Net Man-Games Lost as a Coaching and GM Evaluation Metric in the NHL

Abstract

NHL front offices often treat injury burden as bad luck. We argue it is partly organizational, and that the right unit is relative: a team’s man-games lost against its opponents’. On a 17-season corpus with 2024–25 as the focal year, we introduce Net_MGL and its cap-weighted counterpart Net_MGL_Money and evaluate coach and GM tandems on both. The single-season signal is weak, but across 430 full-length team-seasons qualification falls monotonically with burden (odds ratio 0.470 per SD). A crossed random-effects decomposition places the persistent component at the organizational level: about a fifth of the variance in season-long burden is stable franchise and front-office structure, while the head coach’s share is near zero and neither travels between teams nor responds to coaching changes. At the game level a team wins about 1.5 percentage points more often per net man-game of health advantage, an effect that survives within-roster controls yet is small against tandem-quality differences of 6.4 win-rate points. Net_MGL surfaces a health dimension that records obscure and, for long-tenured tandems, a durability screen for hiring.