Wins Above Average (WAA / GAA)

The companion to WAR: the same all-in-one value, measured against an average NHL regular instead of a freely-available replacement - with the two numbers verified to differ in exactly one place.

Why two numbers

WAR and WAA answer two different counterfactual questions with the same underlying value engine. WAR asks 'how much better than a freely-available replacement is this player?' - the right lens for roster construction and cap value, because a call-up is what you actually fall back on when someone gets hurt. WAA asks 'how much better than an average NHL regular is this player?' - often the more intuitive comparison when you put two established stars side by side, where nobody involved is anywhere near replacement level. Both roll up the identical set of components documented on the GAR / WAR page; they differ in exactly one place.

How they relate

The only thing that changes between the two numbers is the baseline the even-strength piece is measured against. WAR subtracts a real, below-average replacement floor - the bottom slice of qualified NHL regulars at each position. WAA subtracts league average instead. Every other component - power play, penalty kill, penalties drawn and taken, faceoffs, and finishing - is carried through unchanged, and both totals are converted to a wins scale by the same goals-per-win factor. So the difference between a player's WAR and WAA is one of altitude, not of who is better: it is purely the gap between 'average' and 'replacement' on the even-strength piece, scaled by how much a player plays.

When to use which

Use WAR when the counterfactual is a real roster decision - surplus value against a cap hit, or what a team actually loses to an injury. Use WAA when you want a cleaner 'how far above the middle of the league' read, especially near the top of the league where replacement level is a distant abstraction and everyone clears it comfortably. Both are published per season and pooled across our 2020-21 to 2025-26 window (roughly 900 to 1,000 qualified players a season, on the same six seasons of public play-by-play that feed every other card on the site).

How we keep it honest

Because the two numbers share an engine, the cleanest check on WAA is internal consistency: recomputing both directly from the component tables, the total gap (WAR minus WAA) reconstructs the even-strength-baseline difference to floating-point precision - a mean residual on the order of a billionth of a win across roughly 1,575 pooled player-rows. In plain terms: nothing else moves between WAR and WAA, exactly as intended, and we verified it rather than assumed it.

The two metrics also behave the way the design predicts. Because replacement sits below average, WAR is at least as large as WAA for every single player in the pool, and the gap widens with ice time - a player who logs big minutes clears the replacement floor by more total goals than a part-timer does. At the very top, where the metrics are meant to compare established stars, the ordering is nearly identical: 24 of the top 25 players are common to both lists, and the two agree almost perfectly on how to rank the elite tier.

What we checked	Result	Reading
WAR ≥ WAA (every player)	1,575 of 1,575 (100%)	Replacement sits below average, as designed
Gap (WAR − WAA) vs. ice time	correlation r = 0.95	The gap is an ice-time effect, not a talent re-ranking
Top-25 overlap	24 of 25 shared	Near-total agreement on the elite tier
Top-50 rank agreement (Spearman)	0.986	The two metrics rank stars almost identically
Overall linear agreement (Pearson, n=1,575)	0.989	The numbers track each other closely across the pool

Two honest caveats. First, 'nearly identical ordering' is true at the top, where the page focuses, but not across the whole league: rank the entire pool and the two metrics agree only moderately (season-by-season Spearman of about 0.82 to 0.89), because the ice-time-dependent baseline shift reshuffles the crowded mid-pack where players are separated by fractions of a win. Second, WAA is a descriptive value metric, not a betting or forecasting signal - the only quantitative validation here is the internal WAR-versus-WAA consistency above and the shared team-goal-differential relationship that underpins WAR, not a predictive claim. On naming: it is a common observation that many public 'WAR' models baseline their dominant even-strength piece at league average and are therefore, under the hood, closer to a wins-above-average number - but that is a general observation about other people's models, whose internals we cannot inspect, not a measured fact. We split the two apart and label each for exactly what it is.

Validation figures reflect the data through 2025-26 and are recomputed each season. Head-to-head benchmark tests (log-loss, AUC, bootstrap intervals) are dated snapshots of a specific held-out season, labeled where they appear.