How Scores Work
The site uses two different confidence measures. They answer different questions and are deliberately named differently:
- Data confidence (per hospital) — how usable is this hospital's published file?
- Comparison confidence (per service context) — how solid is this cross-hospital comparison?
Neither is a legal-compliance finding, and neither says anything about quality of care.
Data usability score (per hospital)
Each hospital gets a 0–100% usability score: the simple average of five component scores, each 0–100%.
| Component | What it measures |
|---|---|
| File freshness | How recently the file says it was updated: within 90 days = 100%, within 180 days = 75%, within 1 year = 50%, older = 25%. |
| Code coverage | Share of the hospital's charge items carrying a billing code usable across hospitals. |
| Dollar-amount coverage | Share of price rows published as dollar amounts rather than percentages or formulas. |
| Insurer-name matching | Share of payer rates whose insurer name we could match to a canonical insurer. |
| Comparison readiness | Share of classified rows with full service context (specific code, setting, and billing type). |
Data confidence bands come straight from the overall score:
| Band | Threshold |
|---|---|
| High | 85% and above |
| Moderate | 70% – 84% |
| Limited | 50% – 69% |
| Low | below 50% |
Read the score as "how much of this file can honest comparison actually use." A low score often reflects file format or labeling choices — for example, publishing percentages instead of dollars, or omitting billing types.
Comparison confidence (per service context)
Set by cohort size and description availability:
| Band | Rule |
|---|---|
| High | 10+ hospitals report the context and the service has a description |
| Moderate | 5–9 hospitals |
| Limited | 3–4 hospitals (meets the floor, but barely) |
| Low | below the 3-hospital floor — no market statistics published |
Market position bands (per hospital, per service context)
Where a hospital's price sits against the corpus median for one exact context, published only above the 3-hospital floor:
| Band | Rule |
|---|---|
| Well below market | 25%+ below the median, or bottom decile |
| Below market | 10–25% below the median, or bottom quartile |
| Near market | within 10% of the median |
| Above market | 10–25% above the median, or top quartile |
| Well above market | 25%+ above the median, or top decile |
The same ±10% / ±25% thresholds apply to insurer-market positions (a negotiated rate vs. the same insurer's median across hospitals).
Variation bands describe a whole context's spread: "very high variation" means the 90th-percentile price is at least 3× the 10th-percentile price; "high variation" means at least 2×.
Why we publish the thresholds
Publishing the cutoffs makes each label reproducible. Every cutoff above is implemented in the open-source data layer and covered by automated tests. If a threshold changes, this page must change with it; the build identifier on the downloads page identifies the corresponding code version.
What the scores are not
- Not legal-compliance findings — we do not assess conformance with federal requirements, only practical usability for comparison.
- Not care-quality measures.
- Not price levels — a hospital can publish superbly usable data and have high prices, or vice versa.
- Not precise to the decimal — treat small score differences between hospitals as noise; the bands matter more than the ranks.