Speculation: Caps Roster General Discussion (Coaching/FAs/Cap/Lines etc) - 2022-23 Season Part 3: Drop the puck!

[twabby]

Are they really? Just based on what gets posted on twitter whenever a player signs/gets traded, I can't remember ever seeing the models differ significantly in their evaluation of a player. Either they converge because they're all perfect or because they all use similar inputs and methodology and thus are all subject to the same issues and biases. I know which one I'd put my money on.

They are all restricted to the same dataset as far as I'm aware: the NHL's real-time scoring system (RTSS) data that was introduced in the 2007-08 season. It's why a lot of these databases and models have 07-08 as the starting point. That's when they started tracking location events such as shots, and that's also when shift information was available. Before that the only data provided publicly were the simply boxcar counting stats.

I think these models all share a similarish methodology in that they are trying to regress for the outputs from the RTSS data. In other words, given the entirety of the output data that we have (all of the shots, penalties, goals, hits, blocks, etc.), what values for the inputs (e.g. player quality) explain these data the best? Sure it's possible that Jack Johnson has just been getting incredibly unlucky year after year and that's why his on-ice results have been so bad. But it's incredibly unlikely. Similarly, it's possible that Alex Ovechkin has seen a ton of shots, chances, and goals go in against his team because he's just had rotten luck for 7 seasons straight. But it seems the more likely explanation is that he's just a hugely negative impactful player defensively.

 

[YippieKaey]

They are all restricted to the same dataset as far as I'm aware: the NHL's real-time scoring system (RTSS) data that was introduced in the 2007-08 season. It's why a lot of these databases and models have 07-08 as the starting point. That's when they started tracking location events such as shots, and that's also when shift information was available. Before that the only data provided publicly were the simply boxcar counting stats.

I think these models all share a similarish methodology in that they are trying to regress for the outputs from the RTSS data. In other words, given the entirety of the output data that we have (all of the shots, penalties, goals, hits, blocks, etc.), what values for the inputs (e.g. player quality) explain these data the best? Sure it's possible that Jack Johnson has just been getting incredibly unlucky year after year and that's why his on-ice results have been so bad. But it's incredibly unlikely. Similarly, it's possible that Alex Ovechkin has seen a ton of shots, chances, and goals go in against his team because he's just had rotten luck for 7 seasons straight. But it seems the more likely explanation is that he's just a hugely negative impactful player defensively.

So if the Caps switched goalies between let's say Kuempers best performances and then Samsonovs worst, wouldn't that alter the stats? And let's say the Caps switch between Boudreaus tactics and Trotz, does that change things? Wouldn't that alter the stats?

And if it does, doesn't that indicate that there are many other variables at play than just Ovis performance?

I really dont know how these stats are produced so sorry if it's a stupid question.

 

[twabby]

So if the Caps switched goalies between let's say Kuempers best performances and then Samsonovs worst, wouldn't that alter the stats? And let's say the Caps switch between Boudreaus tactics and Trotz, does that change things? Wouldn't that alter the stats?

And if it does, doesn't that indicate that there are many other variables at play than just Ovis performance?

I really dont know how these stats are produced so sorry if it's a stupid question.

When measuring defensive results typically these models only look at the expected goals against rather than goals against. This is an attempt to separate the defensive quality of a skater from the goaltending. Because skaters can really only affect the quality of chances against, not whether their goalie blows easy saves for instance.

Coaching certainly does impact these GAR measures, though I know Hockeyviz models coaching separately so I believe their player isolates are at least somewhat independent of the coach. Of course GAR measures don’t try to explain why a player performs the way they do. It just measures how poorly or how well they are playing. It could be due to coaching, injury, or a multitude of other issues.

Again I think it’s important to realize that Ovechkin has always struggled defensively. Between 2007-08 and now he’s only registered two seasons with a positive defensive GAR: in 2009-10 and 2015-16. The rest of the years he’s been below replacement defensively. And that’s fine! His offense almost always more than makes up for it. But I think it’s important to realize Ovechkin’s weaknesses because I think he needs to be paired with a center that is more defensively responsible. It’s unsurprising that he and Strome have formed a good duo this year, for instance, while Kuznetsov and Ovechkin have struggled together.

 

[YippieKaey]

When measuring defensive results typically these models only look at the expected goals against rather than goals against. This is an attempt to separate the defensive quality of a skater from the goaltending. Because skaters can really only affect the quality of chances against, not whether their goalie blows easy saves for instance.

Coaching certainly does impact these GAR measures, though I know Hockeyviz models coaching separately so I believe their player isolates are at least somewhat independent of the coach. Of course GAR measures don’t try to explain why a player performs the way they do. It just measures how poorly or how well they are playing. It could be due to coaching, injury, or a multitude of other issues.

Again I think it’s important to realize that Ovechkin has always struggled defensively. Between 2007-08 and now he’s only registered two seasons with a positive defensive GAR: in 2009-10 and 2015-16. The rest of the years he’s been below replacement defensively. And that’s fine! His offense almost always more than makes up for it. But I think it’s important to realize Ovechkin’s weaknesses because I think he needs to be paired with a center that is more defensively responsible. It’s unsurprising that he and Strome have formed a good duo this year, for instance, while Kuznetsov and Ovechkin have struggled together.

Thanks for the explanation! And i agree Ovi is no Carbonneau but it doesn't matter.

 

[YippieKaey]

And given how he looked with Ovi in the beginning of this season, and his point production i honestly wouldn't mind Strome as #1 C. 50 points in 68 games whilst being moved up & down the lineup is pretty good.

Recently he has great 1C numbers.

So if Kuzy gets a good return, trade him.

 

[um]

And given how he looked with Ovi in the beginning of this season, and his point production i honestly wouldn't mind Strome as #1 C. 50 points in 68 games whilst being moved up & down the lineup is pretty good.

Recently he has great 1C numbers.

So if Kuzy gets a good return, trade him.

Not a great 1C, but certainly better than Kuzy and Backstrom this year.

Lavi cost us a few points with his treatment of Strome when Backstrom came back.

 

[Kalopsia]

They are all restricted to the same dataset as far as I'm aware: the NHL's real-time scoring system (RTSS) data that was introduced in the 2007-08 season. It's why a lot of these databases and models have 07-08 as the starting point. That's when they started tracking location events such as shots, and that's also when shift information was available. Before that the only data provided publicly were the simply boxcar counting stats.

I think these models all share a similarish methodology in that they are trying to regress for the outputs from the RTSS data. In other words, given the entirety of the output data that we have (all of the shots, penalties, goals, hits, blocks, etc.), what values for the inputs (e.g. player quality) explain these data the best? Sure it's possible that Jack Johnson has just been getting incredibly unlucky year after year and that's why his on-ice results have been so bad. But it's incredibly unlikely. Similarly, it's possible that Alex Ovechkin has seen a ton of shots, chances, and goals go in against his team because he's just had rotten luck for 7 seasons straight. But it seems the more likely explanation is that he's just a hugely negative impactful player defensively.

I was trying to avoid getting into a discussion about the merits of the models cause that's been done to death here and just make the point that saying "three completely separate models" agree ergo we should have higher confidence in their conclusion is kind of disingenuous when the three models always agree cause the pull from the same inputs and use similar methods. Because they're not really independent of each other, their agreement doesn't tell you any additional information about how likely it is that they're correct.

When measuring defensive results typically these models only look at the expected goals against rather than goals against. This is an attempt to separate the defensive quality of a skater from the goaltending. Because skaters can really only affect the quality of chances against, not whether their goalie blows easy saves for instance.

Coaching certainly does impact these GAR measures, though I know Hockeyviz models coaching separately so I believe their player isolates are at least somewhat independent of the coach. Of course GAR measures don’t try to explain why a player performs the way they do. It just measures how poorly or how well they are playing. It could be due to coaching, injury, or a multitude of other issues.

Again I think it’s important to realize that Ovechkin has always struggled defensively. Between 2007-08 and now he’s only registered two seasons with a positive defensive GAR: in 2009-10 and 2015-16. The rest of the years he’s been below replacement defensively. And that’s fine! His offense almost always more than makes up for it. But I think it’s important to realize Ovechkin’s weaknesses because I think he needs to be paired with a center that is more defensively responsible. It’s unsurprising that he and Strome have formed a good duo this year, for instance, while Kuznetsov and Ovechkin have struggled together.

I know I just said I don't want to get the merits of the models, but I can't help myself here. The idea that you can isolate a player from the impact of their coach when in the vast majority of cases they play every minute of the season with the same coach seems nonsensical to me. How can you possibly adjust for a variable when you have nothing else to compare it to?

 

[twabby]

I was trying to avoid getting into a discussion about the merits of the models cause that's been done to death here and just make the point that saying "three completely separate models" agree ergo we should have higher confidence in their conclusion is kind of disingenuous when the three models always agree cause the pull from the same inputs and use similar methods. Because they're not really independent of each other, their agreement doesn't tell you any additional information about how likely it is that they're correct.


I know I just said I don't want to get the merits of the models, but I can't help myself here. The idea that you can isolate a player from the impact of their coach when in the vast majority of cases they play every minute of the season with the same coach seems nonsensical to me. How can you possibly adjust for a variable when you have nothing else to compare it to?

They use similar input and they use regression but their underlying assumptions and weights are different. I mean yeah, if regression in general is shown to be a highly flawed way of measuring a player’s impact then all of these models will be inaccurate. But no one has really shown this to my knowledge. Saying that these models all have the same flaws is like me saying everyone’s eye test is flawed because you’re all using your eyes.

Re: coaching. I believe Micah just treats coaches as another variable in a regression model. For instance if player plays well under coach A, then stinks under coach B, then does well again under coach C, then we might have some evidence that coach B is a negative influence on a player’s on-ice results. Of course with one player it might just be a coincidence, but running a regression model with all players results and all coaches over every single season will paint a much clearer picture of how coaches impact players performance.

Here are a few results of his coaching model:

Spoiler: “Coaching impact”

All of these results make sense to me. Oates was bad, Boudreau was certainly a fun ‘n gunner, and Laviolette is squeezing everything he can out of his limited talent (curious personnel choices aside) with a lean toward the dull side.

 

[hockeykicker]

 

[Brian23]

I get he's hunting for a spark, but I don't know what part of Sheary's recent game has given any indication he can provide one on the top line.

 

[Silky mitts]

If they could somehow make the playoffs and get Carlson and Connor Brown back they'd have a decent lineup

 

[Random schmoe]

Ow.

 

[g00n]

We speculated about a fracture, but severed artery was not on my radar. Damn.

 

[Random schmoe]

I had actually wondered if his injury could've been a subdural hematoma, or other such major "internal" injury. I don't think this is the same thing, since it sounds like it wasn't bleeding internally, but still .. yeah, "Damn"

 

[Calicaps]

Ow.

sweet jeebus

 

[CapitalsCupReality]

Woooooow……severed artery…..crazy….

 

[usiel]

Ow.

Damn figured it would be something at this point of level. Nice scoop by Tarik sometimes we have to wait till the end of the season for the injury details.

 

[twabby]

Jesus Christ get better John. That sounds awful.

 

[marcel snapshot]

I mean, golf is pretty fun. You got nothing to prove to nobody. If there’s any lingering risk, just dont

 

[dreadpirateroberts]

Hawks fan here who has been to plenty of Caps games this year. Do y’all think they’ll fire Laviolette at season’s end?

 

[twabby]

Hawks fan here who has been to plenty of Caps games this year. Do y’all think they’ll fire Laviolette at season’s end?

No because his contract is up at the end of this year

Serious answer: I could see them keeping Laviolette but I think the odds are very low.

 

[Misery74]

Hawks fan here who has been to plenty of Caps games this year. Do y’all think they’ll fire Laviolette at season’s end?

Hopefully sooner. But his contract expires once we are done. No chance he comes back.

 

[Holtbyisms]

Hawks fan here who has been to plenty of Caps games this year. Do y’all think they’ll fire Laviolette at season’s end?

I pray but expect a big extension announced shortly because that's the most caps thing to do.

 

[usiel]

No because his contract is up at the end of this year

Serious answer: I could see them keeping Laviolette but I think the odds are very low.

Oh sweet thought he had another year.

 

[marcel snapshot]

Does anyone think this organization is taking a hard look at why both Samsonov and Vanecek have been able to do with their new teams what they could not do with the Caps - which is to each become #1 goalies able to win tough games down the stretch.

How are the years which each are having - which, in terms of results and confidence, are galactically better than what they did here - not a serious indictment of how goalies are managed in this organization