clock menu more-arrow no yes mobile

Filed under:

An All About the Jersey Hockey Stat Primer: Teammates and Competition

Continuing the series of primers about hockey stats, this post focuses on teammates and competition. This includes an explanation of why they are not used at a season level, why they are more valuable at a game-to-game level, and a couple of tools used for both with Devils players used as examples to demonstrate these concepts.

NHL: OCT 17 Rangers at Devils
Subban! Zibanejad! Actually relevant within this post!
Photo by Rich Graessle/Icon Sportswire via Getty Images

A lot of hockey stats are focused on the individual player or the whole team. This includes the “basic” ones like goals and points to the more “advanced” ones like Corsi and expected goals. However, while every player contributes to the team, the player does not actually play the game by themselves. They have teammates at forward and defense. They tend to get matched to an opposition’s line or defensive pairing in games. Is this accounted for in stats and analysis of player? In this primer, I will do my best to go through each topic, how they are incorporated, and use Devils examples where applicable. Mostly with P.K. Subban. (Note: I did say in the last post this would include passing. I have decided it is too important to throw in here and it deserves mention elsewhere.)

How Are Competition and Teammates Represented in Stats?

When competition or teammates are brought up, this is usually in reference to their quality. Competition can be short for Quality of Competition, also known as QoC or QualComp. Teammates can be short for Quality of Teammates, also known as QoT or QualTeam. As a stat, both have undergone quite a few changes. Mostly because there has not been much of a consensus on how to measure either.

I first read about QoT and QoC as concepts not long after Corsi broke through to the public sphere. There were two main ways to determine it: One method used a player’s time on ice percentage relative to the team. If a coach uses one player a lot, then we assume that the player represents higher quality than their teammates. And playing against another team’s most used players represents higher quality than other players on the opposition. The other method was to use a player’s Corsi value (and later, Corsi For percentage) as the mark of quality for teammates and opponents alike. The actual value of QoC and QoT does not have a unit; it is usually just a number that makes sense when you compare it with others. Higher numbers for either means the player had faced or played with better quality players. These stats would be primarily calculated for 5-on-5 or even strength situations as special teams stick to primarily offensive or defensive players, depending on the team.

(Quick aside: This was back around the late 2000s, when some teams like New Jersey still had dedicated checking lines and match-ups were not necessarily power for power ones like first line versus first pairing that we have seen more often. That may have also influenced the thinking back then.)

Both were imperfect methods. Take the 2019-20 Devils for example. Specifically, defenseman P.K. Subban. Subban finished second on the team in 5-on-5 ice time with 1,157:00 per Natural Stat Trick. Prior to his trade, defenseman Sami Vatanen finished 13th in that same category with 727:31 for New Jersey in 2019-20. By the Time on Ice method, this would mean whoever played with Subban would get a higher value towards their QoT than if they played with Vatanen. This also means whoever played against Subban would get a higher value towards their QoC than if they played against Vatanen. However, Vatanen’s CF% last season was 47.78%, which was a bit better than Subban’s 47.36%. A QoT or QoC value based on Corsi would favor Vatanen as higher quality as a teammate or opponent than Subban. And neither would be seen as particularly high quality as no Devil finished last season with a CF% above 49.61% (owned by Blake Coleman).

Which is more accurate? I cannot tell you. We did nothing to modify the stats. But the ice time method’s assumption of more ice time equals better player does not really hold. Would Subban even play an average of 17 minutes of 5-on-5 ice time per game with those 5-on-5 stats if he was not on the Devils? Given that the one Devil who played more than Subban last season was another right-sided defenseman in Damon Severson, is Subban even really a top-pairing defender as implied by his ice time amount? There are doubts. However, the Corsi method undercuts teams like the Devils from the get-go. Again, no Devils defenseman finished even above the break even mark in CF%. Does this mean the Devils did not have a top-pairing? Maybe one that is not worthy to be seen as one, but they had one nonetheless.

Application and Sample Size Issues

A bigger issue with the QoC and QoT stats is their application. There really is not one I am aware of that uses it to directly adjust a stat. Even if we decide on one method or another, both have been used as values for context. Unless I am unaware of one, there really is no method that incorporates QoC or QoT as values to adjust ones stats in a calculation. You can use it to create bins of data and then present a player’s stats when they were on the ice against certain players or with certain ones. But, really, they are something for the reader to keep in mind. For a blogger like me, the benefit of listing it would mean just stating it and then including some line about how Player X faced tougher competition than Player Y, so that should ease some of the concerns about Player X having crummier stats than Player Y. Can I actually adjust the stats for either and have it make sense? Not really.

The even bigger, and perhaps the biggest, issue with QoC and QoT is their sample size. Teams do not play each other often enough to get a lot of ice time against a common opponent. You would see a division-based opponent at most five times in a season and it is not a guarantee you will see a defenseman matched up with a forward for very much time in each game. Check out P.K. Subban’s list of on-ice stats against opponent players at Natural Stat Trick. The most common opponent he saw in 5-on-5 last season was goaltender Jonathan Bernier for 49:12. The most common opponent who was a skater was Zdeno Chara for just 29:15. Players see each other so little over the course of a season such that even if they had a bad outing against a particular opponent, it alone will not do much to impact Subban’s stats. This also means Subban’s far-from-impressive 5-on-5 stats is a result of a lot of far-from-impressive nights in 5-on-5 hockey more than just struggling against a particular opponent.

For teammates, there is potentially more ice time together. Teams can and do change over the course of a season, but even last season’s Devils team had most of the players at the end of the season start with the team in that season. And coaches tend to stick with lines and pairings after wins and/or successful nights. Even then, the amount of ice time is not as much as you may think. Sticking with Subban, his most common teammate was Andy Greene. They played 420:06 together. This also means Subban played more minutes in total (but less individually) with other Devils than Greene. While the Greene-Subban pairing would have more of an impact on Subban’s 5-on-5 stat line, it is not the sole driver. (Aside: That said, this does suggest that a season of Subban without Greene may help Subban have a better season. Maybe. Possibly? Hopefully?)

What this means is that both QoC and QoT are not really used to analyze how a player did in a season. If Subban does not have a good CF% value among other stats - and he does not - then it is not really the fault of his teammates and definitely not the fault of the individual competition he faced. Over a whole season, Subban would - and did - have a lot of different teammate combinations and even more opponent match-ups such that even the common ones are not representative of the whole. In other words, the question of the quality of his teammates and especially opponents does not answer a lot for Subban’s or anyone else’s performance in a season.

This is why if you are brand new to this and you go to Natural Stat Trick or Evolving Hockey or HockeyViz other sites, you may not even see QoT or QoC represented or featured as a stat on its own.

But Don’t Coaches Specifically Match Players to Opponents and Set Lines and Pairings? So Doesn’t That Matter?

They do. What I just wrote explains why QoT and especially QoC are not used to evaluate seasons or players over a long period of time. I do think it has a lot more value in smaller groups, such as in a single game. So, yes, it does matter at that level.

If you read a recap by me for the last several years, then you will note that I do take time to highlight specific match-ups and their results in a game. Natural Stat Trick’s game pages include on-ice stats for each player. There are even sections where you can filter out players and you can see the on-ice stats when the player was on the ice against different opponents and with different teammates. Here’s the game page from the last Devils’ win of 2019-20, a 6-4 win over Our Hated Rivals. I can go to the Linemates section and see the following (this is just a snip of the screenshot)

Subban’s linemates from March 7, 2020, a 6-4 Devils win over Our Hated Rivals
Subban’s linemates from March 7, 2020, a 6-4 Devils win over Our Hated Rivals
Natural Stat Trick

From this, I can tell you the following:

  • Subban primarily played with Mirco Mueller in this game in 5-on-5 hockey. Given that it was for 17:01, it was the majority of their total ice time too.
  • The Mueller-Subban pairing did not go particularly well. They were nearly even in shooting attempts and neither saw a goal against. However, the Devils were out shot 4-7 when they were together and the expected goals values tells us Our Hated Rivals took more dangerous attempts than the Devils did.
  • Subban (and Mueller) primarily played behind Hischier, Palmieri, and Wood with little success.
  • While it was just for under six minutes, the time with Michael McLeod, Kevin Rooney, and John Haden went badly. It was not as costly or dangerous, but the Devils certainly were not attacking either.
  • The few shifts with Zacha, Gusev, Bratt, Hughes, and Zajac went much better.

Over a whole season, these details may get lost. But this is valuable for determining whether Alain Nasreddine, who was the team’s interim coach for this game, made the right decision to pair up Subban with Mueller and the right choices of which lines to play with. Even with these small amounts of times, a few successful shifts with some may be signs to give them more time together in a future game. This is information that is useful for the next game, too. Is this an acceptable enough result to keep these two together after a win? If so, can they be behind a different forward line? If these combinations have to stay in place to let the other, possibly more successful combinations remain in place, then is this good enough to at least not harm the team? These are questions we can answer seeing these numbers. At a season-level, knowing the QoC or QoT of Subban, Mueller, or anyone else would not really get into that.

Game stats for opponents also yield similar information for the benefit of evaluating the game that was just played and what can be learned for the next game. Here are the on ice stats when Subban played against a certain member of Our Hated Rivals:

Subban’s opponents from March 7, 2020, a 6-4 Devils win over Our Hated Rivals
Subban’s opponents from March 7, 2020, a 6-4 Devils win over Our Hated Rivals
Natural Stat Trick

There is plenty we can immediately see here as well:

  • The game was in Manhattan, so Subban and Mueller saw a lot of Adam Fox and Mika Zibanejad. Our Hated Rivals wanted that match up and you can see why they kept with it for over half of the game Subban played in. When Fox was on the ice, the Devils were pushed back. While the individual attempts and shots were more favorable, there was a goal against them - same with Fox.
  • However, Subban and the Devils did much better when Trouba or Smith were on the ice. They made those two defenders have to defend and the run of play was much better.
  • Surprisingly, Subban also saw plenty of time against Artemi Panarin and the Devils were better in that match-up too.
  • Also surprisingly, Subban and the Devils struggled against the depth of Our Hated Rivals, including 4:12 of Julian Gauthier that did not go all that well. I wonder if this more of a function of the Devils’ fourth line losing in general than anything with Subban.

As this was a road game, we know the Devils did not get the last change and so they did not always get the match-ups they want. That said, we can see it was not all bad across the board for Subban and Mueller. The Devils out-attempting and out-shooting Panarin, Smith, and Trouba for over five minutes each is good. However, we can see that the most common opponents did provide some pain. Fox versus Subban did not go the Devils way and while Zibanejad did not dominate the Devils in the run of play, he (and Fox) were present for a score against them. This information could be used to dig a little deeper to see if there were any common traits about the players Subban struggled against in match-ups as well as those he did well against. It is also recommended, especially for the depth players, to see how other Devils did. It is a bit cumbersome as it is one-to-one comparisons, but this is information that can help show how a player did in a game while acknowledging who they had to play against the most.

Again, the concepts of competition and teammates are not particularly useful to look at a player’s performance overall. For game-to-game evaluations and determining changes to be made, these are valuable concepts.

Right, but What If I Want to Know How a Devil Did Against Tough Competition Anyway?

There is a method for this that I have used here and there: WoodMoney. This is the method of defining competition by three bins - elite, medium, and gritensity for lesser players - and organizing a player’s on-ice data against each. This is meant to estimate the quality of competition and how the player performed with respect to it. It was created by Woodguy and G-Money and the full methodology is explained here. The method is kept up to date Woodguy’s and G-Money’s stat site, PuckIQ.

It is not a popular method since it came to fruition when the discussion among analytics went against the idea of binning and quality of competition as a whole. But the question asked in bold is indeed a question asked by fans. People want to know it, so the WoodMoney method is a way to answer it.

Everyone gets at least a little competition against all three levels, but some definitely get tougher match-ups than others. For the 2019-20 Devils, the elite competition mostly saw Andy Greene per PuckIQ. Greene played 41.9% of his ice time against this defined top tier of opponents. When he was out on the ice against them, the Devils had a CF% of 41.6%. The top tier made Greene and the Devils play a lot of defense in those match-ups. Against the lesser two groups, the Devils’ CF% was better with Greene. But since he was used against the toughest of the three tiers, Greene and the Devils suffered. In fact, no Devil put up a CF% over 50% against the elite tier except for the nearly 14 minutes Frederik Claesson had against them (it was 14 CF to 13 CA, it was not by much). Severson came close, but he missed out on breaking even.

I find that this method is good to answer specific questions like that one. However, it is not particularly useful in determining who did well over a whole season, who the Devils should or should not sign or re-sign, and so forth.

What About Linemates and Defensive Partners? How Do I Know How They Did?

There are a few more tools than that which have been more popular and acknowledged. While common teammates may not be so common over a season, players definitely play with them more often than opponents. Between QoT and QoC, QoT is much more meaningful. And there is a more popular and common way to measure how two or more teammates performed together: With or Without You.

This is a concept I first saw Tyler Dellow use, but I know it predated him a bit. It was one of the hallmarks of David Johnson’s Hockey Analysis site before he expanded on it years later. The method is right in his name. You take a player and you list the on-ice stats (usually in 5-on-5 play) for the player with a teammate, the player without that teammate, and the teammate without the player. The stats can include Corsi, Fenwick, shots on net, goals, expected goals, and scoring chances. No matter the stat, the data is filtered between those three states by ice time together and apart. Through this, we can see who benefitted who as well as how much ice time the player shared with the teammate.

Natural Stat Trick easily has this under Teammates for any player’s page. Here is the page for Subban and his common teammates last season. His most common partner was Andy Greene at 420:06 of ice time last season. When they were together, the Devils had a CF% of 44.86%. That is bad. What makes it worse is that when Subban was with anyone else, the Devils had a CF% of 50.96%; and the Devils had a CF% of 45.5% when Greene was with anyone else. This strongly suggests two things: The Greene-Subban pairing was a bad one and Greene likely dragged down the pairing more than Subban did. I do understand that I only used CF%, but you can make a similar conclusion for all of the other stats included except for Goals or GF%. The With or Without You method is really good when the difference is clear between players together and players apart.

There are drawbacks. First, sample size of ice time does become a real issue as you go deeper into a player’s teammate list. Subban’s 161 minutes with Butcher may not be as representative of either of their true performance compared with Subban’s 420 minutes with Greene. Second, this is only comparing two players. In 5-on-5 hockey, a player has five other teammates including the goalie. This method focuses on just two. Third, when the difference is not so large, the analysis does not really tell you too much. We cannot do much about the third other than to be careful to not jump to large conclusions when we look at the stats. We cannot do much about the first. But the second drawback was addressed when Johnson expanded the concept to five players at Puckalytics, an updated site (I think) from Hockey Analysis. While Johnson has moved to the NHL world, Natural Stat Trick does the same thing through their Line Tool.

There, you can select up to five players and see how the team’s on-ice stats changed with the various combinations. From the game I chose earlier in this post, I noticed that Subban’s most common partner was Mirco Mueller and his second most common skater teammate was Palmieri. I can use Natural Stat Trick’s Line Tool to see how much these three played together and what happened on the ice with and without one or two of them. Again, other stats are used but for the sake of simplicity, I will show the Corsi only.

Subban, Mueller, and Palmieri together and without each other by Corsi
Subban, Mueller, and Palmieri together and without each other by Corsi
Natural Stat Trick - Line Tool

So the combination of Subban, Mueller, and Palmieri did exist but only for just 62 minutes last season. At least by Corsi, it was not a particularly good group. What is more interesting are the splits when you take certain players out. This shows how poor Mueller was in general as when you look at the stats without one or both of Palmieri and Subban. While Palmieri and Subban on their own were not that good, the two did play about 254 minutes together and actually broke even in CF%. That is impressive on a bad 2019-20 Devils team. That is another takeaway from a tool like this: it may behoove the incoming coaching staff to have Subban behind Palmieri. Or at least try it.

The same issues with sample size for ice time still apply here. Even if Subban, Mueller, and Palmieri combined for a mythically high CF% of 90%, 62 minutes is not a lot of time to determine whether a line or a pairing or a unit should stay together. However, on a game-to-game basis or even a small group of games (5-10), this method can definitely work to see whether a combination of teammates is working or not. Coaches and teams do not have the benefit of waiting until a larger sample size is met. They have to make adjustments and lineup decisions regularly for a variety of reasons (performances, injuries, etc.). This can at least justify whether something tried out for one game should continue beyond whether the team won the game - something that may have had nothing to do with the particular forward line or defenseman pairing.

This is how I would look for value among teammates. While not perfect, it is a way to identify which partnerships are working, which are not, and by what means. If you prefer a more visual way, then Dr. Micah Blake McCurdy of HockeyViz has a whole bunch of graphs that present a lot of the similar results you would get from NST.

What’s Next & Your Take

This is tricky concept in part that there really is not a widely used stat used for teammates or competition. And the understanding that it is more valuable within a game as opposed to a season does set both apart from the other stats discussed in this primer so far. Both matter, teammates more so than competition, but it may not be as much as what was thought a decade ago and it is not entirely ignorable like it may seem today. They are both contextual.

Passing deserves a larger section and maybe its own post. But the next one will be about the goaltenders. Some of it will be more of a refresher from previous posts and others will be newer ones that have come out and gained in popularity over the past few years.

In the meantime, I would like to know what you think and what you have learned about teammates and competition. Please provide any further questions about these contextual concepts in the comments. Thank you for reading.