Metric · NBA

UB Impact

v2.0.0

Provide a transparent, season-relative descriptive index from available box-score production.

P S A_{36} = 2 (T S % - T S %_{l g}) \times (F G A + 0.44 F T A) \times 36/ M P

Points above a league-average shooter on the same shot load, per 36 minutes. Volume-weighted, so an elite rate on trivial volume is worth little.

What it measures

Per-minute production is standardized against players in the same role bucket, then shrunk toward league average in proportion to minutes actually played.

How to read it

Higher is better within the selected season; compare percentiles within the stated population.

Why it exists

This is the legacy player-card index currently materialized in nba_player_season_stats. It is intentionally auditable and descriptive. It is not an adjusted plus-minus estimate, an on/off statistic, or a points per 100 possessions contribution. UBPM is a separate research model and must not be substituted into this column without a versioned migration.

Limits

This is a rate, not a season total. A good backup can score well on few minutes; rank players by UB Value instead.
Box-score derived; it cannot see positioning, deterrence, matchup difficulty, or off-ball value.
Component weights are chosen, not fitted against any outcome. Nothing here is validated against on/off, RAPM, or a predictive target.
Role buckets are Guard, Wing and Big, taken from the declared position where one exists and derived from box-score shape otherwise.
The index is not points per 100 possessions, on/off, RAPM, or WAR.

Calculation, evidence, and implementation

Calculation

O = 0.34 z (P S A_{36}) + 0.20 z (U S G %) + 0.18 z (A S T %) + 0.12 z (A S T_{36}) + 0.08 z (O R E B_{36}) - 0.08 z (T O V %)

Offensive component, standardized within role bucket rather than across the whole league.

D = 0.30 z (S T L_{36}) + 0.30 z (B L K_{36}) + 0.30 z (D R E B_{36}) - 0.10 z (P F_{36})

Defensive event component; it is incomplete without tracking and matchup data.

I = c l am p (50 + 10 x M I N / (M I N + 1200) x (0.60 O + 0.40 D), 0, 100)

Shrinkage keys on total minutes, not games played, before the affine 0-100 display scale.

Inputs

Ts Pct, Fga, Fta, Usg Pct, Ast Pct, Tov Pct, Ast 36, Oreb 36, Dreb 36, Stl 36, Blk 36, Pf 36, Min, Gp

Methodology

Eligible-season z-score index with common empirical-Bayes exposure shrinkage.

Normalization

0-100 index with 50 as the eligible-population center; it is not a natural-unit impact estimate.

Uncertainty

No confidence interval is published for this legacy index. Reliability should be read from exposure and sample qualification until component-variance calibration is available.

Evaluation

Directionality and stability checks; a validated natural-unit UBPM model is separate.

Sources

Pipelines/NBA Ingest/Stats.Py, Src/Sports Intelligence/Leagues/NBA/Models/Impact.Py

Implementation

Registry identifier: NBA_IMPACT_TOTAL. Version 2.0.0; status drafted.

Metric distribution →General methodology →