Advanced player score based on CORSI

[pdd]

Earlier today I had the idea for developing a "total player score" based on CORSI metrics. It had twofold inspiration; the thread I was in had evolved into a discussion about "Derp, why does DREW MILLER have the highest QoC number?" and a legitimate desire to prove how valid a metric CORSI is.

As a raw metric on its own, both CORSI and RelCORSI are not perfect for ranking player ability. This is because many shots come due to bad line changes, or one player makes a mistake (or good play) and all of his teammates are punished in their CORSI rating.

Excellent examples of this are Cory Emmerton and Justin Abdelkader. Abdelkader had the best CORSI ratings of any Red Wing forward in the 2013 playoffs. He was solid and played like an average second liner (probably the best hockey of his life) but he wasn't better than Zetterberg, Datsyuk, Nyquist, and arguably Franzen, Cleary, or Brunner.

Emmerton is very close to Abdelkader in ability and typical performance, and is the Wings' second most-utilized PKer after Miller (and is very good at it). He's much better than Abdelkader defensively, and if given a chance could probably make a solid third-line center somewhere (he might be out of Detroit this summer, so we'll see).

The stats I was thinking were important to include in a comprehensive metric were as follows:

CORSI, RelCORSI, RelCORSI QoC, RelCORSi QoT, QoC, QoT, and PDO. I just am not sure how to combine them (other than the obvious idea to "normalize" for PDO) and/or whether I should remove/add metrics.

Any suggestions from the experts?

Click to read more...

 

[Micklebot]

Earlier today I had the idea for developing a "total player score" based on CORSI metrics. It had twofold inspiration; the thread I was in had evolved into a discussion about "Derp, why does DREW MILLER have the highest QoC number?" and a legitimate desire to prove how valid a metric CORSI is.

As a raw metric on its own, both CORSI and RelCORSI are not perfect for ranking player ability. This is because many shots come due to bad line changes, or one player makes a mistake (or good play) and all of his teammates are punished in their CORSI rating.

Excellent examples of this are Cory Emmerton and Justin Abdelkader. Abdelkader had the best CORSI ratings of any Red Wing forward in the 2013 playoffs. He was solid and played like an average second liner (probably the best hockey of his life) but he wasn't better than Zetterberg, Datsyuk, Nyquist, and arguably Franzen, Cleary, or Brunner.

Emmerton is very close to Abdelkader in ability and typical performance, and is the Wings' second most-utilized PKer after Miller (and is very good at it). He's much better than Abdelkader defensively, and if given a chance could probably make a solid third-line center somewhere (he might be out of Detroit this summer, so we'll see).

The stats I was thinking were important to include in a comprehensive metric were as follows:

CORSI, RelCORSI, RelCORSI QoC, RelCORSi QoT, QoC, QoT, and PDO. I just am not sure how to combine them (other than the obvious idea to "normalize" for PDO) and/or whether I should remove/add metrics.

Any suggestions from the experts?

You could start by looking at how others have done it;


Corsi HART is the is the one developed by the Hockey analysis guy;
http://stats.hockeyanalysis.com/rat...rs&minutes=100&disp=1&sort=HARTp&sortdir=DESC

Here is his description of how the stat is generated:

http://hockeyanalysis.com/2013/04/11/how-are-hockeyanalysis-com-ratings-haro-hard-hart-calculated/

 

[pdd]

You could start by looking at how others have done it;


Corsi HART is the is the one developed by the Hockey analysis guy;
http://stats.hockeyanalysis.com/rat...rs&minutes=100&disp=1&sort=HARTp&sortdir=DESC

Here is his description of how the stat is generated:

http://hockeyanalysis.com/2013/04/11/how-are-hockeyanalysis-com-ratings-haro-hard-hart-calculated/

Looking both at the numbers generated and at the way it is calculated, I have some definite questions about that.

Such as... why would the expected offense be TmGF20*LAGF20 and not TmGF20/LAGF20? That seems like a major flaw in the rating system. Obviously the same system when used defensively results in the same flaw.

For example, if the TmGF20 for Detroit is exactly 1.05, and the league average is 1.10, then the calculation projects expected offense of 1.155/20. Let's say Anaheim has a TmGF20 of 1.10, projecting to 1.00 (obviously). The next step in the calculation, which is 100*(PlayerGF20-ExpGF20)/ExpGF20, results in the following scores for two players with GF/20 of 1.50:

Detroit: 29.87
Anaheim: 33.33

So the player who plays for the better offensive team but doesn't individually have any more offensive success is scored as the better offensive player by a significant degree.

When any metric says that Dan Cleary is better offensively than Nicklas Lidstrom, Pavel Datsyuk, and Henrik Zetterberg... some serious questions need to be raised. Or that Justin Williams is the second-best offensive player with 162+ minutes.

It also ranks Scott Gomez top-40 offensively for the 2011-12 season, six spots ahead of Jason Spezza. Offensively.

So yeah, I can respect the effort... but it seems kind of flawed.

 

[Micklebot]

Looking both at the numbers generated and at the way it is calculated, I have some definite questions about that.

Such as... why would the expected offense be TmGF20*LAGF20 and not TmGF20/LAGF20? That seems like a major flaw in the rating system. Obviously the same system when used defensively results in the same flaw.

For example, if the TmGF20 for Detroit is exactly 1.05, and the league average is 1.10, then the calculation projects expected offense of 1.155/20. Let's say Anaheim has a TmGF20 of 1.10, projecting to 1.00 (obviously). The next step in the calculation, which is 100*(PlayerGF20-ExpGF20)/ExpGF20, results in the following scores for two players with GF/20 of 1.50:

Detroit: 29.87
Anaheim: 33.33

So the player who plays for the better offensive team but doesn't individually have any more offensive success is scored as the better offensive player by a significant degree.

The expected Goals for per 20, or Corsi for or what ever base metric you're using is for a replacement player. So because you'd expect a player to score more, have a higher corsi or whatever on a better team (in regards to the relevant metric) they end up with a higher expected GF/20. You then compare their actual GF/20 to the expected value, and in the case of the Anaheim player he would meet the expected GF/20, but the Detroit player would exceed it and end up with a higher HARO: HARO(1st iteration) = 100*(GF20-ExpGF20) / ExpGF20)

When any metric says that Dan Cleary is better offensively than Nicklas Lidstrom, Pavel Datsyuk, and Henrik Zetterberg... some serious questions need to be raised. Or that Justin Williams is the second-best offensive player with 162+ minutes.

It also ranks Scott Gomez top-40 offensively for the 2011-12 season, six spots ahead of Jason Spezza. Offensively.

So yeah, I can respect the effort... but it seems kind of flawed.

One thing to point out here is that it is measuring there effectiveness at generating corsi events, morso than how good they are offensively. I'm not a fan of all encompassing metrics as they often produce zany results, and the reason imo is that they aren't actually measuring what you or I think they are or what is intended to be measured. In this case, it seems to be attempting to measure the impact an individual has on his on ice CF%.

 

[HockeyAnalysis]

It also ranks Scott Gomez top-40 offensively for the 2011-12 season, six spots ahead of Jason Spezza. Offensively.

So yeah, I can respect the effort... but it seems kind of flawed.

Part of that is because it isn't measuring offense, it is measuring corsi, or shot attempts. Gomez over the years has consistently had pretty good corsi numbers. The thing is, he also has consistently had very poor on-ice shooting percentage numbers. Over the past 6 seasons of 277 forwards with >3000 5v5 minutes, Gomez has the 15th lowest on-ice shooting percentage (see http://stats.hockeyanalysis.com/rat...00&teamid=0&type=goals&sort=ShPct&sortdir=ASC). He also has the 8th best corsi for rate (shot attempts by his team per 20 minutes of ice time) over the past 6 seasons (see http://stats.hockeyanalysis.com/rat...000&teamid=0&type=corsi&sort=F20&sortdir=DESC).

In English, when Gomez is on the ice his team takes a ton of shots but is terrible at converting them into goals. This is why Gomez generally does well in Corsi metrics but in reality Gomez is a prime example of where corsi fails a player evaluation tool.

I personally prefer a goal based analysis but there are sample size issues with such an analysis so to get something reasonably reliable it is best to use multiple seasons worth of data. Over the past 6 seasons the forwards with the best HARO rating are (from http://stats.hockeyanalysis.com/rat...teamid=0&type=goals&sort=HAROp&sortdir=DESC):

1. Crosby
2. Semin
3. Malkin
4. H. Sedin
5. Toews
6. Horton
7. Spezza
8. D. Sedin
9. J.P. Dumont
10. Datsyuk
11. Ovechkin
12. Downie
13. Lupul
14. Kunitz
15. Backstrom
15. Stamkos
17. Voracek
18. St. Louis
19. Ryan
20. Vanek

There are a few surprises in that list, but it mostly passes the smell test. The reality is you will never be able to develop a perfect all-encompassing player evaluation statistic and I'll never claim HARO, HARD and HART to be that but I do think it gives a decent overview of whats happening.

 

[PoutineSp00nZ]

corsi sucks

 

[tombombadil]

numbers aside, I actually feel that Abdelkader was amongst the best Wings at generating pucks going in the right direction. To my eyes he is miles better than Emmerton. The problem is that Corsi doesn't include the ability to create genuine chances, nor the ability to bury them.,,, two things that Abbie is absolutely no good at. But as far as pressuring people into losing the puck, creating loose pucks with physicality, and driving overall momentum at the net.... he was at least top 3 for the Wings in my eyes.

 

[Rals]

Alex Semin always coming up big in measurements, but then ripped on by media.

 

[Cunneen]

numbers aside, I actually feel that Abdelkader was amongst the best Wings at generating pucks going in the right direction. To my eyes he is miles better than Emmerton. The problem is that Corsi doesn't include the ability to create genuine chances, nor the ability to bury them.,,, two things that Abbie is absolutely no good at. But as far as pressuring people into losing the puck, creating loose pucks with physicality, and driving overall momentum at the net.... he was at least top 3 for the Wings in my eyes.

Corsi and Fenwick correlate very well with scoring chances. You can read this article by Eric T http://nhlnumbers.com/2012/6/26/sho...ation-between-scoring-chances-and-shot-totals

or this article by Vic Ferrari http://vhockey.blogspot.com/2009/05/possession-is-everything.html

.

Both articles show that shot differential metrics (i.e. Corsi and Fenwick) prove to correlate very well with Scoring chances. Shot quality tends to wash out over a season's worth of games. At a team level it is nonexistent, http://vhockey.blogspot.com/2009/07/shot-quality-fantasy.html . At the individual level, there is some persistance in shot quality over a season's worth of games, but not that much.

 

[smackdaddy]

Alex Semin always coming up big in measurements, but then ripped on by media.

Do you really expect sports journalists to make realistic commentary based on stats? They live on the gravy train where no real integrity is required, only appearances of such and then only towards each other. Sensationalists, really. But in Semin's case, you can't beat the Russian Factor even if it is grossly overstated.

Rant over

 

[seventieslord]

Excellent examples of this are Cory Emmerton and Justin Abdelkader. Abdelkader had the best CORSI ratings of any Red Wing forward in the 2013 playoffs. He was solid and played like an average second liner (probably the best hockey of his life) but he wasn't better than Zetterberg, Datsyuk, Nyquist, and arguably Franzen, Cleary, or Brunner.

The biggest takeaway from this is, first, not to get too caught up in small samples, and second, to never forget that situational use (zone starts and quality of linemates/competition) affect CORSI.

If Abdelkader actually had the best CORSI on the team over a large sample, and after adjusting for situational factors, and also after accounting for the sustained improvements in shot% or save% of certain players, then I wouldn't hesitate to call him their most effective player. But that wouldn't happen of course. There was a lot at play in him having the best CORSI on the team for 14 games.

And, how to adjust CORSI for these factors? Couldn't tell ya. That CORSI hart looks terrible though. Yeah, the Leafs were 5th in the conference and had 7 of the 8 worst players in the NHL. Right. And their worst defensive forward got 6th in selke voting. And that fool Randy Carlyle played their worst defenseman way more than anyone else. Sure.

 

[ARSix]

It's Corsi, not CORSI. It doesn't stand for anything. Sorry, pet peeve.

 

[seventieslord]

It's Corsi, not CORSI. It doesn't stand for anything. Sorry, pet peeve.

Yep, you're right.

 

[Juxtaposer]

Corsi is a person. http://www.sportingnews.com/nhl/story/2013-04-23/jim-corsi-hockey-definition-nhl-advanced-stats-primer-behind-the-net-fenwick-pdo