A Partial Defense of Oscar Klefbom’s GAR

I’ll get the bad news out of the way early. It is, in my opinion, pretty unlikely that Oscar Klefbom was the league’s most valuable defenceman last season, despite what GAR says.

Boooo! It’s the guy’s birthday. What kind of present is that?!?!

Ok, it’s not all bad news. I decided, based on the twitter uproar over the last few days, due to Klefbom’s Goals Above Replacement rating, to look at Klefbom’s GAR, and see which components seem to jive, and which don’t. What I found was mostly favourable.

I found, as I usually do, that breaking down a complex model into the smaller, more manageable components from which it’s made, usually sheds some light on the strengths and weaknesses of the model.

My conclusion was that it’s a very well built model. That being said, it is imperfect, as all models are. It’s limitations, it would seem, stem from the fact that hockey is a fluid game, and allocating credit to each of the five players on the ice for each on-ice event is understandably incredibly difficult. I find this model to be a valiant attempt at allocating that credit, but, I also realize that there are likely many other factors at play, for which we are currently unable to properly take into account.

I would like to thank the model’s creator, who goes by the twitter handle DTMAboutHeart, for making so much of his data publicly available:

• GAR/WAR 2008-16
• 2017 GAR
• 2017 Expected Plus-Minus

I highly encourage all of you to read the methodology , before citing or critiquing GAR/WAR. But hey, who am I to tell you how to live your life?

Here’s my component by component breakdown of Klefbom’s 2016-17 season:

Even Strength Offense

Klefbom’s offensive contributions at even strength were estimated to be worth 3.8 goals above those of a replacement level defenceman last year. This ranked second among Oilers blueliners, behind Andrej Sekera’s 4.5, and above Adam Larsson’s 3.2.

In the two seasons prior to 2016-17, Klefbom’s contribution to GAR through even strength offence was 2.26 goals above replacement in nearly 1600 even strength minutes. Last year, he played about 1400 even strength minutes.

Klefbom’s Expected Plus-Minus (XPM) contribution for even strength offense was 4.2 goals, meaning that his Box Plus-Minus (BPM) was probably in the 3.4 range. So both metrics generally agreed about his offensive value.

BPM is based on traditional scoring stats. Last year Klefbom was tied for first (with Andrej Sekera) among Oilers d-men, with 20 even strength points, although Sekera played nearly 100 minutes less. When it came to primary points, Klefbom led the blueline, both overall (with 18) and on a per-hour basis (0.77 per 60 mins). In the two years prior to last season, Klefbom’s primary point per hour rate was 0.52.

Conclusion: seems like a fair rating

Power Play Offense

Last season, the GAR model estimated that Klefbom was worth 1.3 goals above replacement on the power play. Now, it’s important to keep in mind that a replacement level power play defenceman is usually a team’s 3rd or 4th best offensive blueliner, rather than a fringe roster player, who would be considered replacement level at evens. That being said, it shouldn’t be surprising that Klefbom had a career PP GAR of 0 prior to last season, in less than 100 minutes with the man advantage from 2014-16.

In 2016-17, with nearly 200 minutes of power play time, Klefbom was the Oilers’ top GAR contributor with the man advantage. His 2.17 primary points per 60 minutes at 5 on 4 was comfortably ahead of the team’s only other regular PP defenceman, Sekera, who averaged 1.41. Sekera’s power play GAR was 0.9, although if we compare total points per hour, Klefbom only wins 4.35 to 4.22, so you can see why it’s fairly close.

Conclusion: seems like a fair rating

Penalties Drawn/ Taken

Penalties taken is one area where Klefbom is particularly fantastic. The guy has 16 career penalty minutes in 189 career games. That’s unheard of for any player, let alone a defenceman.

In 2016-17, Klefbom’s ability to stay out of the box was estimated to be worth 4.2 goals above replacement, while his ability to draw penalties was worth -1.2 GAR. Obviously a player willing to antagonize and take penalties is going to have a natural advantage in drawing penalties, but Klefbom was still a net positive of 3 goals, which is outstanding. Honestly I’m a little surprised that 6 PIM in 82 games wasn’t worth more, but Klefbom is unsurprisingly a huge factor in the Oilers’ ability to have a positive penalty differential.

Last season was par for the course for the Swede, as well. In 90 games, over the previous two seasons, Klefbom’s 10 PIM were worth 6.45 goals in penalties (not) taken, and a net of 3.62 goals, after factoring in the below average penalties drawn.

It’s a little surprising that Klefbom’s ability to stay out of the box was worth more in 2014-16 (10 PIM in 1880 total minutes) compared to 2016-17 (6 PIM in 1800 minutes). Maybe that’s a result of more fringe roster spots going to skill guys, as opposed to tough guys each year.

Conclusion: seems like another fair rating

Even Strength Defence

Well, I obviously got to this one last because it was the source of the most controversy. Klefbom’s defensive GAR was 8.8, the highest among any player in the league. Among blueliners, Marc-Edouard Vlasic, Niklas Hjalmarsson, Mattias Ekholm, and Anton Stralman round out the top five. All of whom would qualify as usual suspects for defensive impact.

So, does the model just love Swedes?

Probably, but let’s look at some stats anyway. First of all, it’s important to remember that the model bases all of its defensive value on defensive XPM, which is based on DTMAboutHeart’s expected goals (XG) model. So, while critiquing Klefbom’s defence using Corsi or on-ice goals may be a useful exercise, it’s not going to help us understand his GAR, because those stats aren’t used to determine it.

To my knowledge, the only xG-type metric that’s currently available for public consumption is Puck IQ’s Dangerous Fenwick For/ Against (DFF%). According to Dangerous Fenwick Against, Klefbom was tied for fourth, with his usual partner Adam Larsson, among the six Oilers defencemen with at least 600 even strength minutes played.

Now, it’s worth mentioning that the Klefbom/ Larsson duo often played the opponents’ top competition, a factor that the GAR model attempts to account for, with a multivariate linear regression. It’s also worth mentioning that the GAR model uses prior seasons’ data, in order to establish a baseline for Quality of Teammates (QoT) and Quality of Competition (QoT), as well as to mitigate against collinearity issues.

With this in mind, I established a hypothesis as to how Klefbom’s defence was given so much credit.

• The xG model GAR is based on liked him better than publicly available metrics.
• The multivariate linear regression thinks he played very tough minutes.
• Larsson had a legitimate breakout year, and the model allocated most, if not all of the credit to Klefbom, because they rarely played apart.

The first factor is impossible to determine without access to that data, although I think I recall better on-ice numbers for Klefbom and Larsson being posted on Twitter/ on Corsica, when that iteration on xG was publicly available.

Again the second factor isn’t something I can just prove or debunk with other data. That’s the drawback of regression analysis. Assuming it’s based on sound logic, a regression model will be more accurate than simply eyeballing numbers, or using a crude formula. But, on the other hand, it’s tougher to see where something went wrong, if it did. For what it’s worth, the Oilers’ regular defencemen all played against pretty similar competition by opponents’ goals for per hour. Furthermore, Puck IQ’s WoodMoney numbers have Klefbom facing off with elites 34% of the time, and ‘Gritensity’ (bad) 28% of the time, which isn’t that tough compared to other top pairings.

I can actually address the third factor to a certain extent. In his 4 qualified seasons with New Jersey, Adam Larsson had a defensive GAR impact of 4.35 goals in just over 4000 even strength minutes.

In 2016-17, in roughly one third the ice time, Larsson had a defensice GAR of 2.5. So, the model did give him credit for some improvement.

However, it wasn’t as much credit as Klefbom received. In 90 games between 2014-16, Klefbom’s defensive GAR was 3.12, less than half of 2016-17’s 8.8 in a similar amount of ice time.

It’s not as if Klefbom and Larsson spent no time apart in 2016-17. Together they allowed 55.9 shot-attempts per hour in over 1000 minutes together, while Klefbom allowed 52.4 shot-attempts per hour in nearly 350 minutes apart. His best ‘apart’ split of any skater he played with.

That being said, I don’t think it’s fair to deem Larsson an anchor. His shot-attempt numbers were no worse apart from Klefbom, despite a slightly more defensive role, per zone starts. Furthermore, both players had a significantly worse goals-against rate when they were separated. Perhaps that’s telling, despite the small sample size. Klefbom’s been plagued by a high opponents shooting % for most of his career, while Larsson’s genreally had a low opponents shooting %. Klefbom’s generally been a better shot suppressor though. A combination of his gap control and neutral zone prowess with Larsson’s ability to clear the crease is a nice combination.

Still, none of that really explains Klefbom’s ascension to elite defensive status. He and Larsson were assigned a combined 11.3 defensive goals above replacement last year. A number significantly higher than their combined career totals coming in, despite the fact that neither had career years in either goal or shot suppression relative to their teammates, who received less credit.

So, I’m stumped, does that mean Klefbom’s bad?

Actually no, it doesn’t. Even if he were assigned a replacement level defensive rating, Klefbom would have had a 8.2 GAR just based on his offense and impact of penalty differential last year. Add a defensive GAR of 3, to remain consistent with his career average before last season, and he was still worth about 11.2 goals above replacement last year.

That would have been good enough for 14th in the league, among defencemen. Right below Doughty, right above Josi. Hell, even if Klefbom had a defensive GAR of 0, he would have still been tied for 30th overall in GAR for defencemen.

Klefbom just had a great overall year last season. He was tied for 12th among defensemen in primary points per hour at evens, was the most successful d-man on a top-5 power play, and took only three minor penalties in 82 games! His defensive impact is still very much up for debate, but there’s little doubt in my mind that he’s a number one defenceman.

Happy birthday, Oscar!