Is Erik Karlsson a good Defender? Does it matter?

I read an article today discussing whether “role player” still have a role in the NHL today. A significant part of the discussion revolved around Erik Karlsson, whether he is a good defensive player, and essentially whether it mattered. Former offensive NHL defenseman Brian Leetch discussed Karlsson’s play and that overall he contributes everything a team could ask of him.

“It’s a game of mistakes. Something’s going to go wrong out there and that goes for every player,” he continued. “But the amount of things [Karlsson] does right and the amount of things he does to impact his team in a positive way is ideal. That is just such an easy thing to say about someone that’s putting up offensive numbers, but when you watch him play on a regular basis, everything he does is what any team wants a defenseman to do.”

The article continues on with how the league is moving away from defensive role players to players that can play an all round puck possession game. It is an an interesting discussion but it offered no numbers to support the discussion. Being a numbers guy so I decided to investigate Karlsson’s impact and whether (and how much) his offensive ability offset his defensive liability. Here are the Ottawa Senators 5v5 GA/60 statistics over the past 3 seasons (so far this season and previous two).

  • With Karlsson on the ice: 2.60
  • Without Karlsson on the ice: 2.17

Wow. When Karlsson is on the ice Senators opponents have scored at a rate nearly 20% higher than when Karlsson is on the bench. He has a net impact on GA/60 of -0.43 (0.43 goals/60 worse with Karlsson on the ice).

Of course Karlsson generates a lot of offense too. Does his offense make up for his defense?

  • With Karlsson on the ice: 2.67
  • Without Karlsson on the ice: 2.07

Ok, not so bad. The Senators score at a rate 29% higher with Karlsson than without. His net goals for impact is +0.6 (0.6 goals/60 better) which gives him a net goal differential impact is +0.17 goals/60. He is a net positive overall but not a huge amount.

How does this compare to some other top defensemen? Let’s have a look.


Subban has the highest net positive GF60 impact but also the worst negative impact on GA60. Karlsson is next best in GF60 impact but also next worst in GA60 impact. Of all these players Doughty is the only one with a positive impact on goals against, albeit only slightly positive, while Phaneuf was even and Weber only marginally negative.

Ranked by overall impact we have Subban, Keith, Phaneuf, Doughty, Karlsson and Weber.

Now this is a pretty high level analysis and doesn’t fully take into account quality of team mates each play with (only quality of team) but I think offers a good first look at the players impact on their teams performance. There is no doubt about it. Karlsson is an elite offensive talent but it also seems clear that there is a cost defensively. Same for P.K. Subban. Would the Senators and Canadiens be better off if Karlsson and Subban let up a little bit on their offense and focused a little more on defense? We can’t know for sure but I think it is a fair question to ask.


This article has 7 Comments

  1. Interesting analysis as it debunks a bit a the value this new generation of offensive defence man have on the game. This trend may be increasing as teams are looking for more offence as the 3 forwards tend to neutralize themselves. It would be interesting to add a few more people to this analysis …. Say Bobby Orr…or a Paul Coffey it the stats are available. These are considered the standards when talking about offensive defence man.

    Thank you

  2. Sorry if I am a bit ignorant but I am new to the hockey analytics forum. Is there a way to control the “quality” minutes v. top opponents? I imagine that there would be opponent quality effects that would be heterogenous across teams, divisions, etc. A minute played against Oveckin versus a minute played against…Brooks Laich…. would have different importance.

    1. It is generally believed that over the long haul quality of competition largely balances out. There have been many studies that quality of competition is not a significant factor in a players overall performance (quality of teammates is a far more significant factor).

      Now, with that said, I am having some second thoughts about that research and I hope to investigate some new methodologies that will refine how we measure quality of competition. I am not sure where that will lead though so for now we have to assume it won’t be a huge impact.

  3. I would like to see a dynamic model of how players GF60 and GA60 changes based on game state (e.g. tied, trailing, ahead, etc,). Here are the last three seasons of situational 5v5 defending for 4 top defensemen with very different skill sets.

    Karlsson’s GA/60 goes up to 3.31 when up by one goal and 3.51 when up by two or more goals. Similarly, the Canadiens give up 3.89G/60 when up by two or more goals and Subban is on the ice. GA60RelTM consistently rank near the bottom of the teams when leading.

    Kings are excellent when Doughty is on the ice and they are either up by one (2.16) and even better when up by two (1.76). Similarly, Hawks’ with Keith are great in close games (2.07 when up or down 1 goal). In GA60RelTM, they both hover around +/- 0.5 in all situation, performing no worse in this metric than their teammates. Score effects will obviously affect the GA/60 rate in these situations, but when comparing these two offensive defensemen with their more defensively-minded peers, it’s a stark difference.

    I look at a guy like Adam Larsson’s stellar GA numbers (esp. relTM) and see a team that was close to playoff contention despite having one a roster devoid of scoring talent. I also look at Pronger taking the Oilers to the finals in 2006 with Fernando Pisani providing the bulk of the goal scoring. I’m a fan of the work some people have done on GAR/WAR, but it just seems like there’s something we have yet to figure out from an analytics perspective in terms of the value of defensive defensemen as it relates to winning hockey games.

  4. Karlsson is a great offensive defensemen and ya makes up some mistakes pinching and is on the ice for a lot of the goals for Ottawa. But I think one thing you missed in this article, Erik Karlsson plays against teams top lines, so it’s not like he is going to be a perfect “+” player. Good article! Just wanted to put my comment on here and see what you think?

  5. Doesn´t this measurement more show how uneven in skill a teams lines are. For example if a teams line 1 and 2 are much better than line 3 and 4 and one defender plays more with line 1 and 2 he´ll get a larger net impact all else being equal. Then how you match defenders with different lines of relative strength might have a big impact on these numbers. It wouldn´t surprise me if you checked these numbers for all the players in the NHL if some mediocre defenders beat all of these great defenders in net impact. Imagine a team with one top line and three bad lines, the top line will put up much bigger relative numbers than a top line in a team with three great lines.

    It´s very hard to measure individual skills in a team sport like hockey. You can be the greatest defender in the world, if your goalie is bad and your forwards don´t defend well, you´re gonna get punished with many goals against. A guy like Doughty has teammates that are better defenders AND better offensive players than Karlsson has around him. Imagine if Karlsson and Doughty switched teams, Karlsson wouldn´t have close to as many goals against, but he´d probably have pretty similar amount of points with great offensive players around him (a small sample, but check his stats for team Sweden). If Doughty played in Senators, he´d have many more goals against, but I doubt he wouldn´t improve his offensive numbers very much. Doughty wouldn´t put up close to 82 points in Senators.

    1. I wrote about the differences between Rel and RelTM stats here. In brief:

      Rel = On ice – off ice
      RelTM = On-ice – Cumulative WOWY (average of teammates ‘without’ stats)

      As you point out, Rel would make good players look better on an team with little depth and the depth players on a team with a superstar top six would be penalized.

      RelTM is more impacted by the players you play with as it looks at whether you make those players better. If you play with superstars it will be difficult to make them look better even if you are a very good player. That same player playing with average players should have a better RelTM (in theory).

      For most players both Rel and RelTM would be fairly highly correlated as there is a lot of parity and balance in teams right now. I am a little biased towards using RelTM as it is my stat but I do recognize that it may penalize players who play with elite players. That said, in this type of analysis most of these players are given top roles with the teams best players and most are on good teams so I am not sure there would be a huge impact to consider.

      It is difficult to answer how many points Doughty would put up if he played with the Senators in Karlsson’s role where he was allowed to free-wheel offensively with little worry about defensive play. Similarly it is difficult to predict how many points Karlsson would get if he had Doughty’s minutes and role for a more conservative offensive team. They are certainly interesting questions to ponder.

Comments are closed.