Methodology

Provider records come from the CMS National Plan and Provider Enumeration System (NPPES) Full Replacement Monthly NPI File, specifically the June 2026 snapshot: npidata_pfile_20050523-20260607.csv (11.5 GB). Population denominators and county definitions are the Census Bureau Vintage 2025 county population estimates (POPESTIMATE2025). Map geometry is drawn from Census cartographic boundary files, cb_2024, 1:20,000,000.

A record is included when all four hold (1,481,478 providers in this snapshot; 12,990 otherwise-qualifying records excluded):

1. It is an individual (NPPES Entity Type 1), not an organization.
2. The NPI is active, carrying no deactivation date (3,532 deactivated records excluded).
3. The practice address is in the 50 states or DC (7,975 in territories, 967 at military postal codes, and 516 foreign/unknown excluded).
4. At least one of the record's 15 taxonomy slots carries an in-scope behavioral-health code:

One group per provider. A provider listing several in-scope codes is counted once: under the code NPPES flags as primary, or failing that the first in-scope code by column order. Provider-type mixes therefore sum to 100%.

NPPES practice addresses carry USPS ZIP codes, which don't nest inside counties. Each provider is allocated fractionally across the counties their practice ZIP touches, using the HUD USPS ZIP–County crosswalkweighted by business-address share (with total-address share as fallback for ZIPs without business addresses), restricted and renormalized to counties within the provider's practice state. County "providers" values are therefore fractional; state and national counts are exact integers.

HUD USPS ZIP–County crosswalk 2021 Q4 (business-address ratios; last direct-download vintage), with legacy county codes remapped to the Vintage 2025 county set, including Connecticut's nine planning regions, via ZCTA-centroid containment against cb_2024 boundaries. The remap matters for Connecticut: 2022+ Census products replaced its eight legacy counties with nine planning regions, and this dataset reports planning regions consistently across counts, population, and geometry.

1,798 providers (0.12%) had a practice ZIP that could not be allocated to an in-state county. They are included in state and national figures but absent from county rows, so county columns sum slightly below state totals.

Rates are providers per 100,000 residents against the county or state population estimate above. County per-100k rates are withheld in two cases. First, when a county has fewer than 3 allocated providers or fewer than 1,000 residents, because numbers that small make a rate unstable rather than informative. Second, when a single practice address accounts for 50% or more of a county's allocated providers and that address carries at least 100 providers, because the resulting rate describes one organization's registration practice rather than the local workforce. Provider counts are always shown. 3,144 counties are reported in total. Counties without a published rate render it as "n/a" on pages and carry a rate_suppressed flag plus arate_suppressed_reasoncode in the CSVs ("suppression" is the statistical term for withholding small cells); the underlying count is always shown.

The second rule is a consequence of one address per NPI. A regional employer that enrols every clinician at its headquarters puts a whole multi-county service area's workforce in the county holding that one building, which both inflates that county and thins its neighbours. 42 counties meet the test at the 50% threshold; it applies only to addresses carrying at least 100 providers, so a genuine single-campus employer in a small county keeps its rate. State and national figures are unaffected — the providers are counted, just not turned into a local rate.

Cohort charts group active providers by the year their NPI was first enumerated ("Provider Enumeration Date"). This measures entry into the NPI registry, not licensure, because most clinicians register when they first bill insurance. Charts start at 2010 because the registry opened in 2005 and its first years are dominated by already-practicing clinicians signing up. The current year is always a partial cohort. 0 records lacked a parseable enumeration date and are omitted from cohort charts only.

Practice addresses are normalized (suite/unit designators stripped, USPS suffixes folded) and providers grouped by identical address key. "Only provider at their location" means no other in-scopeprovider shares the key, which is a deliberately narrow proxy. Office buildings without suite numbers can merge distinct practices; addresses that can't be normalized fall into the visible "unknown" bucket. This is a location statistic, not an employment or group-affiliation statistic.

• One address per provider. Telehealth panels, satellite offices, and community-based work don't appear; low registered density is not proof of an unserved population.
• Employer addresses concentrate counts. The same one-address rule pushes the other way: large organizations often register every clinician at a single headquarters, so a county's count can include providers who work across a whole region. The suppression rule above catches the clearest cases, but milder concentration remains in the published rates, and it is why a county rate should be read as registrations per resident rather than clinicians available locally.
• Headcounts, not FTEs. Part-time and full-time providers count equally; no capacity or availability signal.
• Taxonomy is self-designated at registration and is not verification of an active state license. Board license tallies will differ, usually running higher per state because multi-state licensees appear here only once.
• Registry lag. Retirements and moves reach NPPES late; deactivated records are excluded, but deactivation itself is often delayed.
• Crosswalk vintage. ZIP–county weights predate 2022 boundary changes except where remapped as described; ZIPs created after the crosswalk go unallocated (the disclosed unmatched share).

The dataset, tables, and chart images are licensed CC BY 4.0. Cite as: certification.app, "US Mental Health Workforce" (June 2026 NPPES snapshot), with a link to https://www.certification.app/data/mental-health-workforce. NPPES and Census source data are US public domain.

Versioning promise: snapshots are annual and additive. This page and the CSVs for the June 2026 snapshot stay at their current URLs when the next snapshot ships, so citations keep resolving to the vintage they cite.