Quality of Competition in Hockey (QoC) and Teammates (QoT)

Context stats arrived early in public analytics. By 2014, Behind the Net alone carried three measures of competition and three of teammates, by one Hockey-Graphs writer's count. Rob Vollman's player usage charts put quality of competition on one axis and zone starts on the other, which made deployment something you could see at a glance.

Eric Tulsky changed how they're built. In an August 2012 post at NHL Numbers, the earliest War-on-Ice could find, he measured competition by ice time, and a month later his histograms showed how narrow the spread in competition was across the league. Two things followed, in Garret Hohl's summary at Hockey-Graphs. Ice time became the preferred measure over shot metrics, and the impact of line matching went from important to overrated to "inconsequential."

The follow-up work filled in the edges. A 2014 Hockey-Graphs piece noted that nearly every player's average opponent landed within one shot attempt per 60 minutes of average, with the extremes around two. Hohl's 2015 study found teammate quality had about five times the variance of competition quality, and that a one-point change in either moved a player's own shot share about 1.5 points (up for teammates, down for opponents). In 2016, Ryan Stimson and DTMAboutHeart split both by position and found that what matters depends on the player. For a forward's shot generation, the defensemen he faced and the forwards he played with carried the weight.

Evolving-Hockey publishes QoC and QoT tables built on ice time, RAPM, GAR and expected goals above replacement, with every opponent and teammate weighted by the minutes he shared with the player, and this page uses its even-strength numbers from 2025-26 throughout.

Every version is a weighted average. Pick a measure of quality, look up every opponent a player faced, and weight each one by the share of the player's ice time he spent against him. Evolving-Hockey describes its tables exactly that way. The inputs:

• Every opponent (for QoC) or teammate (for QoT) a player shared the ice with
• The seconds they shared, from the NHL's shift charts
• A quality rating for each of them, such as TOI%, time on ice per game, RAPM, GAR or xGAR per 60
• The game state, usually even strength or five-on-five

Here's a worked example with invented round numbers. A forward spends 600 even-strength minutes against opponents who play 30% of their team's minutes, and 400 against opponents who play 27%. His ice-time QoC is (600 × 30 + 400 × 27) ÷ 1,000, or 28.8%, which is about what the average forward faces. Two lineups that look nothing alike on the bench can produce nearly the same number.

Ice time is the rating Tulsky's work made standard, and it's a blunt one. As Hohl put it in 2015, a 13-minute forward on a contending team doesn't have the same talent as a 13-minute forward on a lottery team, which is why the RAPM, GAR and xGAR versions exist.

QoC measures how hard a player's minutes were, and for almost everyone the answer is close to average. If you line up the 372 forwards with 500 even-strength minutes in Evolving-Hockey's 2025-26 table, their opponents' TOI% ran from 27.23% to 29.86%, with a standard deviation of 0.56 points. That's a narrow band. In minutes, the toughest-faced forward's opponents averaged 14.99 a game and the easiest-faced forward's 13.65.

Teammates spread out more. The same forwards' QoT ran from 26.21% to 33.19%, with a standard deviation of 1.20 points, about 2.2 times as wide as the competition's, or 4.6 times the variance, close to the five times Hohl measured with shot-share ratings in 2015. Rated by xGAR per 60, teammates varied 5.6 times as much as opponents did.

Detroit's numbers follow the league's shape. Dylan Larkin faced the toughest competition on the team (29.71%, sixth among the 372 forwards), and James van Riemsdyk the softest (27.85%, 346th). In minutes, Larkin's opponents averaged 14.9 a game and van Riemsdyk's 13.95. The whole range of Detroit's forward assignments came to less than a minute of opponent ice time per game.

So does QoC matter? Over a full season it matters a little, and you can put a number on it. Hohl's slope says a one-point bump in competition shot share costs a player about 1.5 points of his own, and the league spread in competition is narrow. Teammates move more, because their spread is wider.

Testing the tough-minutes defense

Larkin's competition was the sixth-toughest among regular forwards, and Detroit had 47.2% of the expected goals with him on the ice at five-on-five. Is that the tough-minutes story? It's half of it, since his teammates ranked 41st by the same measure, and Hohl's research says a better linemate helps about as much per point as a better opponent hurts. When you see a top-line center with top-line competition, check who he played with before you grade him on a curve.

Explaining a split within a pair

Moritz Seider and Ben Chiarot faced nearly the same competition last season (29.12% and 28.87%). Their teammates were another story, 15th and 168th among 214 defensemen by the same measure. Detroit had 55.3% of the expected goals with Seider on the ice at five-on-five and 45.2% with Chiarot. The competition column puts them side by side, and the teammates column puts them 153 places apart.

Checking whether a player was sheltered

Evolving-Hockey's table is season-only, so for its Axel Sandin-Pellikka research this site rebuilt his QoC game by game from NHL shift charts, matching EH's five-on-five ice time to the hundredth of a minute. His opponents' xGAR per 60 averaged +0.156 over his first 34 games and +0.253 over his last 34, while his ice time fell. Was he being hidden? His competition got tougher as his minutes shrank, which is the opposite of sheltering.

It's an average of a distribution

A single QoC number blends a few shifts against the other team's top line with a lot of shifts against everybody else. Hohl conceded the mean-only problem in 2017, and argued that splitting opponents into tiers doesn't fix it, since the cutoffs are arbitrary.

The rating is only as good as its input

Ice-time QoC rewards facing players who play a lot, which isn't the same as facing good ones. The RAPM and GAR versions fix that by rating opponents on results, and they carry whatever error those models have.

It describes context and adjusts for nothing

Knowing a player's QoC doesn't tell you how much to move his numbers, only which direction. Regression models like RAPM, which control for teammates, opponents, zone starts and score at once, do the adjusting, and Hohl called them superior to either kind of QoC.

Teammates aren't one thing

Hohl found that a teammate's effect on a player isn't the same for every player, the way competition's is. Stimson and DTMAboutHeart found the useful split runs by position, and a single QoT number hides both.