Signal Matrix
How basketball metrics agree, differ, and depend on one another
Correlation describes relationship, not truth. Metrics can agree because they measure similar constructs, share inputs, or respond to the same playing-time and team context.
positive
negative
⛓ shared formula / derived
◐ shared prior
⏱ exposure-driven
≈ unstable across seasons
·n common player-seasons
External comparators — lawful availability
A blocked state records lawful unavailability. It is not a criticism of the provider and implies nothing about metric quality.
How to read this page
- Spearman ρ (primary) asks: do two metrics rank players similarly? Each metric is converted to a percentile within its season and eligible population, ranks are pooled across seasons, and ρ is computed on the pooled ranks. The per-season values are always available in the pair view.
- Pearson r (secondary) is computed on within-season z-scores, never on raw values pooled across seasons.
- Dependency badges matter more than big numbers. A high coefficient against a metric's own formula component is a structural fact, not a discovery. The pair view explains every dependency in plain language.
- Cumulative metrics ride exposure. WAR-style totals rise with minutes and possessions; their agreement always contains shared playing time. The pair view carries an exposure diagnostic.
- What this page can never establish: that one metric is better, validated, or more accurate than another. Low correlation can mean a useful distinct construct or a broken metric — the matrix alone cannot decide which.