Speculation: 2020-21 News/Rumors/Roster Thread Part III

[Raccoon Jesus]

Just so I don't have to quote everybody, could somebody please provide some instances where they believe advanced analytics have been properly applied and changed anything, big or small?

I see fans, reporters and broadcasters consistently misuse numbers to justify opinions without context, cannot recall a single instance where they have been applied in a positive manner.

The championship Kings were one of the first teams heavily using analytics, including something they called injury analytics. You don't even have to look beyond this franchise.

 

[SettlementRichie10]

Hockey is an organic game with a lot of variance and not nearly the amount of predictable structure of American football or baseball. What advanced stats we do have in hockey are valuable because they at least attempt to divine some underlying meaning to the game. Possession stats will always be illuminating in my eyes.

Don’t forget that possession stats like score adjusted Fenwick predicted the Kings dominance in 2012 long before the playoffs actually started.

 

[YP44]

Just so I don't have to quote everybody, could somebody please provide some instances where they believe advanced analytics have been properly applied and changed anything, big or small?

I see fans, reporters and broadcasters consistently misuse numbers to justify opinions without context, cannot recall a single instance where they have been applied in a positive manner.

I think the way GM's hand out contracts and managed their teams has changed greatly changed

 

[crassbonanza]

The championship Kings were one of the first teams heavily using analytics, including something they called injury analytics. You don't even have to look beyond this franchise.

I think the Kings are both a great example of utilizing analytics(2012-2014) and an example of why you can not over rely on analytics(2015-2017). From 2012-2014 the Kings led the league in CF% and xGF% and were able to carry that over to a couple of cups. From 2015-2017 the Kings led the league in CF% and xGF% and that carried them to a single playoff win and missing the playoffs twice. Analytics do provide some fun information, but they are not overly predictive in my opinion.

 

[Raccoon Jesus]

I think the Kings are both a great example of utilizing analytics(2012-2014) and an example of why you can not over rely on analytics(2015-2017). From 2012-2014 the Kings led the league in CF% and xGF% and were able to carry that over to a couple of cups. From 2015-2017 the Kings led the league in CF% and xGF% and that carried them to a single playoff win and missing the playoffs twice. Analytics do provide some fun information, but they are not overly predictive in my opinion.

No offense intended but the fact is the correlation is there.

But again, it should come as no surprise that "team that controls the lions share of the shot attempts will win more often than not."

I think everyone will agree that the 2015-2017ish Kings were a good possession team that didn't have the shooting/playmaking talent to take advantage of it, as well as one that often self-sabotaged in the defensive end. Kinda like we said about Muzzin before, not all shots are created equal--if we're taking 15 outside shots down at one end but we completely break down and the opponent gets a couple of shots in the slot, all that possession is for naught.

 

[bland]

The championship Kings were one of the first teams heavily using analytics, including something they called injury analytics. You don't even have to look beyond this franchise.

You mean the analytics that were misused to promote the mistaken theory that they were still a contender when they clearly weren't?

That did far more harm than good to the franchise. The statisticians were looking for a reason to explain the Kings success and settled on corsi, not the other way around. Sutter hockey had been around for ages, and the data never could explain the difference between the teams with Richards, Mitchell and Voynov than those without them despite the fancy numbers being very similar. It just underlines my point, corsi doesn't mean a damn thing between winning and losing.

 

[Raccoon Jesus]

You mean the analytics that were misused to promote the mistaken theory that they were still a contender when they clearly weren't?

That did far more harm than good to the franchise. The statisticians were looking for a reason to explain the Kings success and settled on corsi, not the other way around. Sutter hockey had been around for ages, and the data never could explain the difference between the teams with Richards, Mitchell and Voynov than those without them despite the fancy numbers being very similar. It just underlines my point, corsi doesn't mean a damn thing between winning and losing.

Don't ask questions you don't want the answers to, then.

Gonna step out of this bad-faith discussion, your self-assuredness makes it impossible.

 

[cyclones22]

The difference between those 2 different sets of 3 year periods is that in 2012 to 2014 there was a combination of elite defense and elite possession. They could get away with winning via timely goals. Starting 2015 is when the elite possession sheltered the defense. Both Willie Mitchell and Voynov were gone in 2015. In 2016 Regehr retired and Greene missed almost the entire season. The system made that defense look statistically better than it was. You want to induce vomiting? See that during that 102 point season in 2016 that Jamie McBain and Christian Erhoff got 40+ games each. The corpse of Rob Scuderi got 21 games. All 3 of them retired the next season. They weren't really NHL players when the Kings acquired them. Luke Schenn was getting second pairing minutes. Good grief that defense was the beginning of the end of the Kings defensive identity.

 

[Yog S'loth]

The corpse of Rob Scuderi got 21 games. All 3 of them retired the next season.

I would just like to point out that Scuderi didn't technically retire.

He got one of those "fight club" jobs that you don't need to come into the office for; you can do this job from home.

 

[crassbonanza]

No offense intended but the fact is the correlation is there.

I guess I just disagree with how much correlation there actually is beyond good teams tend to outshoot bad teams.

If you look at the top 10 in xGF% from the past few seasons you find a pretty wide range of results:

2020 had a cup winner(Bolts), one team that fully missed the playoffs(Kings) and two teams that would have missed the playoffs(Habs, Wild).
2019 had a cup winner(Blues) and two teams that missed the playoffs(Habs, Wild).
2018 had four teams that fully missed the playoffs(Canes, Flames, Stars, Blues).

If you look at strictly SF% you see a similar range of results:

2020 had one team that fully missed the playoffs(Kings) and two teams who would have (Preds, Habs).
2019 had a cup winner(Blues), and two teams who missed the playoffs(Habs, Wild).
2018 had four teams that fully missed the playoffs(Canes, Flames, Blues, Hawks).

We also see that the expected stats and PDO seem to be unable to compensate for talent. Assuming a result based on a shot in a particular spot on the ice while ignoring the competition, position of players and the player taking the shot is just flawed in my opinion.

There was a quote from somebody in the Flames org(DJ maybe?) who said that the online analytics we have available to us is like turning the light on for a second when compared to what teams have. I am sure they have developed some wild stuff, but I do believe that hockey is too fluid to fully be able to condense the game to a spreadsheet. It works for baseball because there is always a singular event, pitcher vs batter, and almost no external factors.

 

[Raccoon Jesus]

I guess I just disagree with how much correlation there actually is beyond good teams tend to outshoot bad teams.

If you look at the top 10 in xGF% from the past few seasons you find a pretty wide range of results:

2020 had a cup winner(Bolts), one team that fully missed the playoffs(Kings) and two teams that would have missed the playoffs(Habs, Wild).
2019 had a cup winner(Blues) and two teams that missed the playoffs(Habs, Wild).
2018 had four teams that fully missed the playoffs(Canes, Flames, Stars, Blues).

If you look at strictly SF% you see a similar range of results:

2020 had one team that fully missed the playoffs(Kings) and two teams who would have (Preds, Habs).
2019 had a cup winner(Blues), and two teams who missed the playoffs(Habs, Wild).
2018 had four teams that fully missed the playoffs(Canes, Flames, Blues, Hawks).

We also see that the expected stats and PDO seem to be unable to compensate for talent. Assuming a result based on a shot in a particular spot on the ice while ignoring the competition, position of players and the player taking the shot is just flawed in my opinion.

There was a quote from somebody in the Flames org(DJ maybe?) who said that the online analytics we have available to us is like turning the light on for a second when compared to what teams have. I am sure they have developed some wild stuff, but I do believe that hockey is too fluid to fully be able to condense the game to a spreadsheet. It works for baseball because there is always a singular event, pitcher vs batter, and almost no external factors.

To be clear, I don't like the expected stats at all, xgf and others. And I agree they aren't perfect and we saw how unable they were to compensate for (ranges of, really) talent.

But outliers happen regardless, I don't think you can suggest outliers blow up the entire thing. No one is suggesting they're perfect, either. It's not a 1-to-1 "do this and you'll make the playoffs" otherwise those 2015-2017 Kings would have been in the mix.

With CF specifically the thing is you're not just counting shots, its all shot attempts. So it's a proxy for how much time and effort you're spending in each zone (remember that it initially started as a goalie workload stat). If you're doing that nearly 60% of the time for 82 games, you should be in a good spot (and those teams mentioned often had a crucial flaw in the system--bad shooting talent/luck, bad goaltending talent/luck, defensive breakdowns, often easy to discern).

I have no doubt the teams have much better than us. We're just working with proxies. They now have full access to full tracking. Hell even some of the analysts are working with Sportlogiq and can tell you player locations at all times and have all sorts of proprietary things.

Ultimately I don't even want to dive this deep, haha. I think my thing is I don't understand how people can be so dismissive, imo it's pretty important to reconcile the eye test with the stats so you can question one or the other. I'm kind of over discussions of predictive value/application because people are so dug in, it's not productive talk. I just don't think you can throw out the baby with the bath water, that's all.

I think anyone could have told you the 2nd line got caved in vs. the Blues, especially night one. But a stat illustrating to what degree is worth discussion--25% CF% is awful, and a trend that had been going on for a few games. To the surprise of no one, all three were -2 last night. I didn't need a stat to tell me they were doing poorly, but it sure helped illuminate to what degree, even before the actual goals against started. Don't see why that info makes anyone grumpy.

 

[DoktorJeep]

Current hockey analytics models are at best signals and at worst novelties. They use secondary data (shots, etc.) and rudimentary statistical methods to come up with insights and predictions. But they are a good start and can be valuable input when paired with human experts.

I think once player and puck tracking technologies mature to be in multiple leagues, you’ll be able to apply more advanced algorithms and deep learning to mine for insights and predictions.

 

[Fishhead]

The usefulness of advanced stats will vary with importance as the playstyle of the NHL goes through cycles. Corsi was an excellent predictor of success from the mid 2000s until the game started shifting away from a bruising, defensive style. There was a period there where cup winners were consistently in the top 5 in CF%. Once the game started migrating towards skill and speed, possession lost a little of it's importance. Pinning a team in and controlling the zone was still effective, but lessened by the fact that teams were starting to rely more on rush scoring with high speed.

Like all stats it needs context, but the more the game slows down the more important possession stats become.

 

[lumbergh]

The most useful stat is goals for vs. goals against, aka Goals For % (GF%). It generally will tell you who is gonna win, because the team with a high GF% generally outscores the other team.

The only "advanced stat" that has a strong correlation with the goals for % is PDO (shooting percentage + save percentage). It doesn't take an advanced degree (although I have one) to figure that out. This is important: PDO is NOT an indication of "puck luck" or whatever that is. PDO indicates how good your team is at outscoring the other team.

You can determine correlation coefficients, or how well something tracks with something else, on Excel using the data from Natural Stat Trick, but I'll just give you a synopsis:
Correlation coefficients with GF%
PDO >0.8 (strong correlation)
Corsi For % ~0.4 (very weak)
Fenwick For % < 0.4 (little correlation)
Shots For % ~0.4 (very weak)
Expected Goals For % (xGF%) ~0.5 (weak)
Scoring chances for % (SCF%) ~0.5 (weak)

That bold part tells you that the person who invented xGF% only has a mild understanding of what makes a good player or a good team. All the other stats are only meaningful in telling you about possession, but possession is only mildly correlated with the score.

This is all to say, advanced stats, as they currently exist, are only mildly indicative of quality of a team or a player. Take them with the grain of salt they deserve. Put all your weight on them, and you will fall over spectacularly, because they are NOT dependable.

 

[bland]

Don't ask questions you don't want the answers to, then.

Gonna step out of this bad-faith discussion, your self-assuredness makes it impossible.

Just looking for an actual instance of analytics being used properly. This was not one.

I am thoroughly confident here because I have already considered, debated and dismissed some of the arguments made while discussing this topic over the years.

If there is evidence otherwise I would like to hear it. But being the quoted "corsi darlings" didn't have a damn thing to do with the Kings Cups, focused team energy, passion and hard-headed accountability won those Cups, and the only stat I have ever seen that reflects it is Mike Richards CV.

 

[King'sPawn]

Just looking for an actual instance of analytics being used properly. This was not one.

I am thoroughly confident here because I have already considered, debated and dismissed some of the arguments made while discussing this topic over the years.

If there is evidence otherwise I would like to hear it. But being the quoted "corsi darlings" didn't have a damn thing to do with the Kings Cups, focused team energy, passion and hard-headed accountability won those Cups, and the only stat I have ever seen that reflects it is Mike Richards CV.

How about from the mouth of an actual executive?

The unsung hero in the process, if you will, is the analytical component.
It’s no secret that analytics have an increasing role in how hockey teams are managed, and the Kings have built a strong presence within the field, led by the work of Rob Vollman within their hockey operations team.
When the analytics and scouts come to the same conclusion, you’re made in the shade.
“A scouting opinion doesn’t factor into the analytics, and the analytics don’t factor into the scouts,” Yannetti said. “If they jive and we mesh, there’s not really another step of the process after that. If our analytics and our scouting tell us the same thing, we feel very confident and the success rate when those two things happen is quite good.”
When they don’t say the same thing, however, it’s perhaps just as beneficial, because it forces Yannetti and his team to pause and take a deeper dive into why there is a difference of opinion between the two.
Sometimes, it’s because analytics isn’t able to identify an elite trait, such as skating, through the numbers. Sometimes, however, it’s the opposite way in surfacing that scouting reports haven’t seen a player enough, or seen a player in a wide enough variety of settings
“We look and say why are the analytics saying something that the scouting side isn’t,” Yannetti explained. “You can usually see where the disconnect is, you can usually see where further work is needed, or a bias popped in. Sometimes it can be something as simple as the analytics didn’t factor in skating, because it can’t, or as simple as the scout didn’t see the games against his peer group. It could be simple things like that.”

Talking Kings historical success throughout the draft with Mark Yannetti - LA Kings Insider

If you want specifics, you might be able to ask him directly. You clearly know more than him since it's such a useless tool and you dedicate your analysis so much more effectively than he.

 

[BigKing]

All I know is that I can look at Matt Roy's old-timey statistics and see two hits per game and a team-leading +4 with 18 minutes of ice a game and have them back-up the eye test.

 

[All The Kings Men]

This right here is my problem with "advanced metrics"

Corsi is just a slightly "better" way of tracking Shots on Goal, Fenwick is just that minus any shots the other team blocks, PDO is just save% + shooting %

There's nothing "advanced" about most of them. So the push back on them, to me, is rooted in a lack of effective communication.

The stats like Goals Above Replacement or Wins Above Replacement or Expected Goals etc, the ones that require equations and complex explanations.... those are the ones I understand the pushback on.

If you think goals, assists, shots on goal, hits etc are valuable tools for evaluating a players impact on a game but don't see the value in corsi, fenwick, SF%, PDO etc... well you're either not understanding how simple those stats are or you're being obstinate.

 

[Steve Zissou]

I see you guys trying to pad your so-called 'STATS'....

 

[BringTheReign]

Sort of a random entry to the discussion, but one area of advanced stats I find particularly intriguing are zone exit/entries and, even fancier, successful passes made out of the d-zone or once in the o-zone. Anyone who has played or coached knows how important puck management is within 10 feet of either blueline, and these manually tracked stats seem incredibly valuable to me.

Its flaw is that some systems openly discourage puck-carrying and prefer dump-and-chase, but it is an excellent way to show a player's value with the puck on their stick given the right team structure.

 

[BringTheReign]

This right here is my problem with "advanced metrics"

Corsi is just a slightly "better" way of tracking Shots on Goal, Fenwick is just that minus any shots the other team blocks, PDO is just save% + shooting %

There's nothing "advanced" about most of them. So the push back on them, to me, is rooted in a lack of effective communication.

The stats like Goals Above Replacement or Wins Above Replacement or Expected Goals etc, the ones that require equations and complex explanations.... those are the ones I understand the pushback on.

If you think goals, assists, shots on goal, hits etc are valuable tools for evaluating a players impact on a game but don't see the value in corsi, fenwick, SF%, PDO etc... well you're either not understanding how simple those stats are or you're being obstinate.

This is a common issue in the data science/analytics field in general, not just in sports. You see tons of entry-level scientists/analysts try to throw deep learning at an issue that only needs simple descriptive statistics. The former can be incredibly powerful if used appropriately with the correct amount of humility, but it will almost always fail if it skips first principles.

 

[Lt Dan]

I see you guys trying to pad your so-called 'STATS'....

That's an all you can eat buffet of rebounds

 

[Basilisk]

Did anybody else read the Morris Lukowich story? It's here.....

Catching up with Kings alum Morris Lukowich - LA Kings Insider


I have great, great memories of watching this on TV. Back then, it was 6-4 league. So any game that was 1-1 after regulation was special, indeed. Dave Taylor scored our other goal. I'm going to say this was easily one of the top 5 Kings games I ever watched......

 

[King'sPawn]

This right here is my problem with "advanced metrics"

Corsi is just a slightly "better" way of tracking Shots on Goal, Fenwick is just that minus any shots the other team blocks, PDO is just save% + shooting %

There's nothing "advanced" about most of them. So the push back on them, to me, is rooted in a lack of effective communication.

The stats like Goals Above Replacement or Wins Above Replacement or Expected Goals etc, the ones that require equations and complex explanations.... those are the ones I understand the pushback on.

If you think goals, assists, shots on goal, hits etc are valuable tools for evaluating a players impact on a game but don't see the value in corsi, fenwick, SF%, PDO etc... well you're either not understanding how simple those stats are or you're being obstinate.

I have mixed feelings on some of the analytics (I hate saying "advanced stats", because I think people have self-aggrandizing motives behind the term).

I'm not a fan of the "Above Replacement" stats. I think statistics, when used well, can adequately put people in a grouping without worrying about WAR to make a point.

I like some basic equations, like /60, because it provides some context and levels the playing field.

xGF and xGA is essentially scoring chances or shots from high danger areas. I think it can be valid.

The problem, to me, is analytics are still very young. Skating speed, shot type, shot location (where the shot goes), and other unforeseeable ways of measuring performance aren't public yet.

For example, does a certain player fail to make controlled zone entries because he's skating too fast and loses control of the puck? Is a player constantly taking high danger shots, but it's a backhand into the goalie's midsection?

I know you weren't polling for opinions. For the most part I agree with you. I think that reasonable equations can provide good data if utilized and interpreted well.

 

[Docgonzo]

So I’m watching the Senators/Canucks and it got me thinking about where the Kings are in their rebuild compared to other teams when they went through them.

Obviously talent is important in prospects because if you don’t have talent your rebuild won’t go far.

But unlike Ottawa, Edmonton, Buffalo, etc is unlike them who bring in Vets as mercenaries or have those like 3-4th line grinder Vets. We have Stanley Cup campion Vets 4 out 5 who’ve only ever played for this team.

I mean hearing Andersson talking about how much he’s watched and admired Kopitar growing up. It makes it so these guys can help the prospects grow more than say a NHL journeyman brought in to help the kids learn the game. Think Chicago is in the same spot but their drafting pool just isn’t as strong yet.