EPM is an all-in-one player impact metric published by DunksAndThrees. It uses regression techniques to combine box score inputs with on/off lineup data to estimate each player's contribution in points per 100 possessions above a league-average player, in both offensive and defensive contexts.
EPM combines a box score-derived prior — similar in spirit to BPM — with regularized plus-minus estimates derived from lineup-level play-by-play data. The model is designed to be accurate on a single-season basis, making it suitable for retrospective evaluation rather than just projection.
Widely regarded within the analytics community as among the most accurate public single-season impact estimators. Incorporates on/off data, controlling for teammate and opponent quality in a way that pure box score metrics cannot. Validated through out-of-sample testing against future RAPM.
Full methodology is not publicly documented, limiting independent reproducibility. Now requires a subscription to DunksAndThrees for access (as of 2024-25), creating an access constraint for APEX's live data pipeline. EPM is optimized for single-season retrospective evaluation, unlike DARKO which is forward-looking.
EPM is APEX's highest-weighted single metric, at approximately 30% of the Overall Impact pillar (~9.9% of composite score). Deshpande & Jensen (2016) establish RAPM-family metrics as the most predictive public estimates of player value. Zimmer, Snyder & Zimmer (2025, Economics Bulletin) independently confirm that methods adjusting for opponent quality — which EPM does — produce more valid player valuations than raw alternatives. The EPM paywall is a documented open issue in APEX's data sourcing plan; resolution requires either a subscription or an alternative metric with comparable methodology and coverage.