2021 Blues Regular Season multi-purpose thread

[ChicagoBlues]

The key thing with any model of advanced stat breakdown is to compare players of similar usage, quality of linemates, and quality of competition. A lot of the advanced stat truthers will ignore all that on HF. I still remember all the discussions I'd have on here with Rangers posters that were big on advanced stats that would try and convince me that while both were in St. Louis, Shattenkirk was better than Pietrangelo. They simply couldn't understand how usage would impact corsi.

Great points.

We’ve seen models and graphs that attempt to capture multiple scenarios, but they are way too messy.

This may sound kinda kooky, but until humans can perceive in more than three dimensions, it is best to break down each question individually and then compare one question to another to capture dimension.

 

[bleedblue1223]

Baseball is the place where advanced stats really took off and when that data became more public and user friendly, people started applying it to other sports. The thing is, baseball is perfect for advanced stats. It ultimately breaks down to a 1v1 matchup with the batter and pitcher, with a changing environment of defenses and ballparks, which can easily be quantified themselves. It doesn't mean the models can't be useful in hockey, they absolutely are, they are just way more complex for the average person to make an accurate conclusion from.

 

[BadgersandBlues]

I like these a lot and it is nice that the Athletic is finally making this data easily available for every player in one location. Cobbling them together from his various "best value, worst value" articles was annoying and his annual "here is how my model ranks everyone" article didn't do nearly as good a job as these cards do.

With that said, my problem with Dom's model is that (like most stat models) really struggles to quantify the value/impact of defensive roles. This model suggests that Parayko's defensive play has been worse than any other D man on the team. Not just this year, but also for the last 3 years. It is a modeling issue when 3 years of data reaches the conclusion that Parayko is barely an NHL caliber defensive player (his expected defensive performance based on the past 3 years has him just above "fringe NHLer"). I agree that he hasn't been at his best, but he just hasn't been the worst defender on the roster. Swap the usage of Krug and Parayko and I'd bet all the money I have that Krug would have allowed more goals and chances and wound up with a worse defensive score than Parayko has. It is incredibly difficult to balance "the situations a player is used in" against "how he has performed in those situations." His model struggles to reconcile those things in regards to defense.

None of his top 10 defenders from the start of the season began less than 40% of their shifts in the offensive zone and almost all of them start at least 55% of their shifts in the offensive zone. Only 2 of the 10 were below 50%. 4 of the 5 best defenders in the NHL according to his model played less than 22 minutes a night. If your coach gives you big minutes as a pure shutdown guy, you just aren't going to be rated high in his model. I'm absolutely not trying to throw out Dom's model as useless and I think it is the best publicly available analytics data in the sport. I just don't think that his market value accurately captures what defensive D men are worth. In general, I find his D valuations really difficult to predict and compare.

All in all, my eye test shares his valuation and performance of most the roster though.

It's the Achilles heel of all of the advanced stats models. So much of the "value" of a player is tied to production (be it actual results or expected results), they still haven't figured out a way to factor in usage and competition nearly to the point where the models make sense. According to Dom's model, Saad sucks defensively. Like, AHL level trash defensively. However, if you dig into the numbers, he's got the lowest offensive zone start % of any forward other then Sundqvist (39.6%), plays the second highest minutes against Elite competition of any forward (Only ROR has more) yet only has a -5 EV goal differential. Plus he averages about a minute on a top 5 PK in the league. That doesn't scream AHL quality defensive play to me. This also cuts both ways too, b/c ofc if all the above is true, you look worse on the offensive side as well then a person getting huge offensive zone starts and playing against lesser competition. Barbashev has had a great season - he's also getting fed much lesser competition (#1 against "middle" and "grit" both, and lowest against "elite" of almost all our forwards) plus he starts in the offensive zone 54% of the time. That's such an enormous difference in usage, but if you look at those cards, you'd think Barbashev is incredible and Saad is trash, which is only true if you don't look at any context - and ironically the whole point of advanced stats is to provide better context lol.

Edit: Holy lord we're starting Peru 82% of the time in the offensive zone? That's absurd.

 

[LGB]

With that said, my problem with Dom's model is that (like most stat models) really struggles to quantify the value/impact of defensive roles.

Have you considered that defensive roles are not as valuable/impactful as you previously thought?

 

[ChicagoBlues]

Baseball is the place where advanced stats really took off and when that data became more public and user friendly, people started applying it to other sports. The thing is, baseball is perfect for advanced stats. It ultimately breaks down to a 1v1 matchup with the batter and pitcher, with a changing environment of defenses and ballparks, which can easily be quantified themselves. It doesn't mean the models can't be useful in hockey, they absolutely are, they are just way more complex for the average person to make an accurate conclusion from.

Another great point. Hockey is far more dynamic than baseball.

I’d imagine basketball has similar advanced stat issues as hockey because it so dynamic.

 

[Brian39]

Have you considered that defensive roles are not as valuable/impactful as you previously thought?

Yup. And I've yet to see any evidence supporting that theory. I have seen an enormous amount of data suggesting otherwise though.

Let's briefly look at corsi for a moment. Ignoring players with less than 400 minutes at 5 on 5, 51 of the top 100 players in CF% have a zone start rate above 55%. 9 of the 100 have a zone start rate below 45%. The highest rated player by CF% with a zone start rate below 45% comes in at 47th on the list.

Similar results when looking at xGA per 60. 13 of the top 20 in the league had at least a 55% offensive zone start rate. No one in that group had an offensive zone start rate less than 45%.

Their is a clear and obvious correlation between offensive usage and strong analytic numbers. There is also the very common sense notion that it is harder to prevent a goal when you are starting in your own zone than when you are starting in the opponent's zone. I'm open to seeing evidence that any player can be equally successful in any role and that the correlation between role and performance is just correlation and not causation. But I haven't seen it yet.

 

[LGB]

Yup. And I've yet to see any evidence supporting that theory. I have seen an enormous amount of data suggesting otherwise though.

Let's briefly look at corsi for a moment. Ignoring players with less than 400 minutes at 5 on 5, 51 of the top 100 players in CF% have a zone start rate above 55%. 9 of the 100 have a zone start rate below 45%. The highest rated player by CF% with a zone start rate below 45% comes in at 47th on the list.

Similar results when looking at xGA per 60. 13 of the top 20 in the league had at least a 55% offensive zone start rate. No one in that group had an offensive zone start rate less than 45%.

Their is a clear and obvious correlation between offensive usage and strong analytic numbers. There is also the very common sense notion that it is harder to prevent a goal when you are starting in your own zone than when you are starting in the opponent's zone. I'm open to seeing evidence that any player can be equally successful in any role and that the correlation between role and performance is just correlation and not causation. But I haven't seen it yet.

Not the point I was trying to make although here is an article talking about why zone starts aren't as impactful on CF% as you might think. The main point being that the vast majority of the time players are starting their shift "on the fly" or in the NZ. I think the best argument for Parayko is that his partners have been sub-par. I think people tend to overrate quality of competition, because generally when the other team has their best players on the ice we will also have very good players on the ice, but this season Parayko's partners are definitely not top pair quality.

What I'm trying to say is that a great "offensive" player can have a greater impact on chances and goals than a great "defensive" player. This just comes down to the fact that you have more control over the play when your team has the puck. Defensive players will always be more dependent on factors they can't control. Therefore will have less of an impact on chances and goals as measured by these models.

 

[AjaxManifesto]

Are you scratching your head because you didn't bother to read the article, you struggled with the middle school level math involved, or because you are simply ignoring the information you didn't like to focus on a headline? It's super basic stuff. 90% of the population in the area has been vaccinated. Despite experiencing symptoms at a much lower rate, there are more total vaccinated people experiencing symptoms since the population is 9 times larger.

Let's equate it to a mandatory game of Russian Roulette. 90 participants are handed a revolver with 1 round in the chamber (and 5 empty slots). At a 1 in 6 chance (16.6%), you would expect 15 of these people to die. 10 participants are handed a gun with 3 bullets in the chamber (and 3 empty slots). At a 3 in 6 chance (50%), you would expect 5 of these people to die. You have to play. Are you grabbing the gun with 1 bullet in the chamber that has killed 15 people or the gun with 3 bullets in the chamber that has only killed 5?

Here are the exact stats laid out in the article:

"As of Thursday, 217 unvaccinated Albertans were in hospital with COVID-19, compared to 282 patients with at least one shot. But almost 90 per cent of the province has one dose, which means the "rate per 100,000 people" is a more telling metric to measure admissions. By that standard, about 24 unvaccinated people per 100,000 are in an Alberta hospital. That rate drops to fewer than eight after just one shot."

And regarding ICU patients:

"Unvaccinated ICU patients with COVID-19 far outnumber those who are vaccinated. The former tallied 47 in Alberta Thursday. The latter, 18. The numbers are just as imbalanced using proper metrics: 5.2 unvaccinated people per 100,000 are in the ICU. That rate drops to 0.4 with two doses."

Make sure you stay up to date on your boosters...

You seem triggered by links to articles and emojis

Maybe your 1st response should have been to ask a question: "Why the scratching head emoji?"

 

[joe galiba]

Not the point I was trying to make although here is an article talking about why zone starts aren't as impactful on CF% as you might think. The main point being that the vast majority of the time players are starting their shift "on the fly" or in the NZ. I think the best argument for Parayko is that his partners have been sub-par. I think people tend to overrate quality of competition, because generally when the other team has their best players on the ice we will also have very good players on the ice, but this season Parayko's partners are definitely not top pair quality.

What I'm trying to say is that a great "offensive" player can have a greater impact on chances and goals than a great "defensive" player. This just comes down to the fact that you have more control over the play when your team has the puck. Defensive players will always be more dependent on factors they can't control. Therefore will have less of an impact on chances and goals as measured by these models.

i think you are somehow assuming that good defensive players don’t possess the puck
great defensive players will control the play every bit as much as a great offensive player
when Rick Meagher played for the Blues he matched up against the best players all the time and the puck always wound up in the other teams zone and stayed there when he was on the ice
ROR is a not great defensive player just because of good defensive positioning, he also wins and controls the puck and it typically winds up in the other teams zone in the Blues possession
it is why he is a much better player than just his offense

 

[ChicagoBlues]

The key thing with any model of advanced stat breakdown is to compare players of similar usage, quality of linemates, and quality of competition. A lot of the advanced stat truthers will ignore all that on HF. I still remember all the discussions I'd have on here with Rangers posters that were big on advanced stats that would try and convince me that while both were in St. Louis, Shattenkirk was better than Pietrangelo. They simply couldn't understand how usage would impact corsi.

Control variables are fairly straightforward, but it's important to not attempt to control too many variables. Instead, what they'll do is set up alternate hypotheses and see if they can knock them out.

Besides that, defining the variables (similar usage, quality of linemates, quality of competition etc) is going to be the most difficult part of this. Operationalizing the variables is not impossible, but it's like a new field of study that hasn't accumulated enough literature for anyone to believe that the definitions are reliable and valid.

The best hockey advanced stats models are, imo, the simplest.

EXAMPLE:
Q1: similar usage vs quality of linemates = Q1A
Q2: quality of linemates vs quality of competition = Q2A
Q3: quality of competition vs similar usage = Q3A
Q4: etc

And then check for interactions between the results of each Q.

Advanced stats models that try to explain a lot in one or two graphs are probably missing the mark. Much better to break it all down into small studies and then measure for interactions.

So, without EVER having visited a website for advanced stats, I already know going into it that the good ones will keep it simple and present lots of data in simple but multiple formats, which can provide the context of dynamism in hockey. They are also going to provide definitions of the variables.

The bad advanced stats sites will try to capture the dynamism in a simple model and in as few graphics as possible. It won't work. Our perception is limited. Better said, our ability to present our fluid perceptions and make it understandable is limited.

 

[Celtic Note]

I like these a lot and it is nice that the Athletic is finally making this data easily available for every player in one location. Cobbling them together from his various "best value, worst value" articles was annoying and his annual "here is how my model ranks everyone" article didn't do nearly as good a job as these cards do.

With that said, my problem with Dom's model is that (like most stat models) really struggles to quantify the value/impact of defensive roles. This model suggests that Parayko's defensive play has been worse than any other D man on the team. Not just this year, but also for the last 3 years. It is a modeling issue when 3 years of data reaches the conclusion that Parayko is barely an NHL caliber defensive player (his expected defensive performance based on the past 3 years has him just above "fringe NHLer"). I agree that he hasn't been at his best, but he just hasn't been the worst defender on the roster. Swap the usage of Krug and Parayko and I'd bet all the money I have that Krug would have allowed more goals and chances and wound up with a worse defensive score than Parayko has. It is incredibly difficult to balance "the situations a player is used in" against "how he has performed in those situations." His model struggles to reconcile those things in regards to defense.

None of his top 10 defenders from the start of the season began less than 40% of their shifts in the offensive zone and almost all of them start at least 55% of their shifts in the offensive zone. Only 2 of the 10 were below 50%. 4 of the 5 best defenders in the NHL according to his model played less than 22 minutes a night. If your coach gives you big minutes as a pure shutdown guy, you just aren't going to be rated high in his model. I'm absolutely not trying to throw out Dom's model as useless and I think it is the best publicly available analytics data in the sport. I just don't think that his market value accurately captures what defensive D men are worth. In general, I find his D valuations really difficult to predict and compare.

All in all, my eye test shares his valuation and performance of most the roster though.

I don’t disagree that usage has its wrench in spokes type impact on these models, but even by Parayko’s defensive abilities he has performed poorly defensively the last 3 years. And, while usage impacts players stats, players are typically used to their strong suits, so if their play shows poorly when they are being played in positions based on their strengths, maybe the model is less broken than either the player or the coach putting them in a position where they don’t succeed?

 

[LGB]

i think you are somehow assuming that good defensive players don’t possess the puck
great defensive players will control the play every bit as much as a great offensive player
when Rick Meagher played for the Blues he matched up against the best players all the time and the puck always wound up in the other teams zone and stayed there when he was on the ice
ROR is a not great defensive player just because of good defensive positioning, he also wins and controls the puck and it typically winds up in the other teams zone in the Blues possession
it is why he is a much better player than just his offense

The best defensive players should possess the puck a lot. ROR is a great example of a great defensive player who can transition his defense into offense. I was just speaking in general. I think there are a lot of players that are perceived as great defenders, but who aren't as great statistically because they are defending too much.

 

[Renard]

It's the Achilles heel of all of the advanced stats models. So much of the "value" of a player is tied to production (be it actual results or expected results), they still haven't figured out a way to factor in usage and competition nearly to the point where the models make sense. According to Dom's model, Saad sucks defensively. Like, AHL level trash defensively. However, if you dig into the numbers, he's got the lowest offensive zone start % of any forward other then Sundqvist (39.6%), plays the second highest minutes against Elite competition of any forward (Only ROR has more) yet only has a -5 EV goal differential. Plus he averages about a minute on a top 5 PK in the league. That doesn't scream AHL quality defensive play to me. This also cuts both ways too, b/c ofc if all the above is true, you look worse on the offensive side as well then a person getting huge offensive zone starts and playing against lesser competition. Barbashev has had a great season - he's also getting fed much lesser competition (#1 against "middle" and "grit" both, and lowest against "elite" of almost all our forwards) plus he starts in the offensive zone 54% of the time. That's such an enormous difference in usage, but if you look at those cards, you'd think Barbashev is incredible and Saad is trash, which is only true if you don't look at any context - and ironically the whole point of advanced stats is to provide better context lol.

Edit: Holy lord we're starting Peru 82% of the time in the offensive zone? That's absurd.

On the subject of Saad, I rewatched the Dallas game yesterday, paying particular attention to Saad. ( I had called Saad floater and people disagreed because he was used in a defensive capacity)

Saad was used mainly with Sundqvist. Sunny was centering the third line, which I suppose is our checking line.

The best thing I can say about Saad is that he was positionally responsible. He knew where he was supposed to be and he went there.. Beyond that, he never seemed to be out of breath.. I don't remember seeing seeing him get up from the ice. I wonder if he needed a shower after the game.

 

[tfriede2]

Saad is one of those guys who, like a steady defenseman, you don’t really notice throughout the game, until he scores because he has good hockey sense and knows where to be. I’m glad he’s on the team.

 

[Brian39]

Not the point I was trying to make although here is an article talking about why zone starts aren't as impactful on CF% as you might think. The main point being that the vast majority of the time players are starting their shift "on the fly" or in the NZ. I think the best argument for Parayko is that his partners have been sub-par. I think people tend to overrate quality of competition, because generally when the other team has their best players on the ice we will also have very good players on the ice, but this season Parayko's partners are definitely not top pair quality.

What I'm trying to say is that a great "offensive" player can have a greater impact on chances and goals than a great "defensive" player. This just comes down to the fact that you have more control over the play when your team has the puck. Defensive players will always be more dependent on factors they can't control. Therefore will have less of an impact on chances and goals as measured by these models.

I vehemently disagree with the conclusion drawn in the article because it is the result of a flawed premise. His conclusion is a result of the assumption that all on-the-fly starts are all created equal. His entire conclusion is that even though an increase in defensive zone starts drastically lowers your expected CF% on those shifts, they don't meaningfully impact your Corsi overall because they make up such a small minority of the player's deployment. He reached this conclusion by assuming that all on-the-fly shifts are created equal and he plugged in a 50.8% expected CF for every shift every player took so long as it didn't begin with a faceoff. Because these 50.8% expected CF% shifts made up the vast majority of the total ice time (and thus the large percentage of total on ice shot attempts), weighting the defensive draws didn't move the numbers much overall.

But it is wildly inaccurate to suggest that all on-the-fly starts are created equal. NHL coaches don't line match their D pairs for faceoffs and then just rotate guys with no thought on-the-fly. When the Blues dump and go for a change and the opponent's top line comes out for a controlled breakout, we are putting out the shutdown pair if they had only been off the ice for 50 seconds but Perunovich had been off the ice for 1:30. If we were instead able to breakout, get fresh forwards out and gain the line with possession, we are much more likely to throw out Perunovich to try and keep that possession and drive offense. A huge number of "who is going out next" decisions are made within 5 seconds of a guy coming off the ice and are driven by where the puck is, who has the puck, and who the opponent has on the ice. There is an assistant coach whose primary job is to make those decisions in real time and his entire goal is to get your offensive minded D men out there when the play is tilting toward offense and your defensive-skilled guys out there when it is tilting toward defending. This is exactly how you get to a big gap in ice time. Your better pairs have more situations where you are comfortable using them while your worst pairing has fewer. So there are a bunch of times where the worst pair gets skipped over because the situation is deemed too risky for them.

The only way his conclusion is accurate is if you believe that the only difference in usage from Scandella and Perunovich is who starts a shift for a D zone draw vs an O zone draw. I am not aware of any model or stat company that has attempted to measure or track expected results from on-the-fly deployment. It would be highly time consuming because it can't be scraped from NHL stats the way shot attempts and faceoffs can. I assume that this is one of the things that team's in-house analytics departments are doing, but none of that info is public. But from decades of watching the NHL and a few years playing vaguely competitive hockey, I can say with 100% certainty that faceoff location isn't the only difference in usage between the D men who make up a 6 man unit.

Zone start rate isn't a perfect measure of that usage (just like corsi isn't a perfect measure of possession), but it is the best stat we have. Coaches who are throwing a guy out there for a 60-70% D zone start rate are almost always also using that guy as a shut down D man on the fly.

 

[bleedblue1223]

Yeah, that whole flawed argument of zone starts don't matter because on the fly starts account for way more of a player's usage and those can just be equalized because for some reason a coach has no control over lines during on the fly changes because they will just go down the line and play each line equally is an exact rehash of the arguments I'd have on here with those Rangers fans about Petro and Shattenkirk.

I put together a chart that plotted CF% and zone starts, and the correlation is clear. I don't think I have it anymore since it was years ago. Another flawed argument they pulled was comparing CF% rel from a player on one team to a player on another team and thinking it had any relevance. It was clear they had no statistical background at all.

 

[Brian39]

What I'm trying to say is that a great "offensive" player can have a greater impact on chances and goals than a great "defensive" player. This just comes down to the fact that you have more control over the play when your team has the puck. Defensive players will always be more dependent on factors they can't control. Therefore will have less of an impact on chances and goals as measured by these models.

I wanted to discuss this point separately from my last post.

I agree that defensive statistics are more dependent on factors outside of a defender's control. And I agree that as a result they will have less of an impact on chances and goals as measured by these models. But my opinion/belief given those assumptions is that the models don't accurately capture the value of defensive players nearly as well as they do offensive players. I don't believe that the actual value of defensive players is less. I don't think that players who excel at defending are less valuable to teams because there are more factors out of their control. I don't think that allocating all your resources on a bunch of offensive skillset guys at the expense of defensive skillsets will make you a better team.

 

[LGB]

I vehemently disagree with the conclusion drawn in the article because it is the result of a flawed premise. His conclusion is a result of the assumption that all on-the-fly starts are all created equal. His entire conclusion is that even though an increase in defensive zone starts drastically lowers your expected CF% on those shifts, they don't meaningfully impact your Corsi overall because they make up such a small minority of the player's deployment. He reached this conclusion by assuming that all on-the-fly shifts are created equal and he plugged in a 50.8% expected CF for every shift every player took so long as it didn't begin with a faceoff. Because these 50.8% expected CF% shifts made up the vast majority of the total ice time (and thus the large percentage of total on ice shot attempts), weighting the defensive draws didn't move the numbers much overall.

But it is wildly inaccurate to suggest that all on-the-fly starts are created equal. NHL coaches don't line match their D pairs for faceoffs and then just rotate guys with no thought on-the-fly. When the Blues dump and go for a change and the opponent's top line comes out for a controlled breakout, we are putting out the shutdown pair if they had only been off the ice for 50 seconds but Perunovich had been off the ice for 1:30. If we were instead able to breakout, get fresh forwards out and gain the line with possession, we are much more likely to throw out Perunovich to try and keep that possession and drive offense. A huge number of "who is going out next" decisions are made within 5 seconds of a guy coming off the ice and are driven by where the puck is, who has the puck, and who the opponent has on the ice. There is an assistant coach whose primary job is to make those decisions in real time and his entire goal is to get your offensive minded D men out there when the play is tilting toward offense and your defensive-skilled guys out there when it is tilting toward defending. This is exactly how you get to a big gap in ice time. Your better pairs have more situations where you are comfortable using them while your worst pairing has fewer. So there are a bunch of times where the worst pair gets skipped over because the situation is deemed too risky for them.

The only way his conclusion is accurate is if you believe that the only difference in usage from Scandella and Perunovich is who starts a shift for a D zone draw vs an O zone draw. I am not aware of any model or stat company that has attempted to measure or track expected results from on-the-fly deployment. It would be highly time consuming because it can't be scraped from NHL stats the way shot attempts and faceoffs can. I assume that this is one of the things that team's in-house analytics departments are doing, but none of that info is public. But from decades of watching the NHL and a few years playing vaguely competitive hockey, I can say with 100% certainty that faceoff location isn't the only difference in usage between the D men who make up a 6 man unit.

Zone start rate isn't a perfect measure of that usage (just like corsi isn't a perfect measure of possession), but it is the best stat we have. Coaches who are throwing a guy out there for a 60-70% D zone start rate are almost always also using that guy as a shut down D man on the fly.

Fair argument. I still think in the aggregate on the fly starts can be assumed to be about neutral in terms of possession. I'll do some more research though.

 

[bleedblue1223]

Fair argument. I still think in the aggregate on the fly starts can be assumed to be about neutral in terms of possession. I'll do some more research though.

The play is always flowing in one direction. A change could happen when a team is cycling and hemming the defense in, that's an on the fly change where it's clearly offensive. Another change could be when the opposition gathers possession behind their net and waits for their own team to change lines, that's a case of an on the fly being change being defensive. Coaches are cognizant of this and will try and get their offensive players in as many offensive situations as possible and their shutdown line in as many situations where they are matched up with the oppositions top offensive line.

We could review the tape and breakdown each of the on the fly starts for each player and figure out if we'd categorize them as offensive or defensive, but I don't think anyone is going to go through the tape and do that.

 

[BlueKnight]

A new article from JR on the Athletic about Armstrong. While I'm sometimes critical and down on Army I much rather have him over someone else. We are so are fortunate to have him here.

 

[LGB]

I wanted to discuss this point separately from my last post.

I agree that defensive statistics are more dependent on factors outside of a defender's control. And I agree that as a result they will have less of an impact on chances and goals as measured by these models. But my opinion/belief given those assumptions is that the models don't accurately capture the value of defensive players nearly as well as they do offensive players. I don't believe that the actual value of defensive players is less. I don't think that players who excel at defending are less valuable to teams because there are more factors out of their control. I don't think that allocating all your resources on a bunch of offensive skillset guys at the expense of defensive skillsets will make you a better team.

If you accept that they are more dependent on factors outside of their control it follows that they would have less of an individual impact.

 

[LGB]

The play is always flowing in one direction. A change could happen when a team is cycling and hemming the defense in, that's an on the fly change where it's clearly offensive. Another change could be when the opposition gathers possession behind their net and waits for their own team to change lines, that's a case of an on the fly being change being defensive. Coaches are cognizant of this and will try and get their offensive players in as many offensive situations as possible and their shutdown line in as many situations where they are matched up with the oppositions top offensive line.

We could review the tape and breakdown each of the on the fly starts for each player and figure out if we'd categorize them as offensive or defensive, but I don't think anyone is going to go through the tape and do that.

I accept that not every on the fly start is the same. Just don't think there would be a significant difference overall. Like I said though I'll do more research about usage's impact on chances and goals.

 

[ChicagoBlues]

I like these a lot and it is nice that the Athletic is finally making this data easily available for every player in one location. Cobbling them together from his various "best value, worst value" articles was annoying and his annual "here is how my model ranks everyone" article didn't do nearly as good a job as these cards do.

With that said, my problem with Dom's model is that (like most stat models) really struggles to quantify the value/impact of defensive roles. This model suggests that Parayko's defensive play has been worse than any other D man on the team. Not just this year, but also for the last 3 years. It is a modeling issue when 3 years of data reaches the conclusion that Parayko is barely an NHL caliber defensive player (his expected defensive performance based on the past 3 years has him just above "fringe NHLer"). I agree that he hasn't been at his best, but he just hasn't been the worst defender on the roster. Swap the usage of Krug and Parayko and I'd bet all the money I have that Krug would have allowed more goals and chances and wound up with a worse defensive score than Parayko has. It is incredibly difficult to balance "the situations a player is used in" against "how he has performed in those situations." His model struggles to reconcile those things in regards to defense.

None of his top 10 defenders from the start of the season began less than 40% of their shifts in the offensive zone and almost all of them start at least 55% of their shifts in the offensive zone. Only 2 of the 10 were below 50%. 4 of the 5 best defenders in the NHL according to his model played less than 22 minutes a night. If your coach gives you big minutes as a pure shutdown guy, you just aren't going to be rated high in his model. I'm absolutely not trying to throw out Dom's model as useless and I think it is the best publicly available analytics data in the sport. I just don't think that his market value accurately captures what defensive D men are worth. In general, I find his D valuations really difficult to predict and compare.

All in all, my eye test shares his valuation and performance of most the roster though.

In about an hour and a half, I'm joining a Zoom meeting for a dear friend's dissertation defense.

Her research is qualitative in nature, so coding (quantifying) the findings was her biggest challenge. As you know, coding assigns numbers (values) to answers of questions or player usage scenarios or whatever.

It's easy to see how messy these advanced stats models are in sports.

Dom, et al. are doing groundbreaking work in the coding of dynamic, qualitative situations and scenarios in order for us to better understand a hockey player and predict their future success.

Cuz that's what this is all about, right? The likelihood of future success.

If there are not any already, I can see conferences on advanced stats in sports at some point.

 

[joe galiba]

The best defensive players should possess the puck a lot. ROR is a great example of a great defensive player who can transition his defense into offense. I was just speaking in general. I think there are a lot of players that are perceived as great defenders, but who aren't as great statistically because they are defending too much.

Rick Meagher, in his Selke season, would have been on the ice against players that year included Savard, Lemieux, Cullen, Yzerman, Gretzky, Nicholls, Francis, Sakic, Stastny, Howerchuck, Nieuwendyk, Gilmour, Janney, Turgeon, Messier, LaFontaine,
basically the Golden Age of centers
He had 20 even strength points that year and was a +4
so when he was on the ice he was used against these players and they essentially didn't score against him in an incredibly high scoring era
Guy Carbonneau was another defensive center from the same era
he scored at a 40 something to 50 something pace when the top centers were putting up 100+, and completely shut them down

great defenders have skill - it is not allowing other teams to score when they are on the ice
there is just no easy way to see statistical info to evaluate it

 

[bleedblue1223]

The same issues you see with getting a quantitative evaluation of a defensive player in hockey, you see with getting a proper evaluation of a catcher in baseball. That's why you see catcher's WAR vary from site to site and generally not be in align with other positions with players of similar tier. It's tough to quantify a catcher's value when it comes to how they manage their pitchers, which is probably their most important skill.