EWA
Estimated Wins Added is one number for every NBA player: how many wins they've added to their team this season, based on what actually happened on the court possession by possession — not box scores, not vibes.
Note on the name:“Estimated Wins Added” was used in the early 2000s by John Hollinger as a linear function of PER (PER−11) × Minutes / 67. That version is no longer maintained. The version on Alleygorithm is a different methodology entirely — ridge regression on possession-level win-probability change, with role-aware decomposition and empirical-Bayes shrinkage. Same goal (per-game wins-added), modern math.
Plus-minus says “point differential while you were on the court.” That breaks immediately. In recent seasons, players like Payton Pritchard and Luke Kornet posted higher raw plus-minus than Curry, Giannis, or Luka — not because they're better, but because they shared the floor with stars on winning teams.
EWA splits the credit fairly. It controls for who's on the floor with you, who you're playing against, and how much each possession actually mattered to the outcome. A bench player on a great team can't inherit his teammates' impact.
EWA isn't just one number. We decompose every player's value into the moments that produce it — scoring, playmaking, rebounding, rim protection, perimeter defense. The same engine that fits headline Wins / 100 poss also fits a role-by-role breakdown so you can see where a player adds wins, not just whether they do.
Nikola Jokić's rate over the last three seasons is +8.16 EWA per 100 possessions. Decomposed by role, 84% of that comes from assisting— not scoring, not rebounding. Two players with similar headline Wins / 100 poss can have completely different role profiles, and that's the part most ratings hide.
Two great players don't always equal a great pair. We compute the same EWA for every observed pair, trio, quartet, and 5-man lineup, then compare what each group actually produced to what stacking their individual EWAs would have predicted. The public read starts with observed contribution; the baseline gap is secondary context.
Jokić + Murray adds +1.4 EWA together. Strong on its own, but they underperformwhat stacking their individual numbers would predict — the baseline gap is negative. Other duos overperform by more than their individual sum. Knowing which is which is the difference between “great roster” and “great team” context.
We tested EWA honestly: the prediction layer is fit on past games only, then scored on future ones. (One technical caveat — the underlying win-probability labeler isn't yet retrained per fold — is documented on /methodology.)
In rolling-origin backtests, roster-aware EWA improved probability quality over team-only baselines across four chronological folds. Vegas odds are still substantially stronger — they have information we don't — and we don't pretend otherwise. Full methodology, validation tables, and known limitations are published here.