Why we find 2.8x more PE-owned clinics than the published estimate

First, what the two counts agree on

Both find private equity operating in exactly 42 states, reached from entirely different data. The disagreement is about depth, not footprint: the two methods find the same map and fill it in to very different densities. In the five states the letter reports individually, Fundprint is higher in every one, but by wildly different factors, which is the first sign that the gap is not a uniform scaling error.

The letter reports only its five largest states, so a full per-state reconciliation is not possible from the published text. Fundprint's column is filtered to private equity alone. The site's own state map counts every institutional financial owner, including a pension fund, a family office and two search funds, and therefore reads higher. Comparing that column against a figure labelled “PE” would manufacture a disagreement out of our own broader scope.

The two methods converge to within 6 sites

Fundprint's clinics can be restricted to the ones the federal provider registry can see: sites whose building appears in the monthly NPPES bulk file. That subset is 580 PE sites. The published estimate, built from a commercial deal database and press releases, is 574.

Two censuses sharing no data source, no unit definition and no time period land 6 sites apart nationally. They are not the same 574 sites, and the overlap cannot be checked, because the letter's site list is not published. But the convergence does one useful thing: it bounds how much of the 2.8x can be method noise. Both approaches, run against records that companies file with somebody else, arrive at roughly the same number.

The remaining 1,041 clinics come from a source neither method uses: the operator's own public list of its centers. Not a leak, not a purchase, not an inference. A chain that runs 400 centers publishes all 400 on its website, because it wants families to find them. It does not separately register all 400 with the federal registry, and the transaction that created it recorded whatever it owned that day.

A second estimate, counted a different way, agrees with us

One comparison against one estimate is the weakest form of this argument. It cannot be told apart from a pipeline that simply inflates. So here is a second published estimate, and it counts in a different unit: the PESP appendix, April 2026 lists a facility count per platform, the operating company a sponsor buys, rather than per deal. Unlike an aggregate, that can be diffed line by line.

Across the 16 platforms both cover, PESP counts 1,521 facilities and Fundprint publishes 1,640 clinics, a ratio of 1.08x. And the differences run both ways: Fundprint is higher on 10 platforms, lower on 4, and exact on 2 (Acorn Health and Centria Healthcare), reached from opposite directions. A dataset that only ever revised counts upward would deserve no trust at all, and these are the receipts that this one does not.

That 1.08x is a precision check, not a national total, and the two must not be confused. It asks a narrow question: counting the same 16 companies, do the two sources land in the same place? They do. It says nothing about the national gap, which is still 2.8x. Summed across all 32 in-scope platforms, PESP's own facility counts come to 2,047, against the published estimate's 574. So the two outside sources disagree with each other by more than we disagree with either, and Fundprint sits between them, nearer PESP:

Our figure is the lower of the two current-state counts, and the reason is already published: PESP lists platforms we have not read. Read them and this dataset would move toward PESP, not away. The three numbers are not strictly like-for-like, since PESP counts facilities open or opening soon and its appendix, like our scope, includes a few owners that are not private equity. The ordering is not close enough for either caveat to change it.

Independence here is partial, and pretending otherwise would not survive a reviewer. The platform list is shared: Fundprint adopted PESP's appendix as its coverage denominator. The counts are not: ours come from operator directories and the federal registry without reference to PESP's numbers, and PESP's come from PitchBook and LevinPro. The two are also not like-for-like, since PESP counts facilities open or opening soon. Key Autism Services is excluded from the totals: we publish zero centers for it under the in-home rule, which is a scope disagreement rather than a counting one, and folding it in would let a definition masquerade as measurement error.

So the estimates sort by unit, not by rigor

Put the four numbers side by side and they do not scatter. They fall into two groups, and the thing that separates the groups is what each method counts.

The bottom two rows cover 16 platforms rather than the whole country, so they are not national totals and the four numbers are not a single series. What the grouping shows is the mechanism: a census built on deals or on registrations sees a company frozen at a moment, and a census built on platforms or directories sees it today. A chain that was bought with 40 centers and now runs 400 is counted at 40 by the first kind and at 400 by the second. That is where 2.8x comes from, and it is a property of the unit, not of anyone's care.

The registry's blind spot is not a constant

This is the part that matters more than the headline. If the federal registry undercounted every operator by the same factor, any registry-based or deal-based count could be corrected with a multiplier and there would be nothing more to do. It does not. Across the 26 brands with 10 or more published centers, the ratio between what the operator publishes and what the registry can see runs from 1.0x to unbounded: the registry sees essentially every center some operators run, and none at all of Wisconsin Early Autism Project.

So there is no adjustment factor to apply, and an operator's visibility cannot be predicted from its size. The only way to close the gap is to read each operator's directory, one at a time, which is what Fundprint does and what takes the time.

“Registry sees” is a visibility test, not a provenance one: it asks whether the building appears anywhere in the federal bulk file, resolved to the registry's own address key. It is the same test that produces the 2.8% national share, so the rows here and that share cannot drift apart. Brands with fewer than 10 published centers are omitted from this table, because a ratio over three sites is noise.

What this does not settle

• Time is inside the ratio and cannot be removed. The published estimate describes 2024-12-31; this dataset is current. Real centers opened in the months between, and some part of the 2.8x is growth rather than method. Separating the two would need the letter's site list, which is not published. We say so instead of claiming the whole factor.
• The counts are not of identical objects. The letter counts ASD service delivery sites attached to acquisitions it identified. Fundprint counts physical ABA therapy centers currently operating, keyed on owner, normalized street and ZIP, excluding head offices, diagnostic-only sites and in-home service areas. Where the definitions differ, they are disclosed rather than reconciled away.
• Our count is a floor too, for a reason of our own. Fundprint has not read every PE-backed platform. The ones it has not are named individually, with their facility counts, on the coverage page. Any number here would rise if that work were finished.
• The federal registry never marks a closed clinic closed. Both counts inherit that. Fundprint removes centers its owners' own directories no longer list, which is why some of its per-operator counts run below other published figures.