Who deserves the Selke strictly based on numbers?

[The Moose is Loose]

OP is also the same person who started 2 threads this summer "Leon Draisaitl is not elite" and "John Carlson is not elite" and throughtout this season doubled down on his argument because his hubris won't allow him to admit when he has made a mistake . These 2 guys will probably be taking home 4 of the major awards for the season.
Just goes to show how easily manipulated advanced stats can be.

My point is what he says should be taken with a grain of salt. If you honestly believe that Draisaitl is a 3rd liner without McDavid and that the Oilers are better with Draisaitl on the bench (yes he said these both verbatim) then that shows how much his opinion is worth about Aho and others.

 

[Dan Kelly]

Blake Coleman deserves an award just for being traded out of New Jersey

 

[TomasHertlsRooster]

OP is also the same person who started 2 threads this summer "Leon Draisaitl is not elite" and "John Carlson is not elite" and throughtout this season doubled down on his argument because his hubris won't allow him to admit when he has made a mistake . These 2 guys will probably be taking home 4 of the major awards for the season.
Just goes to show how easily manipulated advanced stats can be.

My point is what he says should be taken with a grain of salt. If you honestly believe that Draisaitl is a 3rd liner without McDavid and that the Oilers are better with Draisaitl on the bench (yes he said these both verbatim) then that shows how much his opinion is worth about Aho and others.

1. OP is @Love who asked who deserves the Selke based strictly on numbers.

2. The thread about Draisaitl was started in the summer of 2018.

3. The thread about Draisaitl made clear that he was an elite winger, but had not proven himself as an elite #1C in the minutes that he had played at center.

4. I have never said Draisaitl was a third liner without McDavid.

5. I never said that they were a better team as a whole with Draisaitl on the bench. What I said was that prior to this season, when McDavid was not present in either scenario at 5V5, there was a 3-year sample of the Edmonton Oilers controlling a larger share of the shots, goals, and equal share of expected goals when Leon Draisaitl was on the bench than the share which they controlled when he was on the ice.

In the future, if you come into a thread to add absolutely nothing to the discussion, and instead just say “this poster’s opinion is worth nothing because of what they’ve said in the past”, you might actually want to, you know, accurately recapitulate what they’d said in the past? Your post did more to hurt your own credibility than it did mine.

 

[The Moose is Loose]

1. OP is @Love who asked who deserves the Selke based strictly on numbers.

2. The thread about Draisaitl was started in the summer of 2018.

3. The thread about Draisaitl made clear that he was an elite winger, but had not proven himself as an elite #1C in the minutes that he had played at center.

4. I have never said Draisaitl was a third liner without McDavid.

5. I never said that they were a better team as a whole with Draisaitl on the bench. What I said was that prior to this season, when McDavid was not present in either scenario at 5V5, there was a 3-year sample of the Edmonton Oilers controlling a larger share of the shots, goals, and equal share of expected goals when Leon Draisaitl was on the bench than the share which they controlled when he was on the ice.

In the future, if you come into a thread to add absolutely nothing to the discussion, and instead just say “this poster’s opinion is worth nothing because of what they’ve said in the past”, you might actually want to, you know, accurately recapitulate what they’d said in the past? Your post did more to hurt your own credibility than it did mine.

You go ahead and believe that. Your reputation is of a poster whose opinion has to be questioned because you follow advanced statistics with such tunnel vision you can't even realize how ridiculous some of your statements are. I do respect how well written you are though, and your arguments are nicely articulated.

I'm not going to go dig through years worth of your consistent posting, but you absolutely without a doubt said "Draisaitl produces at a 3rd line rate without McDavid". I remember it very clearly because I was absolutely stunned that someone could actually believe that. In the same argument you concluded from your statistics that seperated from McDavid, the Oilers were a better team with Draisaitl on the bench, than having Draisaitl (no McDavid) on the ice. Which again is crazy, he is objectively their 2nd best player.

Yes you are right about it being the summer prior, my bad. Regardless the point still stands, you have an apparent vendetta against certain players and you got proven wrong by results. Drai followed your thread by posting the only 50+50 year of the past 8 seasons.


And I am adding something to the thread. That advanced statistics can be mis/over interpreted.
As said previously, I think your a smart poster and your arguments have their merits especially when we are comparing close players. Something like Eichel vs Matthews which everyone can agree is a very close debate I would 100% defer to defensive gar, evo, evd, ppo, shd and the like to determine who the more effective player is.
But some of the conclusions you draw from your findings are just too absurd.
Even in this thread you said Nuke is the best defensive forward in the NHL?!? He's been a beast for the thus far Avs but come on, that's not even close to true.

I know I'm not going to change your mind because you've been the same for years now, but I think being open to opposing ideas and setting aside the cavalier attitude that your always right because of "overly specific staistical point #23" would go a long ways.

 

[Bear of Bad News]

Be nice, guys.

 

[Love]

OP is also the same person who started 2 threads this summer "Leon Draisaitl is not elite" and "John Carlson is not elite" and throughtout this season doubled down on his argument because his hubris won't allow him to admit when he has made a mistake . These 2 guys will probably be taking home 4 of the major awards for the season.
Just goes to show how easily manipulated advanced stats can be.

My point is what he says should be taken with a grain of salt. If you honestly believe that Draisaitl is a 3rd liner without McDavid and that the Oilers are better with Draisaitl on the bench (yes he said these both verbatim) then that shows how much his opinion is worth about Aho and others.

I never said any of this stuff or made those threads

 

[The Moose is Loose]

I never said any of this stuff or made those threads

Sorry I mixed you up with someone else.
Thought Joe Thortons Rooster was OP (which he kind is since he made it his thread on the 2nd post)

He got the point tho since he did make those threads.

 

[TomasHertlsRooster]

You go ahead and believe that. Your reputation is of a poster whose opinion has to be questioned because you follow advanced statistics with such tunnel vision you can't even realize how ridiculous some of your statements are. I do respect how well written you are though, and your arguments are nicely articulated.

I'm not going to go dig through years worth of your consistent posting, but you absolutely without a doubt said "Draisaitl produces at a 3rd line rate without McDavid". I remember it very clearly because I was absolutely stunned that someone could actually believe that. In the same argument you concluded from your statistics that seperated from McDavid, the Oilers were a better team with Draisaitl on the bench, than having Draisaitl (no McDavid) on the ice. Which again is crazy, he is objectively their 2nd best player.

Yes you are right about it being the summer prior, my bad. Regardless the point still stands, you have an apparent vendetta against certain players and you got proven wrong by results. Drai followed your thread by posting the only 50+50 year of the past 8 seasons.


And I am adding something to the thread. That advanced statistics can be mis/over interpreted.
As said previously, I think your a smart poster and your arguments have their merits especially when we are comparing close players. Something like Eichel vs Matthews which everyone can agree is a very close debate I would 100% defer to defensive gar, evo, evd, ppo, shd and the like to determine who the more effective player is.
But some of the conclusions you draw from your findings are just too absurd.
Even in this thread you said Nuke is the best defensive forward in the NHL?!? He's been a beast for the thus far Avs but come on, that's not even close to true.

I know I'm not going to change your mind because you've been the same for years now, but I think being open to opposing ideas and setting aside the cavalier attitude that your always right because of "overly specific staistical point #23" would go a long ways.

[MOD DELETE] but the stuff that I said about Draisaitl were objective facts at the time.

It was a fact that the non-McDavid Oilers controlled a larger share of the 5v5 goals with Draisaitl on the bench than they did when he was on the ice; here is proof. Unless the better hockey team is no longer the one that controls a larger share of the goals, I'm pretty sure they were a better hockey team when he is off the ice.

And, I certainly wasn't proven wrong by his 2018-2019 year, where just about all of his offensive production in his 50+50 year came playing winger alongside McDavid; and his 5-on-5 scoring rate of 1.47 points/60 without McDavid would tie him for 222nd in the NHL among forwards with at least 500 minutes, which would comfortably fail to meet a 2nd-line cut-off of top-186. A guy who I specifically referred to as an elite winger had an elite season when he was playing the wing position; I didn't make any predictions about future output, but if I had suggested that the future would mirror the past, then I would've correctly said that the elite winger who struggles at center would continue to be elite on the wing and struggle at center. This 2019-2020 season is the first time in his career since 2016 with Taylor Hall that he's actually posted positive results at the center position, but it came in a fairly small sample, and with a very high PDO. It's very possible that he maintains this, and solidifies himself as a legitimate elite #1 center, but doing so wouldn't have proven me wrong because I never said that he couldn't do that; only that he hadn't yet.

In this thread, which asks who deserves the Selke strictly based on numbers, I said that Valeri Nichushkin is deserving based strictly on numbers from this season. I looked at even-strength and short-handed defensive value added, both as they were quantified by GAR and xGAR, and concluded that Nichushkin had comfortably provided the most defensive value according to these metrics. I did note that it is easier to play a lower-event style of hockey and suppress shots from opponents when you play in bottom-6 minutes, as Nichushkin did for the majority of the season, and that I would probably feel more comfortable giving the award to somebody in a bit of a tougher role, playing more frequently against top line players, but all of the players who fit that bill were so far behind Nichushkin in defensive value. The highest ranked player who fits that role is Gabriel Landeskog, and the average between his defensive xGAR and defensive GAR is less than 2/3 the value of Nichushkin's average.

I'm plenty open to opposing ideas, and I think people too often are confuse my opposing ideas for a cavalier attitude. In this thread, I've stated that I'm well open to opposing ideas regarding Sebastian Aho's defensive play, but I haven't been provided with anything besides "everybody knows this isn't true", which effectively boils down to argument ad populum, which is a logical fallacy. tarheelhockey asked me to provide him with the data set for RAPM so that he could objectively shoot him down, and I did so and let him know I was open to hearing him shoot it down. On a similar note, if you can provide me with a counter argument that Valeri Nichushkin does not deserve the Selke - especially in a thread which is titled "strictly based on numbers" - I'd be happy to hear it, and consider the opposing perspective. When your opposing perspective is just "this guy is wrong because I misinterpreted an argument he made 2 years ago about a guy who scored a lot of points this season", then what the hell is there for me to consider?

 

[Dan Kelly]

having said all that, watch Leon get awarded the Selke !!

 

[tarheelhockey]

Well, first off, the adjusted metric is not saying that Aho is one of the worst players in the league. The fact that you've said that, despite Jeff going out of his way to emphasize that Aho is indeed a good player due to his offensive impact, and me also going out of my way to emphasize that he is an excellent offensive player, is rather telling that you're either being disingenuous or just not taking the time to listen to the perspective of the other side.

Or, a third possibility: that you need to go back and read the post you quoted, where I clearly said "defense".

The fact that you jumped straight to to "you're either lying or not reading posts" tells me this conversation isn't heading in a productive direction.

Massive wall of text

Yeah, no. I'm not here for an argument where "winning" means spewing text till the other guy gives up trying to write a dissertation-length response. The one part of that I'm going to address is: Aho isn't the matchup center, Staal is.

Yes, that's correct. Staal is a better defensive center than Aho, and invariably going to be above Aho on a defensive metric. And Wallmark, a solid defender who skates low minutes on a conservative 4th line. We would predict, through sheer common sense, that these players would put up better defensive metrics than Aho... even if Aho plays fine defensively.

So before we run a "relative to teammates" analysis, we should probably step back and ask if that's the right way to judge Aho's defensive performance. We should also ask if there's a systemic difference between the events when these players are on the ice -- for example, if one of them is frequently deep in the offensive zone at the precise moment when an goal-oriented defenseman makes a bad pinch, causing an un-defendable rush in the other direction.

We should be asking those questions especially in a case where the model is producing fluke results that are obviously incorrect. Like Aho being a terrible defender or... I don't know, Brett Pesce being the team's worst defenseman in his own end... or maybe Nino Niederreiter being the top defensive winger. It's not "feelings" to say those are ludicrous conclusions, it's simply being connected to the reality of what these numbers represent.

 

[TomasHertlsRooster]

Or, a third possibility: that you need to go back and read the post you quoted, where I clearly said "defense".

The fact that you jumped straight to to "you're either lying or not reading posts" tells me this conversation isn't heading in a productive direction.



Yeah, no. I'm not here for an argument where "winning" means spewing text till the other guy gives up trying to write a dissertation-length response. The one part of that I'm going to address is: Aho isn't the matchup center, Staal is.

Yes, that's correct. Staal is a better defensive center than Aho, and invariably going to be above Aho on a defensive metric. And Wallmark, a solid defender who skates low minutes on a conservative 4th line. We would predict, through sheer common sense, that these players would put up better defensive metrics than Aho... even if Aho plays fine defensively.

So before we run a "relative to teammates" analysis, we should probably step back and ask if that's the right way to judge Aho's defensive performance. We should also ask if there's a systemic difference between the events when these players are on the ice -- for example, if one of them is frequently deep in the offensive zone at the precise moment when an goal-oriented defenseman makes a bad pinch, causing an un-defendable rush in the other direction.

We should be asking those questions especially in a case where the model is producing fluke results that are obviously incorrect. Like Aho being a terrible defender or... I don't know, Brett Pesce being the team's worst defenseman in his own end... or maybe Nino Niederreiter being the top defensive winger. It's not "feelings" to say those are ludicrous conclusions, it's simply being connected to the reality of what these numbers represent.

The relative-to-teammate analysis is far from the only argument I made, or the first thing that I used to judge his defensive performance. Outside of the regression analysis, the other half of the raw picture is the one where the rate of 5-on-5 expected goals that the Hurricanes allow with Aho on the ice is higher than every team in the NHL besides the Jets, Blackhawks, and Rangers. The two of them put together is pretty damning, as is the fact that you didn't address any other argument. So is the fact that you are yet again reverting to "this is obviously incorrect" without providing any evidence as to why this is so obvious outside of argument ad populum.

 

[tarheelhockey]

The false positive may just be your assessment of Aho being good defensively.

As for the collective thinking Aho is good defensively, I'm not so sure that's accurate. I'm happy to be proven wrong but truth be told I haven't seen a lot of discussion about Aho's defensive game, good or bad. If the collective truly does believe that Aho is an elite defensive player, they sure have been quiet about it.

Nobody claimed Aho is an elite defensive player. Even I didn't claim that.

What's missing here is ANY support from the hockey world -- analysts, coaches, fans, literally anybody -- for the idea that Aho is flat-out terrible defensively. Given the way centers get publicly skewered for being weak, especially high-profile centers on huge new contracts, the total absence of any such observation from any credible source speaks volumes about the truth value of that claim.

And frankly, it's a little weird and telling that this conversation hasn't included a single phrase along the lines of "he doesn't backcheck" or "he cheats in the zone" or "he's weak in the slot". Those phrases have not appeared because they aren't true. The statistical results being cited here don't correspond to observable traits.

As for your point on Pesce being worse defensively than Gardiner, it is only a 1 year sample, and this season was uncharacteristically bad for Pesce.

Hang on. In what way was this season uncharacteristically bad for Pesce?

As for the van Riemsdyk vs Slavin thing, while van Riemsdyk does have better defensive numbers than Slavin this year, he does have better raw defensive numbers than him (2.25 xGA/60 for TvR, 2.31 xGA/60 for Slavin). While their deployment is different, the numbers are actually a lot more skewed in favor of van Riemsdyk. While Slavin has faced the tougher competition, and van Riemsdyk has been relatively sheltered, the spread isn't huge. The spread in teammates, however, is much bigger, with Slavin getting to play most of his minutes with the best defensive players on the team (Hamilton, Staal, Teravainen, etc), while van Riemsdyk has spent the majority of his time with some of the worst (Aho, Necas, Dzingel, etc). So while the "tougher competition" argument favors Slavin, the "shittier teammates" argument favors van Riemsdyk equally. Given that their deployment is essentially a wash, it's not crazy to say van Riemsdyk has been better defensively this year, especially given his raw numbers are better.

I don't even know what to say.

First -- I like TVR. I think he's a nice utility player, a bit of a luxury on the bottom pair. I usually am in the position of advocating for him.

But seriously. TVR is a bottom-pair defenseman with limited upside at either end of the ice. He spent time this season as a healthy scratch, invariably struggles with spot-duty for better players, and is often cited as a player who's likely to be deemed expendable soon.

If this model is showing TVR as a better defensive performer than Brett ****ing Pesce or Jaccob ****ing Slavin, then there is something wrong with the model. Full stop. This is where it's incumbent upon a statistician to listen to feedback and understand the difference between an "I don't like numbers" red flag and a "these numbers don't correspond to demonstrable reality" red flag.

 

[tarheelhockey]

The relative-to-teammate analysis is far from the only argument I made, or the first thing that I used to judge his defensive performance. Outside of the regression analysis, the other half of the raw picture is the one where the rate of 5-on-5 expected goals that the Hurricanes allow with Aho on the ice is higher than every team in the NHL besides the Jets, Blackhawks, and Rangers. The two of them put together is pretty damning, as is the fact that you didn't address any other argument. So is the fact that you are yet again reverting to "this is obviously incorrect" without providing any evidence as to why this is so obvious outside of argument ad populum.

I already addressed this point (whether you acknowledge it or not) by asking whether this Expected Goals Against data is validated for potential systemic biases, and specifically describing the sort of systemic bias that might tilt an Expected Goals Against metric against a player who frequently plays in the offensive zone on a team that emphasizes aggressive D pinching -- which leads to rushes against, which raises the number of shots taken from rushes, which raises Expected Goals Against systemically.

So, again, has the model been validated against that bias?

 

[TomasHertlsRooster]

I already addressed this point (whether you acknowledge it or not) by asking whether this Expected Goals Against data is validated for potential systemic biases, and specifically describing the sort of systemic bias that might tilt an Expected Goals Against metric against a player who frequently plays in the offensive zone on a team that emphasizes aggressive D pinching -- which leads to rushes against, which raises the number of shots taken from rushes, which raises Expected Goals Against systemically.

So, again, has the model been validated against that bias?

I don't exactly understand what you are asking here, but I think I have an idea, so I'll try to respond to what I think you're asking.

The regression analysis (RAPM) does indeed account for systemic variables - it uses every teammate as a predictor variable and then removes the impact of that predictor variable from a player's individual results. So, just as an example here, if Jake Gardiner on the ice as a predictor variable is associated with an increased rate in expected goals against (due to his heavy pinches), the regression analysis will adjust for this and the baseline for a Sebastian Aho shift with Jake Gardiner will be a higher rate of expected goals against than it otherwise would be. So, the validation against these biases is built-in. On a similar note, relative-to-teammate metrics should also be good to look at for this; if every Hurricanes defenseman is heavily inflating their xGA rates, a good defensive center may have raw inflated rates, but higher relative rates.

However, if you are asking about the raw expected goal model itself, then I am still not sure what you are asking. Are you saying that Aho's expected goal against rates are higher because he spends more time with the puck in the offensive zone? And that, by virtue of being a better offensive player, his defensive metrics are inevitably going to shift downwards (as long as he is in the Canes system)?

 

[tarheelhockey]

I don't exactly understand what you are asking here, but I think I have an idea, so I'll try to respond to what I think you're asking.

The regression analysis (RAPM) does indeed account for systemic variables - it uses every teammate as a predictor variable and then removes the impact of that predictor variable from a player's individual results. So, just as an example here, if Jake Gardiner on the ice as a predictor variable is associated with an increased rate in expected goals against (due to his heavy pinches), the regression analysis will adjust for this and the baseline for a Sebastian Aho shift with Jake Gardiner will be a higher rate of expected goals against than it otherwise would be. So, the validation against these biases is built-in. On a similar note, relative-to-teammate metrics should also be good to look at for this; if every Hurricanes defenseman is heavily inflating their xGA rates, a good defensive center may have raw inflated rates, but higher relative rates.

However, if you are asking about the raw expected goal model itself, then I am still not sure what you are asking. Are you saying that Aho's expected goal against rates are higher because he spends more time with the puck in the offensive zone? And that, by virtue of being a better offensive player, his defensive metrics are inevitably going to shift downwards (as long as he is in the Canes system)?

Trying to illustrate the question a little better:

Player A, 100 shots allowed
70 off in-zone possessions
30 off the rush

Player B, 100 shots allowed
30 off in-zone possessions
70 off the rush

All else equal, Player A’s ExpectedGA will suffer for having allowed the higher proportion of shots off the rush, right?

 

[mynamejeff420]

Trying to illustrate the question a little better:

Player A, 100 shots allowed
70 off in-zone possessions
30 off the rush

Player B, 100 shots allowed
30 off in-zone possessions
70 off the rush

All else equal, Player A’s ExpectedGA will suffer for having allowed the higher proportion of shots off the rush, right?

Not really. Rush shots are accounted for with xGA, so if assuming the shot locations were all the same for both Player A and Player B, Player B would still have the worse xGA because of the rush shots.

 

[Michael Farkas]

We should also ask if there's a systemic difference between the events when these players are on the ice -- for example, if one of them is frequently deep in the offensive zone at the precise moment when an goal-oriented defenseman makes a bad pinch, causing an un-defendable rush in the other direction.

Just to say the words, with some of the NHL teams that I talk to and even some higher-end junior programs sometimes, they're at this point and they've been at it. I'm not going to pretend like I know the ins and outs of all 31 teams analytics departments. But there's a reason why all this data became public from the NHL...it was no longer essential to the process for teams. Now, there's some people out here that are doing awesome work and challenging things in a way that they've never been challenged before or demonstrating the game in a way that's never been statistically/analytically captured before...but then there's these situations which have you found yourself in here, where - at a glance - appears to be purely statistical (correct me if I'm wrong). If that isn't course 101 in this curriculum, it's probably 102, or at best, 201.

A lot of teams have a "grading" element to their work...think Pro Football Focus type of stuff if you're familiar with it. Because you can come up with as many ways to twist the data as you want, but you can easily fail at two distinctly separate points: The old garbage in, garbage out stand by. And then the analytics aren't useful if there is no application to the game. You can print out the graph and the spreadsheet and go, "see, this is what it says...so that's how it is...that's my subjectivity that I'm masquerading as pure objectivity..." but if there's no application to the game, you got nothin'...or at least, much less than what you think you have. And that's how this essentially reads to me.

Sometimes analytics will reveal an ugly truth about a player and then sometimes you have to re-evaluate your methodology if you produce, frankly, garbage results...if you got something that says fringe NHLer Trevor vanRiemsdyk is better than Slavin and Pesce at anything besides maybe filling up the bottles, and if that same something says something about Jake Gardiner being better defensively than...anything...and that same something says Sebastian Aho is a poor defensive player...it's extremely likely that you just don't have a great metric here. You can't take "action" on that as a coach or a GM.

And if that metric(s) is reasonably "accurate" on 22 teams, but it "fails" (so emphatically) in Carolina and some other teams, then you want to find out why...because maybe you have something good for teams that play like this, or situations like this...but it's not so good for this/that. This is what happened on the early hype train when all the corsi stuff went public...everyone just assumed that because it was data, that you automatically produced objective conclusions based on it. And it made a lot of people here look very silly because not enough people took a minute to actually make an application to the game. I remember as soon as it got released, I thought, "man, this is really gonna damage defensive centers and really pump up 'streak and shoot', low hockey IQ wingers." and sure enough...

But I mean, imagine if when the VsX metric was released on HoH it produced this list for most productive offensive players of all time:
1. Wayne Gretzky
2. Craig Janney
3. Ron Francis
4. Marc Savard
5. Mario Lemieux
6. Joe Malone
7. Ilya Kovalchuk

You think anyone's gonna go, "Yeah, I always kinda thought Janney [and those other dumb names up there] was in Gretzky territory..."? Of course not. They're going to say, "well, that didn't work..."

 

[TomasHertlsRooster]

Trying to illustrate the question a little better:

Player A, 100 shots allowed
70 off in-zone possessions
30 off the rush

Player B, 100 shots allowed
30 off in-zone possessions
70 off the rush

All else equal, Player A’s ExpectedGA will suffer for having allowed the higher proportion of shots off the rush, right?

This expected goal model actually doesn’t use rush chances as a variable. The full write-up is here: RPubs - A New Expected Goal Model for Predicting Goals in the NHL

However, it uses prior events and recency of prior events as a variable, so if there are certain prior events (for example, an OZ turnover made by a pinching defenseman, 4 seconds before the shot) that are correlated with an increase in goal odds on a future shot that also occur before a rush chance, then these events should increase

I would say that in your scenario, player B will probably allow a higher number of expected goals, since they’re the one who allowed the higher proportion of shots off the rush. (Which I’m assuming is what you meant to ask.) Shots coming off the rush will likely have more dangerous pre-shot events occurring more recently than shots coming off in-zone possession.

However, let’s just say for the sake of discussion that player B would definitely have a higher rate of expected goals against. Theoretically speaking, if all variables are held constant (shot distance/angle, shooter talent, goaltending talent) would they not also have a higher rate of actual goals against? And if so, then what is the issue with them having a higher number of expected goals against?

 

[mynamejeff420]

Nobody claimed Aho is an elite defensive player. Even I didn't claim that.

What's missing here is ANY support from the hockey world -- analysts, coaches, fans, literally anybody -- for the idea that Aho is flat-out terrible defensively. Given the way centers get publicly skewered for being weak, especially high-profile centers on huge new contracts, the total absence of any such observation from any credible source speaks volumes about the truth value of that claim.

And frankly, it's a little weird and telling that this conversation hasn't included a single phrase along the lines of "he doesn't backcheck" or "he cheats in the zone" or "he's weak in the slot". Those phrases have not appeared because they aren't true. The statistical results being cited here don't correspond to observable traits.



Hang on. In what way was this season uncharacteristically bad for Pesce?



I don't even know what to say.

First -- I like TVR. I think he's a nice utility player, a bit of a luxury on the bottom pair. I usually am in the position of advocating for him.

But seriously. TVR is a bottom-pair defenseman with limited upside at either end of the ice. He spent time this season as a healthy scratch, invariably struggles with spot-duty for better players, and is often cited as a player who's likely to be deemed expendable soon.

If this model is showing TVR as a better defensive performer than Brett ****ing Pesce or Jaccob ****ing Slavin, then there is something wrong with the model. Full stop. This is where it's incumbent upon a statistician to listen to feedback and understand the difference between an "I don't like numbers" red flag and a "these numbers don't correspond to demonstrable reality" red flag.

You're not really providing any support from the hockey world that Aho is good or above average or great (or whatever you want to call it) at defense. You say "everyone in the hockey world thinks this" but you =haven't shown any concrete proof either.

As for the Pesce thing, he tends to put up some of the best defensive results in the league, year after year. This year he had poor on ice results from a defensive perspective, whether it was GA or xGA. I still think it's safe to say that he's one of the best defenders in the league, because of his strong priors. But this year, he wasn't that.

As for TvR, the Hurricanes allowed very few goals (expected or otherwise) when he was on the ice this year. That's why the model is spitting out that he is good defensively. No matter what model you use or what metric you choose, one thing that won't change is the goals allowed when TvR is on the ice, or when Slavin is, or when Pesce is. Full stop, the Canes didn't allow much when TvR was on the ice. This has been consistent for a few years now. At a certain point, if he's constantly putting up good defensive results, it may be time to accept that he's really good at defense. Does that mean he should be playing over Slavin or Pesce? No, because defense only account for half of the game and both of those guys are miles better at offense than TvR. But imo it's not that outlandish to suggest that TvR may be on a similar level defensively to guys like Slavin or Pesce, considering he accomplishes the goal of defense (i.e. preventing goals) at a similar rate.

 

[tarheelhockey]

Not really. Rush shots are accounted for with xGA, so if assuming the shot locations were all the same for both Player A and Player B, Player B would still have the worse xGA because of the rush shots.

This expected goal model actually doesn’t use rush chances as a variable. The full write-up is here: RPubs - A New Expected Goal Model for Predicting Goals in the NHL

However, it uses prior events and recency of prior events as a variable, so if there are certain prior events (for example, an OZ turnover made by a pinching defenseman, 4 seconds before the shot) that are correlated with an increase in goal odds on a future shot that also occur before a rush chance, then these events should increase

I would say that in your scenario, player B will probably allow a higher number of expected goals, since they’re the one who allowed the higher proportion of shots off the rush. (Which I’m assuming is what you meant to ask.) Shots coming off the rush will likely have more dangerous pre-shot events occurring more recently than shots coming off in-zone possession.

However, let’s just say for the sake of discussion that player B would definitely have a higher rate of expected goals against. Theoretically speaking, if all variables are held constant (shot distance/angle, shooter talent, goaltending talent) would they not also have a higher rate of actual goals against? And if so, then what is the issue with them having a higher number of expected goals against?

Ok, so the player who allows more shots off the rush will very likely have a higher Expected GA.

A rush is a transitional event, right?

By definition, transition is a rapid change from one offensive zone to the other, right?

So, if a player spent a high proportion of his time in his offensive zone to begin with, he might face a proportionately higher rate of shots off the rush, compared to a guy who usually started in the defensive zone and mainly dealt with in-zone defensive play... and this might be accentuated if he played within a system that was characterized by very strong emphasis on defensive pinches at the risk of rushes-against. Yes?

 

[tarheelhockey]

You're not really providing any support from the hockey world that Aho is good or above average or great (or whatever you want to call it) at defense. You say "everyone in the hockey world thinks this" but you =haven't shown any concrete proof either.

To be honest, I am comfortable assuming you have access to Google and would already have cited credible examples of Aho being skewered for his defense, if any such thing existed.

The fact that this hasn’t been cited, tells me it doesn’t exist, and in turn tells me I don’t need to waste time posting every article under the sun that says something to the effect of “Aho is not elite, but quite competent defensively”.

As for the Pesce thing, he tends to put up some of the best defensive results in the league, year after year. This year he had poor on ice results from a defensive perspective, whether it was GA or xGA. I still think it's safe to say that he's one of the best defenders in the league, because of his strong priors. But this year, he wasn't that.

I’m asking for something more specific than just reading the metrics back.

What specifically leads you to believe that Pesce was worse this year than in prior years?

As for TvR, the Hurricanes allowed very few goals (expected or otherwise) when he was on the ice this year. That's why the model is spitting out that he is good defensively. No matter what model you use or what metric you choose, one thing that won't change is the goals allowed when TvR is on the ice, or when Slavin is, or when Pesce is. Full stop, the Canes didn't allow much when TvR was on the ice. This has been consistent for a few years now. At a certain point, if he's constantly putting up good defensive results, it may be time to accept that he's really good at defense. Does that mean he should be playing over Slavin or Pesce? No, because defense only account for half of the game and both of those guys are miles better at offense than TvR. But imo it's not that outlandish to suggest that TvR may be on a similar level defensively to guys like Slavin or Pesce, considering he accomplishes the goal of defense (i.e. preventing goals) at a similar rate.

I’m not saying this to be insulting or provocative, but because it’s the simple truth.

The statement “TVR is better defensively than Slavin or Pesce, but they’re just miles better than him offensively” would very strongly suggest that you have zero frame of reference for these players as hockey players.

@Mike Farkas made a good point upthread that in order for these metrics to be meaningful, they have to be actionable. A metric showing TVR to be better than either Pesce or Slavin defensively, but unable to keep up with them offensively, is the diametric opposite of actionable.

 

[TomasHertlsRooster]

Ok, so the player who allows more shots off the rush will very likely have a higher Expected GA.

A rush is a transitional event, right?

By definition, transition is a rapid change from one offensive zone to the other, right?

So, if a player spent a high proportion of his time in his offensive zone to begin with, he might face a proportionately higher rate of shots off the rush, compared to a guy who usually started in the defensive zone and mainly dealt with in-zone defensive play... and this might be accentuated if he played within a system that was characterized by very strong emphasis on defensive pinches at the risk of rushes-against. Yes?

So, you think that the Hurricanes are allowing so many expected goals (and actual goals) at such a high rate with Aho on the ice, and you think that Aho’s isolated impact on expected goals (and actual goals) against as quantified per regression is so bad, because he starts more shifts in the offensive zone?

 

[tarheelhockey]

So, you think that the Hurricanes are allowing so many expected goals (and actual goals) at such a high rate with Aho on the ice, and you think that Aho’s isolated impact on expected goals (and actual goals) against as quantified per regression is so bad, because he starts more shifts in the offensive zone?

I’m asking questions about the model.

If a disproportionate number of a player’s shots-against come from the opponent cycling the puck around and taking shots on in-zone possession, his ExpectedGA will be lower than a player who faces a disproportionate number of shots off the rush, correct?

And a shot off the rush is by definition a possession that begins at the other end of the ice, correct?

And a disproportionate number of rushes-against happen in systems where the D pinch aggressively, correct?

Putting those three correct statements together — a player will have a disproportionately higher Expected GA if he faces a disproportionately high number of rushes, which is disproportionately likely to happen if he spends a disproportionate amount of time in the offensive zone with pinching defensemen behind him, correct?

 

[TomasHertlsRooster]

I’m asking questions about the model.

If a disproportionate number of a player’s shots-against come from the opponent cycling the puck around and taking shots on in-zone possession, his ExpectedGA will be lower than a player who faces a disproportionate number of shots off the rush, correct?

And a shot off the rush is by definition a possession that begins at the other end of the ice, correct?

And a disproportionate number of rushes-against happen in systems where the D pinch aggressively, correct?

Putting those three correct statements together — a player will have a disproportionately higher Expected GA if he faces a disproportionately high number of rushes, which is disproportionately likely to happen if he spends a disproportionate amount of time in the offensive zone with pinching defensemen behind him, correct?

The answer to the first question is not something that can be provided with absolute certainty. Because there is no discrete "rush" variable present in this expected goal model, if everything about two shots is equal besides one being off the rush and another being off a possession play, then there is no reason that the one off the rush would be higher.

However, it's very unlikely that everything about a shot off the rush will be equal to everything about a shot off an in-zone possession; the biggest discrepancy will be in the prior event. However, I'm not convinced that the prior events which occur before rush chances are all that much higher than the prior events which occur before in-zone possession chances are actually associated with an increase in expected goal odds. If we look at the factors that drive shot quality, the prior event does play a fairly large role in driving the expected goal value of any given shot, but the prior event has to be one that is more recent, and the prior event with the strongest influence on increasing xG odds is a shot by the same team; the one with the second strongest influence is a missed shot by the same team. Prior events like giveaways by the opposite team or takeaways by the same team do not even make the list. And a giveaway by the opposing team would likely be occurring with a high time before and high distance from the shot, so I think rush chances really wouldn't change much.

I think this is actually one of the biggest flaws of the model; that it doesn't properly assess the value of shots off the rush, due to the increased frequency of passing plays occurring before shots off the rush. When the authors of the model wrote about this, they expressed similar concerns:

This was a classic 2-on-1 sequence with a clean cross-crease pass before the shot was taken. This event, while not necessarily “common”, is something we see somewhat often in the NHL. It is not surprising that this shot resulted in a goal. However, the model does not have access to any passing data (location or time of the pass) and doesn’t know where the players are on the ice (possibly player tracking data would allow this). The model only sees this as single shots taken close to the net.

The full write up is here, and you can scroll to the very bottom of the page to see the specific plays that they are talking about; 3 of the 5 goals that they consider undervalued are those that come off the rush, with a quick pass occurring before the shot.

So, the answer to your very first question, as best as I can put it, is probably no. It's certainly not a definitively correct statement, which means that the other two statements can't really follow from it.

If anything, you could probably make the argument that if a team is allowing a disproportionately high percentage of their chances off the rush, that expected goals against will undersell the quality of chances they're actually allowing. Which anecdotally checks out.

 

[tarheelhockey]

The answer to the first question is not something that can be provided with absolute certainty. Because there is no discrete "rush" variable present in this expected goal model, if everything about two shots is equal besides one being off the rush and another being off a possession play, then there is no reason that the one off the rush would be higher.

However, it's very unlikely that everything about a shot off the rush will be equal to everything about a shot off an in-zone possession; the biggest discrepancy will be in the prior event. However, I'm not convinced that the prior events which occur before rush chances are all that much higher than the prior events which occur before in-zone possession chances are actually associated with an increase in expected goal odds. If we look at the factors that drive shot quality, the prior event does play a fairly large role in driving the expected goal value of any given shot, but the prior event has to be one that is more recent, and the prior event with the strongest influence on increasing xG odds is a shot by the same team; the one with the second strongest influence is a missed shot by the same team. Prior events like giveaways by the opposite team or takeaways by the same team do not even make the list. And a giveaway by the opposing team would likely be occurring with a high time before and high distance from the shot, so I think rush chances really wouldn't change much.

View attachment 342860

I think this is actually one of the biggest flaws of the model; that it doesn't properly assess the value of shots off the rush, due to the increased frequency of passing plays occurring before shots off the rush. When the authors of the model wrote about this, they expressed similar concerns:



The full write up is here, and you can scroll to the very bottom of the page to see the specific plays that they are talking about; 3 of the 5 goals that they consider undervalued are those that come off the rush, with a quick pass occurring before the shot.

So, the answer to your very first question, as best as I can put it, is probably no. It's certainly not a definitively correct statement, which means that the other two statements can't really follow from it.

If anything, you could probably make the argument that if a team is allowing a disproportionately high percentage of their chances off the rush, that expected goals against will undersell the quality of chances they're actually allowing. Which anecdotally checks out.

So rather than have a long strung-out argument, we can simply agree on this statement: there are systemic biases built into the model.

Those systemic biases will, inevitably, conjure statistical phantoms. And identifying those unsupportable results is necessary for refining the model.

All I’m trying to accomplish here is to wave a flag and point out that I see several unsupportable results cropping up on a single team — one that I happen to know very intimately and am (at worst) unlikely to be completely off-base about.

The flag has been waved. Having said my peace, if you still truly think the metrics are correct, that there is no phantom but a real tangible result, then that’s simply a judgment call. Nothing I can do to change it and I’m not bothered by letting it end there.