PER is a per-minute box score composite developed by John Hollinger and published by Basketball-Reference and ESPN. It weights 14 statistical categories with coefficients Hollinger chose by judgment, normalizes for pace, and scales so that the league average is always 15.0. A PER of 30+ represents historically elite, MVP-level production.
Widely published and recognized. Available for all historical seasons. Gives a single number that is calibrated (league average = 15) and therefore interpretable without context. Has successfully identified many all-time great players at the top of its leaderboard.
- Arbitrary coefficients: the relative weights of different stats (e.g., a steal is worth more than an assist) are Hollinger's judgment calls, not derived through regression against winning.
- Essentially no defensive component: DBPM, blocks, and steals are the only defensive inputs. Players who defend well without generating these box score events are systematically undervalued.
- Big-man bias: due to rebounding overweight and position normalization issues, centers dominate PER leaderboards in ways inconsistent with impact metrics.
- Ignores opponent quality, usage context, and team strength.
- Largely obsolete for serious analytical purposes — the analytics community has near-consensus that BPM and RAPM-family metrics are strictly more informative.
PER is excluded from APEX entirely — neither scored nor displayed. Jewell et al. (JQAS) confirm BPM is more robustly significant as a predictor of player value than PER-adjacent metrics. The analytics community consistently rates PER as the weakest of the major catch-all metrics, with its arbitrary weighting, big-man bias, and near-absent defensive component as disqualifying flaws. APEX's model note: PER's leaderboard does generally identify elite players at the very top — the problem is the large population of center-heavy false positives in the top 20-40 range.