Macfax Bayesian Performance Rating
Player impact in points per 100 possessions above Division I average
Macfax Bayesian Performance Rating (BPR) is a player impact estimate expressed in points per 100 possessions above the Division I average. It combines what a player's box-score production predicts about their value with what actual on-court lineup data reveals, then uses Bayesian regularization to produce stable estimates even for players with limited minutes. BPR separates into an offensive component (OBPR) and a defensive component (DBPR).
What It Measures
BPR measures how many points per 100 possessions a player adds to their team's scoring margin when on the court, relative to a Division I average player. Positive values mean the player improves the team; negative values mean the team performs worse with that player on the court. OBPR measures offensive contribution, DBPR measures defensive contribution, and total BPR is their sum. This is fundamentally different from counting stats — a player scoring 18 points per game on poor shooting can have a negative OBPR, while a player scoring 8 on elite efficiency with good decision-making can have a high OBPR.
Why It Matters
Volume stats reward usage, not efficiency. A player who takes more shots generates more raw points regardless of whether those shots help the team. BPR controls for pace and playing time by measuring per possession, and incorporates actual on-court lineup results to capture contributions that box scores miss entirely — spacing, off-ball defense, screening, communication. It also adjusts for schedule strength, so a player putting up big numbers against weak competition is not rated the same as one doing the same against elite opponents.
How to Interpret
BPR is centered at 0, representing replacement level — the typical contribution of a player who can fill a roster spot but does not meaningfully improve the team. Positive BPR means the player adds value; negative means they subtract it in the current model. A player with OBPR +5 and DBPR −1 is a strong offensive contributor who is slightly below average defensively. Always check the source label: RAPM-based estimates (from lineup data) are more reliable than box-only estimates for players with limited minutes.
Formula
BPR = OBPR + DBPR Both OBPR and DBPR expressed in points per 100 possessions above D1 average.
Technical Notes
- BPR combines two sources: box-score predictions (what production patterns predict about value) and lineup-based RAPM (what actually happens on the court with this player in). The two are blended using Bayesian regularization.
- Players with sufficient on-court lineup data receive RAPM-based estimates. Players with limited minutes receive box-score-based estimates. Each BPR value is labeled by its primary data source.
- Multi-year lineup data is pooled where available to improve stability for returning players. Prior seasons contribute context without overriding current-season performance.
- Schedule strength is factored in — players whose teams face stronger opponents are adjusted relative to those producing similar stats against weaker competition.
- On-court performance is compared against the team's overall level to separate individual contribution from team-quality effects.
- Defensive BPR carries materially more uncertainty than offensive BPR. Defensive impact is harder to isolate from box scores and lineup data. DBPR should be interpreted with appropriate caution.
- Exact feature weights, regularization parameters, and calibration values are internal to Macfax and may be recalibrated as additional validation data accumulates.
- •Player ratings are noisier than team ratings by design. College basketball sample sizes are small and lineup combinations repeat infrequently.
- •DBPR is particularly uncertain. Most box-score defensive indicators are weak proxies for actual defensive contribution.
- •Freshmen and low-minute players lack lineup data — their ratings rely more heavily on box-score priors, which carry their own uncertainty.
- •Transfer players have a break in on-court data continuity. First-season transfer estimates carry more uncertainty than returning players.
- •Injuries, foul trouble, and mid-season role changes affect lineup data quality but are not explicitly modeled.
- •BPR should not be the sole basis for player evaluation. Role, system fit, and usage context matter in ways the model cannot fully capture.
Illustrative — not based on a specific live season. Guard A: OBPR +5.8, DBPR +1.4, BPR +7.2. Scores efficiently, protects the ball, draws fouls, and on-court lineup data confirms the team scores more with him on the floor. Source: RAPM-based. Guard B on the same team: OBPR +6.1, DBPR −2.3, BPR +3.8. Higher raw offensive production, but opponents score more when he defends. Box stats alone would overrate Guard B — BPR separates the two clearly. Same team, same playing time, very different actual impact.