How the ratings work
Each home gets a risk score from 0 (least concern) to 100 (most), built from seven kinds of public evidence. Homes are then split into five equal groups. Method version 0.1; every number here comes from the project's open weights.yaml file.
1. Points for each inspection citation
Inspectors grade every citation A to L by how severe and how widespread it is. We use the same points CMS uses for its own inspection rating:
| Letters | Meaning | Points |
|---|---|---|
| A, B, C | No actual harm, minimal potential for harm | 0, 0, 0 |
| D, E, F | Potential for more than minimal harm | 4, 8, 16 |
| G, H, I | Actual harm to residents | 20, 35, 45 |
| J, K, L | Immediate jeopardy to residents' health or safety | 50, 100, 150 |
Older citations count less: a citation's points halve every 18 months. A problem cited again at a later inspection counts 1.5×. Points are adjusted for home size (per 100 beds).
2. Seven kinds of evidence
For each kind, a home's percentile among PA homes is calculated (100 = more concern than every other home). The score is the weighted average.
| Evidence | Weight | What it measures |
|---|---|---|
| Inspection points | 35% | All citation points from step 1 |
| Serious events | 15% | Harm (G–I) and immediate-jeopardy (J–L, counted 3×) citations with the same fading, plus the CMS abuse flag and Special Focus status |
| Staffing levels | 20% | Payroll-reported nurse hours per resident per day, adjusted for how sick residents are: total (50%), registered nurse (30%), weekend (20%) |
| Staff turnover | 10% | Share of nursing staff (50%) and RNs (30%) who left in a year, plus administrators who left (20%) |
| Fines and payment bans | 10% | Federal fines over about three years and Medicare payment denials |
| Complaints and repeats | 5% | Citations from complaint investigations, plus repeat citations, size-adjusted |
| Hospital and ER transfers | 5% | Medicare claims: long-stay hospitalizations per 1,000 days and short-stay rehospitalization rate |
Facility self-reported quality measures are left out on purpose: research has found some homes inflate them. If a kind of evidence is missing for a home, it is skipped and the other weights are scaled up; the confidence label drops when data is missing or old.
3. Five groups and one safety rule
Homes are ranked by score and split into five equal groups: Fewest concerns Fewer concerns Middle More concerns Most serious concerns
Safety rule: a home with an immediate-jeopardy citation in the past 12 months, the CMS abuse flag, or federal Special Focus status is never placed above "Middle", whatever its score.
Limits you should know
- Inspections record problems, not strengths. A home with few citations may have been inspected less often.
- Since 2026 CMS inspects some high-performing homes with a shorter survey, which finds fewer problems by design.
- Staffing comes from payroll data CMS publishes about a quarter late.
- PA Department of Health state-only citations and sanctions are not yet included. They are being added.
- Ratings compare homes within PA only.
Want to see how different weights change the results? Open the weighting sandbox.