Transition-credit RAPM

Isolating one player's impact

Raw on-ice results are muddied by who else is on the ice. They also depend on who the player was matched against and how the coach used him. Our RAPM untangles that. Over several seasons, it accounts for a player's teammates and the quality of his competition. It also accounts for how the coach used him. What is left is an estimate of the impact that belongs to that player alone. On our cards and boards, higher is always better for both.

The transition-credit adjustment

Plus-minus-style ratings tend to hand a rush chance's value to whoever shoots it. That quietly short-changes the player who carried the puck up the ice to create it. The blind spot falls hardest on puck-moving defensemen. A pure on-ice model captures their defense well. But it keeps under-crediting their offense. We correct for it with our own transition credit. On a same-team zone transition, the puck-mover earns a share of the chance that follows. That share is pulled toward average when the sample is small. So a few flashy carries can't inflate a rating. We report the credit as its own measure, with its own percentile. The offense percentile ranks isolated offense alone. The transition percentile sits beside it wherever the offense one appears. So you get two honest measures instead of one blended number. The credit still counts toward goals above replacement, where even-strength offense includes it. So the defensemen who drive play rather than finish it keep that value.

How the pull toward average is set

A rating like this pulls small samples toward the league average. That keeps a player with few minutes from getting a wild number. For the multi-season rating we chose a strong pull on purpose. It keeps short stints from swinging the board. Every rating is set up the same way. So the board compares fairly across positions.

For the single-season rating, how hard it pulls is set from the data.

From a rating to goals and wins

A rating can rank players sensibly and still be off on how much each player is worth. This RAPM is the engine under our goals-and-wins value numbers. Each season's rate for turning goals into wins comes from how many goals the league scored that season. It is worked out on our own data.

The count below is the career-pooled rating, with one row per skater. So it counts players, not player-seasons. A season's percentile is pooled over several seasons up to that one. It is ranked against the pool as it stood then. So an earlier season reads the way it did at the time, without later seasons folded in. The rating on a season's card is that season's own.