All Purpose Analytics and Extended Stats Discussion

[QuadrupleDeke]

Two things correlated to the same thing does not necessarily mean that those two things are related. Celery stalks and Jelly are both positively correlated with Peanut Butter as shopping habits go. That does not mean that people buy Celery when they buy Jelly as well. Just one of the reasons that correlation =\= causation.

If two things are correlated with something, then they are correlated with each other. One directly follows from the other.

If you usually buy jelly when you buy peanut butter and you usually buy celery stalks when you buy peanut butter, then you usually buy jelly when you buy celery stalks.

 

[ChibiPooky]

Advanced stats movement is making its way to beer leagues.

http://thehockeymovement.com/advanced-stats-for-beer-league-hockey/

 

[Hivemind]

Plenty of stats, such as faceoffs, have a correlation with winning. However, faceoffs have a weaker correlation with winning than both goal and shot attempt-based metrics (and sv%, PP%, PK%, etc), and have a weaker year-to-year correlation than the shot attempt-based metrics.
http://www.pensionplanpuppets.com/2...n-corsi-fenwick-PDO-hits-fights-blocked-shots

 

[HunterSThompson]

If two things are correlated with something, then they are correlated with each other. One directly follows from the other.

If you usually buy jelly when you buy peanut butter and you usually buy celery stalks when you buy peanut butter, then you usually buy jelly when you buy celery stalks.

I am looking at this in opposite terms and claiming a lack of relation/causation. I am also not talking specifically, but generally. I tried to expand further and obviously miserably failed. Oh and what you said is not necessarily true.

 

[g00n]

Here is the post I was talking about regarding face offs vs other stats including "possession"...

http://hfboards.mandatory.com/showpost.php?p=87969889&postcount=152

 

[NobodyBeatsTheWiz]

EDIT: According to that chart, it would take Ovechkin almost 4 hours to get to his top speed. Crosby almost 5 hours. Perhaps they meant km/hr/sec and not km/hr^2 ?

Typo in the chart that has since been fixed. Acceleration was measured in m/s^2.

 

[QuadrupleDeke]

Typo in the chart that has since been fixed. Acceleration was measured in m/s^2.

Makes way more sense. Thanks.

Now back to the real issue. Does Ovechkin, one of the heaviest players in the NHL, really accelerate 30% faster than most NHL stars? I'm dubious.

 

[Raikkonen]

Makes way more sense. Thanks.

Now back to the real issue. Does Ovechkin, one of the heaviest players in the NHL, really accelerate 30% faster than most NHL stars? I'm dubious.

I call small sample size (BS)! How could it be so one-sided?!

 

[Dream Big]

Makes way more sense. Thanks.

Now back to the real issue. Does Ovechkin, one of the heaviest players in the NHL, really accelerate 30% faster than most NHL stars? I'm dubious.

I'm going with yes.

Acceleration to red carpet with A list star is fastest and done all the while dragging two Ruskies on his coat tails.

http://www.csnwashington.com/hockey...hotos-ovechkin-hangs-brad-pitt-movie-premiere

 

[QuadrupleDeke]

The major problem with Corsi getting so popular is that the predictive value of it goes away as you start to optimize for it. Hockey is not a game about attempting shots; It's a game about scoring goals. It just so happened that over long sample sizes, shot quality was relatively uniform between teams, which made attempted shots a decent surrogate for possession, which itself was a decent surrogate for goal differential and wins. However, once you start trying to maximize one of those stats, its predictive value goes away. Want to maximize your Corsi? Take a bunch of terrible shots. Your Corsi will be great, but it will no longer correlate with success.

I wonder if the NHL is approaching the point where Corsi/possession based thinking is losing its usefulness.

 

[ChibiPooky]

The major problem with Corsi getting so popular is that the predictive value of it goes away as you start to optimize for it. Hockey is not a game about attempting shots; It's a game about scoring goals. It just so happened that over long sample sizes, shot quality was relatively uniform between teams, which made attempted shots a decent surrogate for possession, which itself was a decent surrogate for goal differential and wins. However, once you start trying to maximize one of those stats, its predictive value goes away. Want to maximize your Corsi? Take a bunch of terrible shots. Your Corsi will be great, but it will no longer correlate with success.

I wonder if the NHL is approaching the point where Corsi/possession based thinking is losing its usefulness.

As long as the team with the most goals wins, and as long as it's easier to score on a better shot, then Corsi and other shot attempt metrics are still valid. Remember that shot attempts are only "equal" over extremely large samples. On any individual play, you still want the best shot attempt you can get, because it gives you a better chance to score.

I wonder if teams start preaching a random shot attempt over a random puck to the corner, though, if that starts to change the dynamic. I'm not sure that it's preferable to create chaos rather than sustaining the cycle, but to me that's a pressure point that could shift things.

 

[Hivemind]

The major problem with Corsi getting so popular is that the predictive value of it goes away as you start to optimize for it. Hockey is not a game about attempting shots; It's a game about scoring goals. It just so happened that over long sample sizes, shot quality was relatively uniform between teams, which made attempted shots a decent surrogate for possession, which itself was a decent surrogate for goal differential and wins. However, once you start trying to maximize one of those stats, its predictive value goes away. Want to maximize your Corsi? Take a bunch of terrible shots. Your Corsi will be great, but it will no longer correlate with success.

I wonder if the NHL is approaching the point where Corsi/possession based thinking is losing its usefulness.

I don't think coaches are advocating taking suboptimal shots to improve Corsi numbers. I do think managers are using these stats to optimize the construction of their roster, and coaches are using these stats to optimize the utility of their deployments.

 

[QuadrupleDeke]

I don't think coaches are advocating taking suboptimal shots to improve Corsi numbers. I do think managers are using these stats to optimize the construction of their roster, and coaches are using these stats to optimize the utility of their deployments.

I can tell you that if I was a player, and Corsi had ever been used as justification to increase or decrease my ice time, or increase or decrease my contract offer, or anything like that, I'd be looking for any way possible to increase my Corsi. Try that pass or take the shot? Take the shot.

There is literally no scenario where coaches or players even knowing what Corsi is is good for the success of the team.

 

[QuadrupleDeke]

As long as the team with the most goals wins, and as long as it's easier to score on a better shot, then Corsi and other shot attempt metrics are still valid. Remember that shot attempts are only "equal" over extremely large samples. On any individual play, you still want the best shot attempt you can get, because it gives you a better chance to score.

I wonder if teams start preaching a random shot attempt over a random puck to the corner, though, if that starts to change the dynamic. I'm not sure that it's preferable to create chaos rather than sustaining the cycle, but to me that's a pressure point that could shift things.

Or always shooting on 2-on-1s, or always taking the easy shot on power plays, or always trying to shoot it towards the goalie on dump-ins. There are many ways that focusing on Corsi can reduce the effectiveness of hockey team, and thus reduce the predictive value of the stat itself.

 

[ChibiPooky]

I can tell you that if I was a player, and Corsi had ever been used as justification to increase or decrease my ice time, or increase or decrease my contract offer, or anything like that, I'd be looking for any way possible to increase my Corsi. Try that pass or take the shot? Take the shot.

There is literally no scenario where coaches or players even knowing what Corsi is is good for the success of the team.

Or always shooting on 2-on-1s, or always taking the easy shot on power plays, or always trying to shoot it towards the goalie on dump-ins. There are many ways that focusing on Corsi can reduce the effectiveness of hockey team, and thus reduce the predictive value of the stat itself.

That would be true if Corsi were the only stat used in negotiations, and/or if simply choosing to shoot all the time had no impact on your other stats. For example, in the 2-on-1 scenario, you'd be giving up a pretty strong chance of an assist in exchange for that shot attempt.

 

[Hivemind]

I can tell you that if I was a player, and Corsi had ever been used as justification to increase or decrease my ice time, or increase or decrease my contract offer, or anything like that, I'd be looking for any way possible to increase my Corsi. Try that pass or take the shot? Take the shot.

There is literally no scenario where coaches or players even knowing what Corsi is is good for the success of the team.

Joel Quennville knows what these stats are, and uses them to help guide his line construction and deployment. Pretty clearly a scenario where knowing Corsi is good for the success of the team, considering Q has led the Blackhawks to two cups.

 

[QuadrupleDeke]

That would be true if Corsi were the only stat used in negotiations, and/or if simply choosing to shoot all the time had no impact on your other stats. For example, in the 2-on-1 scenario, you'd be giving up a pretty strong chance of an assist in exchange for that shot attempt.

Well, previously you had:

chance at a goal VS chance at an assist

Now you have

a corsi event plus a chance at a goal VS chance at a corsi event plus chance at an assist

The trade-off is skewed by the player thinking about Corsi. Regardless of what the chances of the goal and assist are, there are situations where the player will choose wrong because of the presence of the "definite corsi" versus "chance at a corsi". This lowers his team's chance of winning.

 

[Hivemind]

The same argument could be applied to publicly tracking any stat, 4xdeke.

 

[QuadrupleDeke]

Joel Quennville knows what these stats are, and uses them to help guide his line construction and deployment. Pretty clearly a scenario where knowing Corsi is good for the success of the team, considering Q has led the Blackhawks to two cups.

True, I misspoke. In team construction and line construction, Corsi is useful. In actually deciding how to act in different situations (which is the part of "coaching" I was referring to there), focusing on Corsi makes the stat less predictive.

 

[QuadrupleDeke]

The same argument could be applied to publicly tracking any stat, 4xdeke.

This is probably true. My original post was in the Tank thread, after a bunch of posts talking about how the Caps have great Corsi numbers, so the wins will come (someone moved my post to this thread, removing the context). At some point, this will no longer be true. If the league focuses on improving their Corsi numbers enough, wins will no longer be inevitable for a team with a great Corsi%.

 

[ChibiPooky]

The two on one scenario is interesting, though, because it's theoretically a quality shot. In practice you expect the goalie to be set up for your shot but it's still considered a scoring chance. Likewise, even if you didn't see net, why wouldn't you just try the shot pass off the goalie instead of a regular pass?

Regardless, though, I don't think these specific scenarios comprise enough of the event base to have a significant impact on the statistic. These things aren't tracked, so we don't really have access to the "true answer."

 

[ChibiPooky]

This is probably true. My original post was in the Tank thread, after a bunch of posts talking about how the Caps have great Corsi numbers, so the wins will come (someone moved my post to this thread, removing the context). At some point, this will no longer be true. If the league focuses on improving their Corsi numbers enough, wins will no longer be inevitable for a team with a great Corsi%.

That was me. I thought the discussion had progressed beyond the scope of the tank thread, so I moved it (all the way up to the beginning).

It will no longer be true IF the league starts coaching to shot attempts as the primary driver. But I don't see any evidence that they have, nor any evidence that they will. I think this is a red herring.

 

[Hivemind]

Unless one player or the other completely flubs on the 2-on-1, a Corsi For event would be generated regardless of whether or not the puck carrier passes the puck. Corsi doesn't rely on an individual's shots, but the team's while he's on the ice. Passing the puck there still generates a Corsi For, as his teammate would get a shot attempt.

 

[QuadrupleDeke]

Unless one player or the other completely flubs on the 2-on-1, a Corsi For event would be generated regardless of whether or not the puck carrier passes the puck. Corsi doesn't rely on an individual's shots, but the team's while he's on the ice. Passing the puck there still generates a Corsi For, as his teammate would get a shot attempt.

I know what Corsi is. If the pass on a 2-on-1 was always successful, then no one would ever take the shot. It's nowhere near 100% that the teammate gets a shot off.

 

[g00n]

I know what Corsi is. If the pass on a 2-on-1 was always successful, then no one would ever take the shot. It's nowhere near 100% that the teammate gets a shot off.

The point you're making is clear to me and I think some of us have even discussed it in this thread. I've talked about it in recent GDT when posters complain about high possession but no results. Better teams do good things that result in shots and goals. You don't get better or get goals just by shooting.

The stat is only valuable if the players are largely ignorant of it. I believe they do know about it, but their understanding is probably limited. Most will say they've heard of it but don't get into it much. Some may be lying, but I'm guessing what most players are going to come away with is "shooting is good" and they will do what they can to generate shots. Professional athletes in all sports are aware of stats to some degree, even if they don't admit it. Their careers often depend on it, but you can't play effectively when micromanaging stats. So you simplify.

In the 2 on 1 scenario you describe it may even be an unconscious decision to take the shot (even if the better play is to attempt the pass due to defender and goalie position) because of that simplified, positive association.

You're essentially describing what we could call Shrodinger's Shot. The observation of the stat is likely affecting the stat, creating a paradox. Or something.