, which is where most of the argument actually sits, has nothing in this record. The nearest object to it is a table of facilities reporting to the Toxics Release Inventory, fifty-one jurisdictions running from 22 in the District of Columbia to 5,110 in California, and the file that carries that table states in terms that it measures industrial density and not environmental quality. It holds no permit, no application and no date. A debate about the speed of environmental review is being conducted with no environmental review statistic in evidence at all.

Spending series are sometimes offered as a substitute and they are not one. Nonresidential construction spending runs, at a seasonally adjusted annual rate, 856 billion dollars in 2020, 845 in 2021, 968 in 2022, 1193 in 2023, 1295 in 2024 and 1309 in 2025, which is 453 billion a year above where it started. Money put in place is dated to the work, not to the consent that allowed the work. A project held up for four years and then built appears in that line in the year of the pour and in no earlier year. The series records that building happened. It is silent on when permission arrived.

Reform of permitting may well be worth doing, and nothing here is a defence of delay. What I am refusing is the pretence that the case has been measured. On the record available, the sentence that permitting is slow and the sentence that permitting is not the constraint have the same evidentiary standing, which is none, and the argument between them is settled by temperament and by whoever speaks last. An agency confident that its process is quick has the cheapest possible way to demonstrate it and has not taken it. That, rather than any movement in the permit count, is the most informative fact in the file.