Methodology

We download the monthly full release of NPPES, the federal registry of healthcare providers. The file is about 1 GB compressed and 11.6 GB as CSV, covering roughly 9 million records.

From it we keep a record only if all of the following hold:

• It is an individual (Entity Type 1), not an organisation.
• The NPI is not deactivated.
• There is a practice address in the United States.
• At least one of its taxonomy codes is in scope — psychologists, clinical social workers, counselors, marriage and family therapists, behavior analysts and psychiatrists on the mental health side; physical and occupational therapists on the rehab side.

That leaves 2,013,859 providers across 17,106 places and 66 licence types.

One scoping decision worth knowing about

The taxonomy code set puts psychiatry and neurology in the same classification, so taking that classification whole would pull in epilepsy, sleep medicine and vascular neurology specialists. We list the psychiatry codes explicitly instead. The cost of that choice is that a newly added psychiatric sub-specialty would be missed until someone notices, so the pipeline prints the excluded codes on every run.

NPPES gives a city name, a state and a ZIP code — but no coordinates. We resolve a provider to a place two ways and cross-check them: by exact city-name match within the state, and by ZIP code using the Census ZCTA centroid.

Name matching alone is dangerous. There are more than thirty Springfields, and a naive match puts people a thousand kilometres from where they actually are. So any name match whose city sits more than 60 km from the ZIP centroid is discarded and the ZIP is used instead. That check currently rejects a few thousand matches.

Why the maps show areas, not pins

Because ZIP centroids are the only coordinates we have. In Manhattan ZIP 10001 alone, nearly 2,500 providers share a single point while their actual addresses are spread across the neighbourhood. Dropping a pin next to a street address would imply a precision we do not have, and someone would drive to the wrong place.

So city and state pages show density by ZIP code — circle size is the number of providers registered there — and provider pages show the ZIP area with a link out to a mapping service that does geocode street addresses.

Counts, shares, rankings and per-capita figures are computed once when the data is loaded and stored alongside it. Nothing is calculated in your browser.

Per-capita figures use Census population estimates for incorporated places. Census designated places — unincorporated communities such as Bethesda, Maryland — are included on this site but have no population estimate, so their pages show counts without a per-capita figure rather than showing an invented one.

Where a page reports a rank, the denominator is stated with it: "2nd of 921 places in Texas" rather than a bare "2nd".

Providers per 10,000 residents is the most useful number on this site, and for a few hundred places it is meaningless. The numerator counts everyone who registered a practice address in a place; the denominator counts everyone who lives there. For most places those two populations are close enough. For two kinds of place they are not, and the result is a figure that looks authoritative and is wrong.

Office districts and enclaves. Berry Hill, Tennessee is a 1,864-resident municipality of commercial buildings inside Nashville. Divide its registered providers by its residents and you get roughly 1.6 therapists per person. To catch these, we compare each place against the twenty kilometres around it: if it has three or more times the providers that an even spread across that area would give it, the residents inside the boundary are not the population being served. Genuinely dense cities are well under that line — Ann Arbor scores 1.0, Pittsburgh 2.0, Salt Lake City 2.2.

Single-employer towns. In Prestonsburg, Kentucky, 1,393 of the 1,484 registered providers share one street address. The count is accurate; it is the size of one organisation's roster, not the number of practices a resident could choose between. Where the largest single address holds at least fifty registrations and at least 40% of the local total, we treat the per-resident figure as describing that employer rather than the town.

Rural service towns. The catchment test above needs neighbours to compare against, and some places have none: Stanley, Idaho has 124 residents, 37 registered providers and nothing within twenty kilometres, so comparing it to its surroundings compares it to itself. Where a place is that isolated and still has more than 250 providers per 10,000 residents, the practices there are serving a county rather than a town — Bethel, Alaska is the clearest case, a hub for dozens of villages with no road connection. The count is right; the ratio would credit the town with capacity that belongs to a whole region.

And a backstop. Those three tests each name a mechanism, and a list of mechanisms is never complete. Nevada City, California has 379 providers for 3,174 residents and passes all three — its neighbours are dense too, so nothing looks anomalous — yet the ratio still says that 12% of everyone living there is a licensed therapist. So there is a final rule that does not ask why: above 500 providers per 10,000 residents we do not publish the figure, whatever explains it. That threshold sits above the densest towns we can actually vouch for, such as Northampton, Massachusetts at 462. In these cases the page says plainly that we can see the ratio does not hold but not which ordinary explanation applies — which is the honest thing to say, and better than picking one.

In all four cases the page shows the count and explains the omission rather than printing a ratio with a footnote. These places are also excluded from per-resident rankings and from the distribution charts, since leaving them in would stretch the scale until every ordinary town looked under-served by comparison. Rankings additionally require at least a thousand residents: below that, one clinic opening or closing moves a town from bottom to top, which makes for a volatile league table rather than an informative one.

• Registration is not the same as practising. An NPI persists once issued. Someone who has retired or moved may still appear until their record is updated.
• Addresses are self-reported and not always current. Some solo practitioners register a home address.
• The registry records nothing about availability, fees or insurance, and nothing about whether a licence is currently in good standing. Only a state licensing board can tell you that.
• The generic Social Worker taxonomy is ambiguous. It does not distinguish clinical practice from case management or child welfare, so a listing under that code does not by itself mean the provider offers therapy.
• School-based providers — school counselors, school psychologists — are counted here but generally serve enrolled students rather than outside clients.