Wins Above Replacement (WAR), the Hockey Stat for Total Value

WAR is a baseball idea. FanGraphs describes it as an attempt to sum up a player's total contribution to his team in one statistic, measured against the freely available minor leaguer who'd take his place if he got hurt. Hockey's early versions included Alan Ryder's Player Contribution in 2003 and Tom Awad's Goals Versus Threshold (GVT) at Hockey Prospectus, which Awad explained in 2009 as the hockey version of baseball's VORP, a player's value in goals above what a replacement would've contributed.

The modern models arrived in a rush. Andrew Thomas, Sam Ventura and Alexandra Mandrycky built one for their site WAR On Ice in the fall of 2014. Dawson Sprigings released his own in the summer of 2016, and Emmanuel Perry published a version on Corsica in May 2017.

Evolving-Hockey's is the one most people mean today. Josh and Luke Younggren laid it out in a three-part series at Hockey-Graphs on January 16 through 18, 2019, two days after their write-up of the RAPM regression it's built on. They were open about the goal. They'd made a descriptive model, they wrote, and "we don't care whether it's predictive." The expected-goals version, xGAR, followed that October, and Evolving-Hockey describes it as the model for "deserved" or "expected" evaluation. It isn't the only public model. Patrick Bacon's, published at HockeyStats, gives shooting its own component, and Dom Luszczyszyn's Game Score Value Added does a similar job at The Athletic.

The public models run the same three steps. Measure each part of a player's game in goals above average, move the zero point down to replacement level, then convert goals to wins. Evolving-Hockey splits a skater into six components:

• Even-strength offense (EVO)
• Even-strength defense (EVD)
• Power-play offense (PPO)
• Shorthanded defense (SHD)
• Penalties taken (Take)
• Penalties drawn (Draw)

Here's a worked example with invented round numbers. A forward rates at plus-8 goals on even-strength offense, minus-2 on defense, plus-3 on the power play, zero on the penalty kill and plus-1 for drawing more penalties than he takes. That's 10 goals above replacement. At six goals per win he's worth about 1.7 wins, or a little over three standings points across a season.

The hard part is measuring the components. That's where the models split. Evolving-Hockey runs a long-term RAPM regression for each one, trains a statistical plus-minus model (SPM) on play-by-play and box-score numbers to predict it, and uses the SPM output for each season. In the 2019 write-up, offense was built on actual goals for and defense on expected goals against.

Replacement level is a cutoff by ice time. Anyone outside a team's top 13 forwards or top seven defensemen counts as replacement at even strength, and the special-teams components use each team's top 11 skaters. Evolving-Hockey took that shortcut because there aren't enough league-minimum players to define replacement by salary. Goals become wins through a rolling, season-by-season conversion (about six goals per win in the 2025-26 export).

WAR measures total value against a player you could get for nothing.

Zero is replacement level, so a skater at zero is one a team could swap for a call-up without losing anything, and a negative number says the team would've been better off making that swap. On Evolving-Hockey's 2025-26 numbers the median forward with 60 or more games finished at 0.9 wins, and the top 10% cleared 2.4. Defensemen ran a touch lower (a median of 0.7 among 161). Lane Hutson and Quinn Hughes tied for the lead among all 940 skaters at 4.9 wins, with Hutson ahead on goals (30.2 to 29.9).

Detroit had one of the five best players in the league by this measure. Moritz Seider tied for fourth at 4.3 wins, almost a full win clear of any other Red Wing, with Alex DeBrincat tied for tenth at 3.5 and Lucas Raymond at 2.7. The bottom of the scale is crowded, too, with almost a quarter of the skaters who played 60 games finishing below replacement, and Ben Chiarot's minus-1.2 wins ranked 925th of 940.

How much of it carries over? Across every pair of consecutive seasons in Evolving-Hockey's files since 2007-08, for players with 800 minutes in both, xGAR correlated with the next season's xGAR at 0.50 and GAR at 0.44. That's real signal with a lot of one season's luck mixed in, which is why Evolving-Hockey suggests reading any season's number with a margin of about half a win either way.

The W-Value on every player page

This site's W-Value is a 0 to 100 rating built on Evolving-Hockey's components. It blends xGAR and GAR at 60/40, component by component, leaning toward chances because they repeat better from year to year, and scales the result so 50 is the average NHL skater with 20 or more games. Around the middle a standard deviation of value is worth about 23 points. Why blend at all? Evolving-Hockey's own awards use the average of the two models, and it says GAR isn't its preferred model for defensemen because it uses on-ice shooting percentage.

Comparing players across positions

WAR puts a defenseman and a winger on one scale, and the components show you how each got there. Seider's 26.3 goals above replacement split almost evenly between even-strength offense and defense (10.9 and 10.4), with 4.4 more on the power play. DeBrincat's 21.4 leaned on even-strength offense, which supplied 14.1 of it. Different jobs, both near the top of the league.

Separating results from chances

Sometimes GAR and xGAR disagree. The gap is the story. Axel Sandin-Pellikka finished his rookie season at minus-1.9 GAR and plus-3.3 xGAR, so the goals said below replacement while the chances said above it, if still short of the median defenseman. If you only looked at one model you'd have half the picture, and that's the kind of split the W-Value's blend is built to absorb.

Replacement level is a choice

Evolving-Hockey draws the line at a team's top 13 forwards and top seven defensemen by ice time, and Bacon's model uses a similar cut. Move the line and every player's number moves with it. His play didn't change.

One season is a small sample

A year-to-year correlation near 0.5 means a lot of any season's number won't come back, so read a single season's WAR as a record of what happened and give it the half-win margin Evolving-Hockey suggests.

The models don't agree

Evolving-Hockey, HockeyStats and The Athletic each build their own components, choose their own replacement line and convert goals to wins their own way. The same player can carry a different WAR on each site, so compare numbers from one model at a time.

It can't explain itself

WAR tells you how much value a player added, and you need the components to see where it came from. Even those come out of regressions, so they describe what the model credited him with over a season and say nothing about a given shift.

Do goalies get a WAR?

Evolving-Hockey rates goalies separately, on goals above average and replacement built from the unblocked shots they face. This site's goalie rating uses MoneyPuck's goals saved above expected instead, scaled 0 to 100 among goalies with 600 or more minutes.