Andrew Berkshire: Ranking the Top 20 Defencemen in the NHL, by the Numbers

[Human]

The list is backed up by stats...

That being said every time they made Niskanen play first pair minutes because of injuries back when he was with the Pens, he was disaster...

 

[CDN24]

Shot locations matter for individual shots. But when you are looking at a population of 1000 shots, it's unlikely that one player only took shots from the blue like and the other from the slot. If it were true it would show up in their shooting %. No one has ever claimed that 1 break away is equal to a shot from centre ice - other than you. Have you ever seen a box score where one team had 50 shots and another team had 10, but the 50 were all from the blue line while the 10 were all breakaways? Me neither.

Same with quality of competition.One shif vs Crosby compared to one shift vs Glass? Obviously there is a difference. But have you ever seen a season worth of play where one first pairing Dman played their entire TOI vs Crosby but the Other first pairing Dman playd against Glass? Me neither. I have however seen one player play 50% of their ice time with Kopitar while another player played 50% of their ice time with Sean Monahan. Which one do you think had an easier time?

Faceoff locations have minimal impact over a course of a season. That has been shown a countless number of times. Obviously it takes more than one play to see the relationship.



What does "Corsi rates them the same" mean? That's not even true or the reason why corsi is a valuable indicator. Where did you get this bizzare idea?

Corsi works, because over 10 games or more, a team takes hundreds of shot attempts and conceded hundreds of shot attempts. Some are high danger, some are low. The distribution high danger shots/ medium danger shots/ low danger between teams is not nearly as different as their #of shot attempts taken or faced though. Like I said earlier, have you ever seen one team continuesly have breakaways and the other shoot from center ice? It's more likely that if you gave up more total shots, you game up more shots from the slot as well.

I would argue that this is where Corsi fails for rating individual players especially D-men. It does not measure the impact a player has on that shot distribution and assumes it is the same for all players, it is not. The low danger shots and even the medium danger ones don't matter for the most part, Todays goalies stop those. It is the high danger shots that matter, reduce a those by a % point or two and you will reduce goals against. By the same token you see some elite d-men with brutal on ice sav% from their goalies relative to their teammates - hello Erik K. Why? The eye test tells me because he takes unnecessary risk to generate offense that leads to odd man rushes the other way.

Again these stats tell you what (which is damm useful) but not the why (which is more useful)

 

[Bourne Endeavor]

My take away from this list is Karlsson may actually be underrated on these forums. It's borderline laughable how any statistic that gives him a semblance of defensive credit is seen as a joke. Hate to break it to people but transitional play is a massive part of being a top tier defenseman. And Karlsson happens to be phenomenal at it.

 

[Beville]

Weird how, Burns has a 10.0 increase over Karlsson for Offense?

Yet, EK got more points? Which would surely imply regardless that he's more of an offensive player?

I dunno, weird.

 

[fiveonfive]

I love the fact when I blast corsi as an individual stat the stat heads come out and tell me why I'm wrong and use team stats to back it up.

You love the fact that when you say something wrong people come and correct you? That's nice

If you're good a not allowing quality scoring chances, you're goalie is going to have better stats than when you allow a lot of high quality scoring chances. To me, this is common sense.

What's better, giving up 120 low percentage shots against, or 105 higher percentage shots against?

It is common sense, but that is not how the game actually plays out. There is tons of data out there to show it. Players influence shot volume much better, than save percentage. People also have developed scoring chance models and expected goal models based on where shots are coming from - they show a very similar picture that corsi shows.

Obviously in your example, assuming the difference between low and high percentage shots is high enough, it might be better to give up more low percentage shots than 105 higher percentage shots. But that's not how the game ever turns out - we have data on that.

Correlation does not equal causation. If you could show me that Corsi causes winning that would mean something. As it is maybe winning causes good Corsi.

I actually like Corsi as a team stat as there are no quality of competition biases. On the individual level it tells you what happened but not why.

Good coaches use their better defensive d-man against the forwards who can score not just the guys who shoot a lot. It does not balance out, coaches are not rolling 4 lines and 3 d pairs.

Back to this Berkshire list while not based only on Corsi it is based on stats that do not pass the smell test. If this was so predictive of team success one would expect some of these top guys to have helped their teams a little bit more.

Hockey is a team game, winning teams are usually greater than the sum of their parts. Really it requires players who complement each other. A defense full of Karlssons and Burns' will not win a cup. You need a few Vlasics and Webers mixed in to provide the balance.

Shot attempts materialize into goals, which materializes into outscoring, which is what winning is. How can wining cause more shot attempts? That is backwards.

Exactly, corsi tells you what happened and who was on the ice. Results are results.

Better coaches use their best defensemen to shut down the forwards that score a lot, while the coach of the forwards tries to play them away the other team's best defensemen. Also, every coach does it. So at the end, all top pairing dmen play similar competition.

It’s easy to say that a team can’t win with only Karlssons and Burns, because there is only two players in the world as good as them and there can never be a team made up of 6 of them. But yes, Weber and Vlasic are both top pairing players, who can help you win against teams that don’t have as many top pairing players.

 

[CDN24]

My take away from this list is Karlsson may actually be underrated on these forums. It's borderline laughable how any statistic that gives him a semblance of defensive credit is seen as a joke. Hate to break it to people but transitional play is a massive part of being a top tier defenseman. And Karlsson happens to be phenomenal at it.

Ensuring that your team scores more goals than it lets in while you are on the ice is also a massive part of being a top tier defenseman. And Karlsson happens to be pretty ordinary at it.

Last year EK 5-5 was 2.63GF60 and 2.73GA60 for a minus .10
Girodano 2.52GF60 against 2.80 for a net minus as well.

 

[fiveonfive]

I would argue that this is where Corsi fails for rating individual players especially D-men. It does not measure the impact a player has on that shot distribution and assumes it is the same for all players, it is not. The low danger shots and even the medium danger ones don't matter for the most part, Todays goalies stop those. It is the high danger shots that matter, reduce a those by a % point or two and you will reduce goals against. By the same token you see some elite d-men with brutal on ice sav% from their goalies relative to their teammates - hello Erik K. Why? The eye test tells me because he takes unnecessary risk to generate offense that leads to odd man rushes the other way.

Again these stats tell you what (which is damm useful) but not the why (which is more useful)

But if that were the case, then there would be a strong relationship between on ice save percentage and individual players. But there isn't. Why is that? Why does Karlsson's expected GA stats or scoring chance against and rel scoring chance against (which factor in shot location) do not reflect what you are saying?

And it's certainly debatable weather Karlsson's risk are unnecessary. His team is at 42% CF% when he isn't on the ice. Do you think it's possible to win playing that type of hockey?

Ensuring that your team scores more goals than it lets in while you are on the ice is also a massive part of being a top tier defenseman. And Karlsson happens to be pretty ordinary at it.

Last year EK 5-5 was 2.63GF60 and 2.73GA60 for a minus .10
Girodano 2.52GF60 against 2.80 for a net minus as well.

Should we not consider the team impacts on that though? Or are we gonna pretend like teams that are bottom in the league without these players are the same as teams that are top of their league without their first Dman on the ice?

 

[Billy Kvcmu]

Now this list is hilarious even though it's backed up by analytics

 

[6 Karlsson 5]

I love the fact when I blast corsi as an individual stat the stat heads come out and tell me why I'm wrong and use team stats to back it up.

I love the fact that you ducked my question and can't show why your method is any better
And how would you like us to use individual stats? All you are going to do is disagree with the list because
a) eye test
b) your stats are better ( so explain, show why they are better )

So it seems pretty tough to do that

 

[Juicy Pop]

The stats are sound but the assumptions behind the composite scores are suspect.

 

[6 Karlsson 5]

Correlation does not equal causation. If you could show me that Corsi causes winning that would mean something. As it is maybe winning causes good Corsi.

I actually like Corsi as a team stat as there are no quality of competition biases. On the individual level it tells you what happened but not why.

Good coaches use their better defensive d-man against the forwards who can score not just the guys who shoot a lot. It does not balance out, coaches are not rolling 4 lines and 3 d pairs.

Back to this Berkshire list while not based only on Corsi it is based on stats that do not pass the smell test. If this was so predictive of team success one would expect some of these top guys to have helped their teams a little bit more.

Hockey is a team game, winning teams are usually greater than the sum of their parts. Really it requires players who complement each other. A defense full of Karlssons and Burns' will not win a cup. You need a few Vlasics and Webers mixed in to provide the balance.

I'll admit it is tough to show causation, how would you propose that would be done ?
I think logical reasoning is the best solution ( or that I have thought of )
Team who attempt to shoot the puck more...have he puck more...teams who have the puck more ..... Probbaly have better systems, players, choaches....combine all that together and you get teams that win more

I'm all about proof, so I'll admit it isn't the greatest argument for causation

 

[Stuzchuk]

I'm not too sure about that one...

 

[CDN24]

But if that were the case, then there would be a strong relationship between on ice save percentage and individual players. But there isn't. Why is that? Why does Karlsson's expected GA stats or scoring chance against and rel scoring chance against (which factor in shot location) do not reflect what you are saying?

And it's certainly debatable weather Karlsson's risk are unnecessary. His team is at 42% CF% when he isn't on the ice. Do you think it's possible to win playing that type of hockey?




Should we not consider the team impacts on that though? Or are we gonna pretend like teams that are bottom in the league without these players are the same as teams that are top of their league without their first Dman on the ice?

Thing is KEK65's case most of his teammates were better than him GF60-GA60 at 5 on 5

Wideman 2.35vs 2.06 so +.29
Cowen 2.27 vs 2.04 so + .23
Ceci 1.96 vs 1.85 or +.11
Methot 2.56 vs 2.46 or +.10
weircoch 1.99 vs 1.92 or +.07
Karlsson 2.63 vs 2.73 or -0.10
Boro 1.44 vs 2.08 0r -0.64

Phaneuf is worse too but most of his numbers were accumulated on a worse team Toronto.

So at the end of the day at 5 on 5 the only guy he was better than on a net basis is Boro.

Giordano is not that bad relative to his teammates but he was still 4th with JJ, brodie and ENgelland having better numbers.

Quantity of chances against is important but one should not ignore those that go in especially if they seem to go in way more often when certain guys are on the ice.

 

[CDN24]

I'll admit it is tough to show causation, how would you propose that would be done ?
I think logical reasoning is the best solution ( or that I have thought of )
Team who attempt to shoot the puck more...have he puck more...teams who have the puck more ..... Probbaly have better systems, players, choaches....combine all that together and you get teams that win more

I'm all about proof, so I'll admit it isn't the greatest argument for causation

No clue, when I figure it out some team will pay me the big bucks hopefully.

I like the concept of advanced stats, understand it fairly well but I am of the opinion that they make a great tool but a poor master. They tell us what happened but we still need they why to make a useful analysis and decisions. That is no different than it was 30 years ago when the Leafs ran Larry murphy out of town and he went on to det to be a valuable piece. The team that can identify the why (correctly) can still make a killing on trades for what someone considers spare parts or worse.

 

[projexns]

Honestly, if I were to rate defensemen in the defensive zone, I'd use.

Quality of competition measured by number of points from forwards, not by corsi.

Because to me, these are the most important things a defenseman does in the defensive zone.

Quality of competition measured by number of points from forwards, not by corsi.
This is the most important bit. Corsi based QOC stats are weak in my opinion as everyone can make a shot attempt but the truly quality competition are guys like Mike bossy who were burying 1 in 5 of those shots.

Here’s an interesting article on the subject, with emphasis on the 3rd and 4th graphs and their explanations:

https://hockey-graphs.com/2016/03/3...arincin-and-quality-of-competition/#more-8588

 

[triggrman]

You love the fact that when you say something wrong people come and correct you? That's nice

Hasn't happened yet. I've seen more opinions and more team stats but not one slice of evidence that says corsi is a good way to measure defensive play.

I still think the best way to keep scoring down is to keep shots and pressure to the outside, in lower scoring areas.

It is common sense, but that is not how the game actually plays out. There is tons of data out there to show it. Players influence shot volume much better, than save percentage. People also have developed scoring chance models and expected goal models based on where shots are coming from - they show a very similar picture that corsi shows.

Obviously in your example, assuming the difference between low and high percentage shots is high enough, it might be better to give up more low percentage shots than 105 higher percentage shots. But that's not how the game ever turns out - we have data on that.
s

Show me this data proof, for individuals, not team stats.

 

[capebretoncanadien]

Cmon Berkshire do you even watch these guys play? That list is an abomination.

 

[CDN24]

Here’s an interesting article on the subject, with emphasis on the 3rd and 4th graphs and their explanations:

https://hockey-graphs.com/2016/03/3...arincin-and-quality-of-competition/#more-8588

Thanks. I like it. Essentially the problem as I see it in using some variation of Corsi to measure QOC is that the variation in Corsi and its variants is small relative to the actual variation between the quality of players. I can comfortably say that a Crosby Ovi or seguin is more than 2 times as good as a marginal 4th liner. Corsi based QOC will not show that. In fact the marginal 4th liner may be shown as just as good as his Corsi stats were built for the most part playing against equally inept 4th liners.

 

[triggrman]

I love the fact that you ducked my question and can't show why your method is any better
And how would you like us to use individual stats? All you are going to do is disagree with the list because
a) eye test
b) your stats are better ( so explain, show why they are better )

So it seems pretty tough to do that

No, I have no evidence of any of this, it's simple the way I view defense.

I never said my way was perfect, but those are the things I'd like to see measured.

Maybe if I have time (doubtful) I'll do it.

 

[SmellOfVictory]

As a stat lover, I do think this list likely has some flaws; particularly the use of team-relative stats. Playing on a team that has a bunch of crappy players doesn't intrinsically seem like it would make a difference as to how well an individual plays, so long as those players aren't playing with said individual.

Take the contentious Gio vs Doughty ranking: Giordano has played on a team that has a terrrrrrrrrible bottom 3 dmen, a 3rd dman who is quite good but definitely a work in progress defensively, and a bottom 6 that could make a grown man cry with pity. When he's on the ice, he plays mostly with Brodie (who is elite), Backlund/Frolik (excellent two-way players) and the Flames' top line (not great defensively, but they do a good job keeping the puck in the offensive zone when they have it). Doughty has played on a team that is top-to-bottom quite well-balanced, both in forwards and defencemen. Relative to the other Flames players, Gio's stats will likely look better simply because the parts of the team he doesn't often play with suck balls (but I don't see how this would affect his life aside from forcing him to start in the defensive zone more); relative to the other Kings players, Doughty's stats will likely look less impressive simply because overall there are less weaknesses in that lineup.

That's just a lay interpretation of issues I think I see with the stats, but I really do think it'd benefit from different context in the case of some of these measurements.

 

[fiveonfive]

Thing is KEK65's case most of his teammates were better than him GF60-GA60 at 5 on 5

Wideman 2.35vs 2.06 so +.29
Cowen 2.27 vs 2.04 so + .23
Ceci 1.96 vs 1.85 or +.11
Methot 2.56 vs 2.46 or +.10
weircoch 1.99 vs 1.92 or +.07
Karlsson 2.63 vs 2.73 or -0.10
Boro 1.44 vs 2.08 0r -0.64

Phaneuf is worse too but most of his numbers were accumulated on a worse team Toronto.

So at the end of the day at 5 on 5 the only guy he was better than on a net basis is Boro.

Giordano is not that bad relative to his teammates but he was still 4th with JJ, brodie and ENgelland having better numbers.

Quantity of chances against is important but one should not ignore those that go in especially if they seem to go in way more often when certain guys are on the ice.

That's an interesting way to present things, but here is a more complete picture:

Karsson is GF2.63 GA2.73 = -0.10 or 49.0 GF%
Karlsson's TMGF1.85 TMGA1.98 =-.13 or 48.2 GF%

So he improved the team's goal difference by 0.8GF% per 60 minutes in 2015-2016. Also more food for thought, his most common partner, Methot:
Methot with Karlsson, GF 2.84 GA 2.26 so +.48 or 55.7 GF%
Methot away from Karlsson, GF 1.64 GA 3.07 so -1.43 or 34.8 GF%

To say he doesn't provide value in outscoring at 5v5 is pretty bizarre. In addition, the whole issue with volatility, SV%, AND the eye test and all that jazz, which supports the fact that he is a very effective 5v5 player.

 

[CDN24]

That's an interesting way to present things, but here is a more complete picture:

Karsson is GF2.63 GA2.73 = -0.10 or 49.0 GF%
Karlsson's TMGF1.85 TMGA1.98 =-.13 or 48.2 GF%

So he improved the team's goal difference by 0.8GF% per 60 minutes in 2015-2016. Also more food for thought, his most common partner, Methot:
Methot with Karlsson, GF 2.84 GA 2.26 so +.48 or 55.7 GF%
Methot away from Karlsson, GF 1.64 GA 3.07 so -1.43 or 34.8 GF%

To say he doesn't provide value in outscoring at 5v5 is pretty bizarre. In addition, the whole issue with volatility, SV%, AND the eye test and all that jazz, which supports the fact that he is a very effective 5v5 player.

Not that bizarre he doesn't. They give up more than they score when he is on the ice so its hard to see the value he provides in outscoring the opposition. His other dmen are better.

 

[fiveonfive]

Hasn't happened yet. I've seen more opinions and more team stats but not one slice of evidence that says corsi is a good way to measure defensive play.

I still think the best way to keep scoring down is to keep shots and pressure to the outside, in lower scoring areas.


Show me this data proof, for individuals, not team stats.

I'm sorry, but I'm not gonna do your homework for you. You have made claims that Corsi doesn't work and that defense is best played by "keep shots and pressure to the outside, in lower scoring areas" without any data at all. There is hundreds of analytics sites that have ran all sorts of analysis on these things with lots of these topics beaten to death. If you are gonna claim something is trash, at least attempt to understand it first and look at the work that has been done to support/disprove it.

No, I have no evidence of any of this, it's simple the way I view defense.

I never said my way was perfect, but those are the things I'd like to see measured.

Maybe if I have time (doubtful) I'll do it.

So you don't have ANY evidence, but you are confident that those that looked at this are wrong? Sounds like you have a very convincing case.

 

[fiveonfive]

Not that bizarre he doesn't. They give up more than they score when he is on the ice so its hard to see the value he provides in outscoring the opposition. His other dmen are better.

Did you miss the part where they do worse without him? His relStat is positive. What about the year before when he did have a greater than 50% GF and a positive relGF? And also him and Methot being an almost 60 GF% when together when they take on top lines every night when together?

Also, do you actually believe that Wideman, Ceci, Methot, and Cowen are better 5v5 players than Erik Karlsson? You must have misspoke.

 

[LEAFANFORLIFE23]

This is why I hate advanced Stats Matt Niskanen makes the list yet Weber is nowhere to be found.

I can't wait for the advanced fad to die and it will once John Chyka fails in Arizona and he will