Testing a score against something it was never shown.
Construct validity asks one question: does the score separate facilities on something it never got to look at? That last clause is the entire discipline, and it rules out almost every obvious way of checking.
Take the CMS Value-Based Purchasing score, the obvious benchmark. About three quarters of it is built from 30-day mortality, infection rates and the patient survey. Those are exactly what the Clinical and Experience lenses read. Testing the full ProxiScore against it would be a number agreeing with itself, and it would look great. So we do not report that as the headline. We report it as the cautionary one.
Three tests. The honest one is not the flattering one.
Loading the latest run...
Which lenses carry the signal.
A composite can look fine while one lens does all the work and the rest ride along. This breaks the score apart and tests each lens on its own, against both benchmarks. A lens near zero is not necessarily broken: Proximity has no reason to predict clinical quality and should read flat. It is here so you can see which lenses earn their weight.
Loading...
The limits, stated plainly.
This is not a claim that the ProxiScore predicts sales. We have no access to your pipeline, and nobody publishes a national dataset of who bought what. Any vendor claiming a validated link to revenue is describing their own customers, not yours.
It is a claim that the score separates facilities on public outcomes it never read. That is a smaller claim and it is one we can actually stand behind, because you can check it. The measure it is tested against, the Excess Readmission Ratio, appears in no lens. A unit test fails our build if anyone ever adds it.
Correlation here is modest by design, and a modest number is the honest one. The published literature puts hospital financial performance against VBP at roughly r = 0.28. A prospecting score built from beds, margin, shortage and market structure should land in that neighbourhood. If this page ever showed a very high correlation, the first thing we would do is look for the leak.
Recomputed nightly against every acute hospital in the country. If a run fails, this page says so rather than showing you yesterday's number dressed as today's.