PickALeague WAR: how we calculate it
WAR — wins above replacement — answers a simple question: how many wins did a player add to his team, compared with a player it could have called up from the AHL for next to nothing? Here is the calculation, step by step, with the 2026-27 numbers.
The data
Everything starts from two files the NHL publishes for every game: the play-by-play — every shot with its location, every goal, faceoff and penalty, the game state and score — and the shift charts — who was on the ice, to the second. The calculation covers 5 seasons, 2022-23 to 2026-27, each informing the others. For 2026-27: 315 games, 26,185 unblocked shot attempts and 107,949 even-strength stints.
The season has 1,961 regular-season goals, shootouts excluded — the official count. The model sets aside 107 empty-net goals, where expected goals mean nothing, and works on 1,854 goals.
1 Expected goals (xG)
The expected-goals model gives every unblocked attempt its probability of becoming a goal: a logistic regression on distance and angle; shot type; rebounds (a same-team attempt in the previous 3 seconds); rush chances (an event in the neutral or defensive zone in the previous 4 seconds); a faceoff win just before; a takeaway or opponent giveaway just before; and power play or short-handed play.
xG = 1 / (1 + e^-(b0 + b1·distance + b2·angle + … ))
Arena adjustment. Every rink has its own scorekeepers, and they don't all place shots the same way. We compare each arena's distribution of recorded distances with the league's and remap by quantiles. Median-distance gaps range from -2.60 ft (EDM) to +1.90 ft (MIN).
Always out of sample. A season's xG is predicted by a model trained on the other seasons (2022-23, 2023-24, 2024-25, 2025-26), never on its own shots. For 2026-27 it expects 1,854 goals against 1,854 actually scored on those attempts, with an AUC of 0.745.
2 Even strength: offense and defense (RAPM)
Every game is split into stints during which the same players are on the ice. Each stint yields two observations, one per team on offense:
xG for / 60 = intercept
+ Σ offense(players of the attacking team)
+ Σ defense(players of the defending team)
+ home ice
+ game state (4v4, 3v3)
+ score differential (−2 or worse, −1, +1, +2 or better)
+ zone start (offensive, defensive, neutral)Solving this for every player at once, weighted by stint length, is RAPM. The controls absorb what isn't the players' doing: overtime 3v3 is a far more open game; a trailing team pushes and a leading team sits back; a line sent out for a defensive-zone faceoff starts far from the other net (the zone start lasts while the same skaters stay on). Relative to 5v5, tied, on the fly (xG per 60):
| 4v4 | +0.24 |
| 3v3 | +1.17 |
| Trailing by 2+ | +0.34 |
| Trailing by 1 | +0.03 |
| Leading by 1 | -0.12 |
| Leading by 2+ | -0.19 |
| Offensive-zone start | +0.46 |
| Defensive-zone start | -1.13 |
| Neutral-zone start | -0.83 |
Ridge with a prior. A ridge penalty λ reins in each coefficient — pulling it not toward zero but toward the player's value the previous season (2025-26). Established players start from their measured value and one season corrects it; rookies start from zero.
min Σ length × (y − ŷ)² + λ × Σ (coefficient − last season's coefficient)²
λ is picked by five-fold cross-validation on games: λ = 96,000. The minimum is flat: any λ between 48,000 and 256,000 gives the same error within 0.01%. Then to goals:
EV offense = offense coefficient × even-strength hours EV defense = − defense coefficient × even-strength hours
3 Power play and penalty kill
Same method on stints where one team has more skaters, with the same controls and 5v4 as the reference:
| 5v3 | +7.13 |
| 4v3 | +5.31 |
| Offensive-zone start | +1.97 |
| Defensive-zone start | -0.12 |
λ = 24,000, intercept 5.14 xG per 60.
4 Shooting
Shooting = goals scored − expected goals of his shots
5 Penalties
A power-play minute yields 0.1392 goals for the team with the advantage, against 0.0423 at even strength.
Value of a minute = 0.1392 − 0.0423 = 0.0969 goals Penalties = (minutes drawn − minutes taken) × 0.0969
An empirical value and a simplification: a power play ends at the first goal. Only minors and bench minors count.
6 Goalies
Goals saved above expected (GSAx) = expected goals against − goals allowed
7 Replacement level
Inspired by hockeystats.com, in simpler form: replacement players are forwards beyond 13th in team ice time, defensemen beyond 7th and goalies beyond 2nd in games played. Their output per hour (per shot faced for goalies) sets the rate: forwards -0.161, defense -0.119 goals per hour; goalies -0.004 goals per shot.
Goals above replacement = goals above average − replacement rate × hours played
8 From goals to wins
The Pythagorean formula links goals and wins: W% = GFx ÷ (GFx + GAx). Christopher Long estimated x = 2.022 on his own goal definition; since WAR drops empty-net goals, we re-estimate x on that same universe — over 524 team seasons from 2009-10 to 2025-26: x = 2.182. Around an average team scoring g goals per game, one more goal is worth x ÷ (4g) wins:
g = 1,854 goals ÷ 630 team games = 2.943 Goals per win = 4 × 2.943 ÷ 2.182 = 5.39 WAR = goals above replacement ÷ 5.39
9 Confidence interval
WAR is an estimate. To measure its precision we redo the whole calculation 40 times on resampled seasons — games drawn with replacement, some twice, some not at all — recomputing RAPM, shooting, penalties, goalies and replacement. Nine times out of ten a player's WAR lands between the 5th and 95th percentile of those draws: that is the interval shown. Two players whose intervals overlap widely can't be told apart.
A complete example
| Even-strength offense | +2.30 |
| Even-strength defense | +0.02 |
| Power play | +0.97 |
| Penalty kill | +0.03 |
| Shooting | +6.88 |
| Penalties | +0.19 |
| Goals above average | 10.39 |
| Average → replacement gap | +1.11 |
| Goals above replacement (GAR) | 11.50 |
| WAR = 11.50 ÷ 5.39 | 2.13 |
| 90% interval | 0.98 to 3.03 |
Each line is rounded to the hundredth and the totals are computed from those rounded values, so the arithmetic can be checked. The leaderboard keeps every decimal, hence a possible one-hundredth difference.
Updates
Every Monday the current season is recalculated: new games are read, the xG model from completed seasons scores them, and RAPM starts from last season's coefficients. A season is published from 150 games on.
What WAR doesn't say
- It is an estimate: the interval says so.
- Passing and play away from the puck only count through expected goals: the NHL doesn't publish passes.
- Our numbers are not hockeystats.com's or Evolving-Hockey's: same idea, simpler model — a logistic xG model, one replacement threshold.
Sources
Method inspired by hockeystats.com and Evolving-Hockey; arena adjustment after Schuckers and Curro. Data: NHL. Calculation: PickALeague, updated November 17, 2026.
