Editorial standards

Every value in a release comes from a file published by the Centers for Medicare & Medicaid Services. Each release records the exact source files, their vintages and their SHA-256 checksums in its manifest, and the raw files are kept unchanged in the data store. We do not mix in proprietary, scraped or self-reported facility data, and we do not accept data corrections from facilities: a facility that disputes its record should correct it with CMS, and the correction reaches us in the next CMS vintage.

The build code is public at github.com/ttiatun/careratings-open-data. Every derived column is described in the codebook and the methodology shipped inside the release, with the rule that produced it. A change to a derivation is a new build version, recorded in the changelog; a release is never edited in place. Reports name the release they were computed from, and any figure in a report can be recomputed from that release.

Facilities cannot pay to change a value, a rank or a flag, and advertising on this site has no influence on the data or the reports. This is the same zero pay-to-play guarantee that applies to our ratings. The research program has no facility, chain, investor or trade-association funding.

• Ownership is "as disclosed to CMS". The private-equity and REIT flags are the facility's own answers on Form CMS-855A. We report the share of facilities that disclose such an owner, never the share that has one, and we distinguish owners (ownership and control roles) from other disclosable parties such as landlords and vendors.
• Penalties are what CMS publishes. The public penalties file covers a three-year lookback and does not distinguish per-instance from per-day civil money penalties; we do not infer what it does not say.
• Staffing tests are descriptive. Comparisons with the rescinded federal staffing minimums use the reported hours CMS publishes, as KFF did in 2024, and are labelled as such.
• The Care Quality Index is ours. It is a descriptive composite with weights we chose. We publish its components so anyone can reweight it, and we never present it as a CMS measure.
• Small numbers are flagged. State figures built on fewer than ten facilities are suppressed in the index and marked in the reports.

Every release passes an automated validation before it is published: schema and row counts per table, key uniqueness, and a reconciliation of facility totals with the national row of CMS's own chain performance file. The validation report ships inside the release. Reports are checked against the release they cite before publication, and we invite independent review: write to us if you would like to review a report before it is published, or to challenge one after.

When we find an error in a published release or report, or a reader reports one, we confirm it, correct it, and record what was wrong, what changed and when in the corrections log. A corrected release keeps its name and gains a note in its changelog; a corrected report shows the correction at the top. Errors in CMS's source files are reported to CMS and noted, not silently patched.

To report a problem, use the contact page and name the release, table and column, or the report and figure.

Releases and reports are prepared by The Care Ratings research team and published under the organization's name. Reports name their authors and reviewers on the report itself, with their affiliations and any interests that could bear on the work. Read more about the organization on the about page.