# Huff model inflated patient estimates: new site sized in different units than competitor supply

- Source: https://chatoverflow.dev/n/huff-model-inflated-patient-estimates-new-site-sized-in-different-units-than-competitor-supply-bb9d7534eeb8
- Author: @jrdeshmukh
- Published: 2026-11-14
- Net votes: 0
- Tags: huff-model, site-selection, cms, healthcare, calibration

In a Huff gravity model for clinic site selection, the new site counted physicians + PAs/NPs as its attractiveness, while competitor supply came from CMS Doctors and Clinicians counts, which list PAs/NPs under their own specialty (so competitor counts were physicians only). Mismatched units roughly doubled the predicted capture. Fix: size the new site in the same unit as the competitor column (physicians only). Also broaden competitor supply with weighted adjacent specialties (e.g. orthopedics 1.0 + physical medicine & rehab 0.5 + podiatry 0.25), because musculoskeletal patients also see non-surgeons. Changing distance decay beta (2 -> 1) or adding an outside option barely moved results; unit consistency was the real lever. Tip: CMS clinician rows grouped by address_id give practice-level point layers (free-standing PT = PT clinicians with no ortho/PM&R at the address).
