Are possession stats the best way to evaluate a player

[BaseballCoach]

All summer long there have been tens of thousands of posts written about the merits of various moves made by the Habs. Inevitably people post numbers to back up their argument.

As the summer progressed, I became increasingly uncomfortable with the framework within which we were having all of the discussion.

It seems that there is an unspoken assumption that is assumed by many to be correct: that possession stats are the most important stats we can refer to, and should ALWAYS refer to, even for defencemen.

At the same time, the old-fashioned +/- stat is frowned upon as so deeply flawed that one must never use it, and anyone who criticizes its use need not even say why. This viewpoint is considered by some as a kind of "established science", and if you refer to +/- you are looked at like a dinosaur-like climate change denier.

I'd like to have a real debate on the way we measure success in hockey players, especially defencemen, without the emotion of tying this debate to specific players, GMs, coaches, etc.

I start with a hypothetical story, a discussion between a coach and a player.

Why? Here's what I know. If a certain player of mine would be on the ice for A LOT of goals against, I would want to know why. And I won't give him a free pass just because he is also on the ice for a lot of weak shots taken by his team from bad angles at the other end of the ice.

"Hey Johnny, why is it you are on the ice for 60% of our goals against when you only play 50% of the minutes?"

"Coach, that's not fair, my Corsi is excellent."

"Johnny Boy, I don't give a crap about your Corsi, if it's only good because Paul Byron thinks he's doing me a favour by throwing pucks weakly at the net with no chance of going in. What I want to know about is not your Corsi, but my I-see. Why do I see the other team keep getting open in the slot when you are on the ice, and how come you almost never block one of those dangerous shots?"

I contend that possession stats based on shots on goal without regard to how dangerous those shots are, by either team, will yield results that are not correct in measuring the effectiveness of a player.

I'd like to hear counter-arguments. Let's try to make some progress without always coming back to Subban, Weber, Eller, Shaw, Bergevin, Pfeiffer, Babcock, Poile, etc. Maybe possession metrics are the best we have, but let's go through it carefully without using the debate as a weapon in another war. I'm aiming for a real discussion.

How should Johnny Boy answer his coach's I-see challenge?

Click to read more...

 

[Lshap]

Pretty good premise.

I'm pretty sure advanced stats are more accurate for teams than individuals. Even then, a team's stats reflect a coaching strategy more than it does an absolute measure of the players, but at least with team-stats we're seeing how the whole product -- players and coaches -- performs together on-ice.

Individual numbers give us a glimpse of a player's style, but are subjected to too many variables. A player may have the ability to carry or shoot the puck, but decides it's smarter to pass it to his even-faster teammate, or doesn't take a shot in favour of setting up an even better chance. Or his coach preaches a particular system that doesn't key on that player's individual skills. In all those cases, that player's advanced stats are hurt not by his lack of ability, but by his decision-making or coaching. He's making unselfish decisions or following the coach's system, but this shows up as a deficit on his stats columns.

And if a player is the recipient of passes, takes low-danger shots, is the point-man on a PP -- all those factors will artificially bump up his advanced numbers.

Not to dismiss the value of this data. It shows what's actually happening on the ice. However it doesn't show why it's happening, and it certainly isn't a direct correlation to a player's talent.

 

[Price is Wright]

How should Johnny Boy answer his coach's I-see challenge?

Johnny isn't running the team.

The bigger problem is when a coach rejects his own available tools to understand why his team is losing because he goes with his old school gut.

 

[Lshap]

Johnny isn't running the team.

The bigger problem is when a coach rejects his own available tools to understand why his team is losing because he goes with his old school gut.

Rejecting any tool is dumb. But each tool has its limits. Analytics give us the perception of understanding because they open up new levels of the game. But they're not the game itself.

I feel like I'm talking in circular eastern philosophy-talk -- "See the Corsi as a life lived, but not the meaning of why it was lived".

 

[Sterling Archer]

Rejecting any tool is dumb. But each tool has its limits. Analytics give us the perception of understanding because they open up new levels of the game. But they're not the game itself.

I feel like I'm talking in circular eastern philosophy-talk -- "See the Corsi as a life lived, but not the meaning of why it was lived".

This and more.

There's no one catch all tool to completely analyse the game of hockey. Corsi and other advanced stats give a deeper look but the interpretation and the science behind them are still in their infancy. Hockey is not baseball where the variables are much easier to take into consideration. Hockey is by far a more dynamic sport and as such, analytics have to account for much more before they can give a truer picture of what's going on with all the different perspectives of the game.

It's a cool new tool but like most things that are new, it's untested and will go through several iterations before it can be consistently used as a larger part of hockey analysis IMO.

 

[Chet-Nick]

No. Analytics is more trivial. Maybe if two players have equal talent. That could be used as a tiebreaker.

A fast moving, bouncing puck is unpredictable at times. All these extra stats don't account for that. Nor do they reflect shot blocking and desire.

If a guy has great possession numbers but is a floater. Not a team oriented person. What good is he?

 

[MXD]

Possession stats are the best tool to evaluate shot differentials. They don't evaluate possession. The same way goals total aren't a very good metric to evaluate shot power.

 

[Apoplectic Habs Fan]

Im not much into the analytics personally. I guess im still old school in a lot of ways.

The biggest thing is the eye test must be a big factor as well if you use them. Statistics on their own can be used to prove almost anything, right or wrong, if you play around with them enough.

Honestly looking at the numbers consistently takes enjoyment out of watching the game. I rather react emotionally as a fan lol. Or irrationally...

 

[LyricalLyricist]

I'm a numbers guy at heart and I think the way people have adopted advanced stats at the expense of everything has been brutal.

Quite frankly +- is a guide and so are 'advanced' stats. No one stat can say everything about someone. This is a fact of life.

Advanced stats are meant to give us MORE tools to evaluate someone, not erase the old ones.

+- relative to teammates is still important because it shows how you either aid or deter the team from winning. However, advanced stats showing your usage also plays a factor.

With hockey advanced stats aren't as clear cut as other sports. It's essential people don't act like one thing says everything or that some things don't matter. Everything matters and watching the game does as well.

 

[Michelangelo]

No. Analytics is more trivial. Maybe if two players have equal talent. That could be used as a tiebreaker.

A fast moving, bouncing puck is unpredictable at times. All these extra stats don't account for that. Nor do they reflect shot blocking and desire.

If a guy has great possession numbers but is a floater. Not a team oriented person. What good is he?

But how do you measure talent? How do you know how much talent a player has? Are we talking about executional talent, or potential? Are we talking peak talent or on an average basis?

 

[Kriss E]

Not every stat means the same and most are useless without context.
There isn't one best stat to evaluate a player. I think everything needs to be considered, as well as the eye test.
If all you have are 2-3 stats, then really, you shouldn't be making any evaluation or drawing conclusions.

 

[scrubadam]

Possession stats tell you what happened that game. Not what will happen in the future. I would say the same for other advanced stats. IMHO you can't plug someone's possession or advanced stats into a different time and environment and it will play out then same.

Use the less stats to explain an analyze but not as some predictor of the future. Player A isn't guaranteed to have the same possession stats playing on a different team under a diff coach with different teammates on the ice and a different goalie.

 

[Michelangelo]

Not every stat means the same and most are useless without context.
There isn't one best stat to evaluate a player. I think everything needs to be considered, as well as the eye test.
If all you have are 2-3 stats, then really, you shouldn't be making any evaluation or drawing conclusions.

Exactly.

Even the eye-test has limitations, just as every single stat does.

But they can and should be used in a complementary fashion, to provide a complete and thorough analysis, not on a stand-alone basis.

 

[Michelangelo]

Possession stats tell you what happened that game. Not what will happen in the future. I would say the same for other advanced stats. IMHO you can't plug someone's possession or advanced stats into a different time and environment and it will play out then same.

Use the less stats to explain an analyze but not as some predictor of the future. Player A isn't guaranteed to have the same possession stats playing on a different team under a diff coach with different teammates on the ice and a different goalie.

It's about distinguishing signal from the noise. No model will accurately predict 100% of future outcomes due to the inherent nature of random variation.

But if we can limit or reduce the error term to a minimum, it would go a long way. This of course, without overfitting to the data, which would then capture the noise in addition to the signal.

 

[TT1]

Yes they are but there's a ton of adv. stats.. Corsi is important but there are plenty of other ways that players can be effective, just take this example where i compare JVR and Gallagher (copy pasting a post from another thread).

I compare their fenwicks, penalty taken/drawn, rebound/60, points per 60 etc. If you wanna properly evaluate a player based on advanced stats you need to do alot of research.

"Gallagher (http://stats.hockeyanalysis.com/showplayer.php?pid=1689)

vs

JVR (http://stats.hockeyanalysis.com/showplayer.php?pid=1182)

Gally FF% (5v5 ES Zone Start Adjusted)

15-16: 55% (TMFF% 49.99% - this is team FF%)
14-15: 54.2% (TMFF% 49.3%)
13-14: 52.4% (TMFF% 47.5%)
12-13: 59.0% (TMFF% 54.8%)

As you can see Gally has elite Fenwick numbers (especially relative to his team), on top of that it's well known that he's one of the best puck possession players in the league and he has one the highest P/60 among RW'ers in the NHL. Oh yea and he's signed for 3.750M for 5+ years.

Bonus stats, hes the best net front presence forward in the league aswell: http://www.tsn.ca/the-value-of-a-forward-with-net-front-presence-1.526616 (better than Parise, Simmonds, Hornqvist etc.)

_________

JVR FF% (5v5 ES Zone Start Adjusted)

15-16: 51.8% (TMFF% 50.1% - this is team FF%)
14-15: 43.9% (TMFF% 47.5%)
13-14: 43.6% (TMFF% 39.8%)
12-13: 45.1% (TMFF% 44.2%)

As you can see JVR's Fenwick numbers can't touch Gally's, and if you notice his FA60 you'll see that his shot suppression is terrible.

_________

Bonus stat in order to quantify Gallagher's effectiveness:

Penalty Taken/Penalty Drawn per 60:
http://www.behindthenet.ca/nhl_stat...3+5+4+6+7+8+13+14+29+30+32+33+34+45+46+63+67# (Gally)

http://www.behindthenet.ca/nhl_stat...+3+5+4+6+7+8+13+14+29+30+32+33+34+45+46+63+67 (JVR)

Gallagher:
15-16: Pens Taken: 0.3 // Pens Drawn: 1.4
14-15: Pens Taken: 0.5 // Pens Drawn: 0.7
13-14: Pens Taken: 0.9 // Pens Drawn: 1.2
12-13: Pens Taken: 1.1 // Pens Drawn: 1.6

JVR:
15-16: Pens Taken: 0.1 // Pens Drawn: 0.4
14-15: Pens Taken: 0.7 // Pens Drawn: 0.6 (negative)
13-14: Pens Taken: 0.9 // Pens Drawn: 0.7 (negative)
12-13: Pens Taken: 0.7 // Pens Drawn: 0.4 (negative)

As you can see Gallagher has always drawn alot of penalty's (avg 1.23) throughout his career (due to the style of game he plays) and he's been getting more disciplined year after year (his Pens taken have dropped after each year).

Compare him to JVR, he draws alot less penalties because he plays soft (avg 0.53) and on top of that he takes more penalties than he draws which ends up hurting his team.

____________________

** Gallagher P/60:

http://www.behindthenet.ca/nhl_stat...+RW&f7=40-&f8=1&c=0+1+3+5+4+6+7+8+17+18+19+20

15-16: 10th in the NHL (4th among RW'ers behind Jagr, Kane, Wheeler.. hes ahead of Tarasenko)

Gallagher is simply one of the most effective/efficient players in the NHL. He's a workhorse."

 

[Lafleurs Guy]

I'm a numbers guy at heart and I think the way people have adopted advanced stats at the expense of everything has been brutal.

I don't think people have adopted advanced stats at the expense of everything else. I think it comes across that way because they can be (and are) used in an argument when comparing players.

The nice thing about these numbers is that they are objective fact. It's not like one guy who says "he plays well" and the other says "he really didn't." At the very least we have numbers we can appeal to.

And people seem to think advanced stats= corsi. There's more to it than that. And a lot of it isn't all that complicated. Points/60 for example is an advanced stat...

Possession stats are the best tool to evaluate shot differentials. They don't evaluate possession. The same way goals total aren't a very good metric to evaluate shot power.

Actually, they mirror zone possession almost perfectly and can be used as a proxy for possession.

 

[V13]

Pretty good premise.

I'm pretty sure advanced stats are more accurate for teams than individuals. Even then, a team's stats reflect a coaching strategy more than it does an absolute measure of the players, but at least with team-stats we're seeing how the whole product -- players and coaches -- performs together on-ice.

Individual numbers give us a glimpse of a player's style, but are subjected to too many variables. A player may have the ability to carry or shoot the puck, but decides it's smarter to pass it to his even-faster teammate, or doesn't take a shot in favour of setting up an even better chance. Or his coach preaches a particular system that doesn't key on that player's individual skills. In all those cases, that player's advanced stats are hurt not by his lack of ability, but by his decision-making or coaching. He's making unselfish decisions or following the coach's system, but this shows up as a deficit on his stats columns.

And if a player is the recipient of passes, takes low-danger shots, is the point-man on a PP -- all those factors will artificially bump up his advanced numbers.

Not to dismiss the value of this data. It shows what's actually happening on the ice. However it doesn't show why it's happening, and it certainly isn't a direct correlation to a player's talent.

Great post and this is pretty much how i see it as well. I value possession stats/Corsi more as a team stats than an individual stat for the same reasons you mentionned. That doesn't mean that possession for an individual player is not important but it must be placed within a context of in games situations and those normally vary a lot because they are affected by many different dynamics and factors , some of whom cannot be quantified.

Don't have much else to say other than that because you post was bang on.

 

[Hoople]

Possession stats are the best tool to evaluate shot differentials. They don't evaluate possession. The same way goals total aren't a very good metric to evaluate shot power.

This exactly.

The word possession was co-opted by the analytics bloggers who used the Internet and social media to force their beliefs and agenda on hockey fans.

Corsi and Fenwick are shot differentials. A +/- for shots. Period.

Possession should be measured as the amount of time that Team A or Team B has control of the puck. That is a more reliable measure of who is controlling play.

The NFL measures time of possession. The NHL is capable of doing the same.

Proxy, by definition, is a substitute. It is not absolutely the same as. It is an estimation, a guess.

 

[Estimated_Prophet]

It really is a pretty useless way to evaluate individual performance. It says more about the team concept and on a smaller spectrum, line performance.

Good teams will use Corsi and Fenwick as a tool to draw attention to a possible issue but they still have to watch video to ascertain cause and effect.

These tools are only useful in the hands of experts. Your average fan is poorly misguided in forming concrete conclusions from these metrics as is all too often on display on these forums.

 

[Hoople]

It really is a pretty useless way to evaluate individual performance. It says more about the team concept and on a smaller spectrum, line performance.

Good teams will use Corsi and Fenwick as a tool to draw attention to a possible issue but they still have to watch video to ascertain cause and effect.

These tools are only useful in the hands of experts. Your average fan is poorly misguided in forming concrete conclusions from these metrics as is all too often on display on these forums.

When you have last season's Canadiens with a higher CF% than Washington, Chicago, Rangers and Florida, then the limitations of using Corsi as an accurate evaluator for teams has to questioned.

The answers and reasons are there when context is applied. And with that, Corsi is no more valuable a tool than any other stat when relied upon singularly.

And yes, Corsi is presented as a singular tool for evaluation all the time here.

 

[habalifeok]

I think video is very important to analyze players and how to correct mistakes. Also to show players successful plays on the power play,etc.
Management can use the back room to try and correlate stats with video. Players want the video.
The future will be chips inserted in uniforms to track position speed etc.and this info will be combined with video to show players how to improve positioning , decision making etc.

 

[Price is Wright]

I think video is very important to analyze players and how to correct mistakes. Also to show players successful plays on the power play,etc.
Management can use the back room to try and correlate stats with video. Players want the video.

Guess what happens when you watch a lot of video on a player and jot down their statistics on every shift?

You get analytics.

 

[DangerDave]

Useless without context and should not be the be all, end all. Taking the Gallagher and JVR example posted above would make you think JVR is some plug while Gallagher is better than guys like Kucherov, Kessel etc... You can't rely solely on this to evaluate a player. According to these stats, McDonough and Weber are merely average.

Your team, quality of competition, zone starts, system and style/ assignment all come into play here. One thing that is overlooked is what exactly you do with that possession. Weak shots from the boards don't amount to anything by they sure look good statstically. That's why quality scoring chances % is important to look at. Most important though is the eye test.

 

[Price is Wright]

Useless without context and should not be the be all, end all. Taking the Gallagher and JVR example posted above would make you think JVR is some plug while Gallagher is better than guys like Kucherov, Kessel etc... You can't rely solely on this to evaluate a player. According to these stats, McDonough and Weber are merely average.

Your team, quality of competition, zone starts, system and style/ assignment all come into play here. One thing that is overlooked is what exactly you do with that possession. Weak shots from the boards don't amount to anything by they sure look good statstically. That's why quality scoring chances % is important to look at. Most important though is the eye test.

Nobody is arguing to remove the eye test except those building a strawman to try to downplay the importance of analytics.

What so many seem to miss is that the only way to compile advanced statistics is through the eye test.

You just don't like it when someone on the Internet makes a point with numbers instead of what you use: your gut.

 

[Lafleurs Guy]

Useless without context and should not be the be all, end all. Taking the Gallagher and JVR example posted above would make you think JVR is some plug while Gallagher is better than guys like Kucherov, Kessel etc... You can't rely solely on this to evaluate a player.

I'm not sure how you come up with that evaluation. They aren't that far off from each other. As for context, we do have the ability to see how players play with/without other guys and look at usage... it's not like stats don't provide us with that ability.

You're looking at one stat and thinking that this is the sum total of analytics... it's not.