Adjusted Plus-Minus (APM)

APM was the first generation of plus-minus metrics that attempted to isolate individual player contributions from teammate and opponent quality. Developed by Dan Rosenbaum in the early 2000s, it uses linear regression on stint-by-stint lineup data to assign each player a coefficient representing their estimated point contribution per 100 possessions, holding other players constant.

For each lineup stint, the point differential is regressed against a matrix of binary player variables (1 = player on court, -1 = player on opposing team). The resulting coefficients are each player's APM rating.

Represented a genuine methodological advance over box score composites — the first public metric to attempt isolation of individual impact from team context using on-court/off-court data.

APM requires very large sample sizes (multiple seasons) to stabilize because of multicollinearity — players who share most of their court time together cannot be easily disentangled. Single-season APM is extremely noisy. This noise problem led directly to the development of RAPM (regularized APM) and its modern derivatives.

APM is not used in APEX because its noise problem makes single-season evaluation unreliable. APEX uses EPM, LEBRON, and DARKO DPM — all of which are regularized APM derivatives that address APM's variance problem through Bayesian priors, box score inputs, and other stabilization techniques. APM is the methodological ancestor of APEX's entire Impact pillar.