Advanced Stats: Corsi, QoC, etc

Les Wynan*

Guest
Are these advanced stats useful in predicting the outcome of a single game/series/season? And how far back do they go?

If someone could link to a study showing the best advanced stats for predicting team success in the season/post-season I'd be very interested.

Corsi Tied (i.e. the percentage of all even strength shots, missed shots and blocked shots attempted while the score is tied in games involving Team X that Team X themselves attempt) has been shown to be a better predictor of future winning percentage than current winning percentage or current goal differential at any point of the season.

Recently, score-adjusted Fenwick has been shown to be an improvement in predictive value over Corsi Tied (and, therefore, winning% and goal differential). It's likely the most accurate measure of team talent in existence. The top ten teams in the league this season by that measure are, in order, Detroit, St. Louis, Pittsburgh, Boston, San Jose, Vancouver, Chicago, Philadelphia, Los Angeles and New Jersey.
 

billybudd

Registered User
Feb 1, 2012
22,049
2,249
They have their place though. If Player A and Player B are tied in points, TOI, play on similar teams, and are equal in all other traditional metrics, but Player A has significantly better Corsi, then yeah that adds something substantial to the argument.

Actually, no. That would be an example of a situation in which raw Corsi tells you absolutely nothing.
 

Egil

Registered User
Mar 6, 2002
8,838
1
Visit site
The issue is no one has been able to find an instance of a team where scoring chances for/against are not proportional to Fenwick for/against. It's not that "all shots are created equal" - that's obviously a ridiculous and completely untrue statement - it's that the notion some teams are able to control shot quality to a greater degree than others has largely been proven to be false. PDO invariably regresses to the mean long term.

Just because some people might misuse advanced stats is no reason to discount them - seems like that was the rationale behind BoredMan's dismissal of the numbers.

I don't think you can do too much with the data made available by the NHL, and to truly bring stats that are useful to predict future performance (especially at the individual level). I do suspect that teams are keeping better tabs on things, and that they are coming up with some useful numbers.
 

NOTENOUGHJTCGOALS

Registered User
Feb 28, 2006
13,542
5,771
Corsi Tied (i.e. the percentage of all even strength shots, missed shots and blocked shots attempted while the score is tied in games involving Team X that Team X themselves attempt) has been shown to be a better predictor of future winning percentage than current winning percentage or current goal differential at any point of the season.

Recently, score-adjusted Fenwick has been shown to be an improvement in predictive value over Corsi Tied (and, therefore, winning% and goal differential). It's likely the most accurate measure of team talent in existence. The top ten teams in the league this season by that measure are, in order, Detroit, St. Louis, Pittsburgh, Boston, San Jose, Vancouver, Chicago, Philadelphia, Los Angeles and New Jersey.

Thanks I'll take a look more later. It would be interesting to see where the four semi-finalists each year finished in fenwick/corsi tied at the end of the regular season.
 

eklunds source

Registered User
Jul 23, 2008
8,323
0
Ed Snider's basement
The problem with these is that they're so new, people don't really grasp how to use them.

I don't claim to be an expert, but I've seen more than a handful of people in this thread explaining a concept wrong, declaring hypothesis as facts, and being out-ad-out 100% wrong about something they're calling a fact.

Discussing advanced stats, their implications, weaknesses and strengths, predictive powers, etc on HFBoards is an exercise in futility.
 

GKJ

Global Moderator
Feb 27, 2002
185,669
37,463
Are these advanced stats useful in predicting the outcome of a single game/series/season? And how far back do they go?

If someone could link to a study showing the best advanced stats for predicting team success in the season/post-season I'd be very interested.

behindthenet.ca (Gabriel Desjardins' site) goes back to 07-08 I think.


He, among others foresaw the "collapse" of the Wild this year, Stars, Avalanche before that. A good place to start.
 

Les Wynan*

Guest
The problem with these is that they're so new, people don't really grasp how to use them.

I don't claim to be an expert, but I've seen more than a handful of people in this thread explaining a concept wrong, declaring hypothesis as facts, and being out-ad-out 100% wrong about something they're calling a fact.

Discussing advanced stats, their implications, weaknesses and strengths, predictive powers, etc on HFBoards is an exercise in futility.

I don't know if this is directed at me but the only "fact" I've alluded to is that team PDO largely tends to regress over a sufficient sample size. Perhaps fact isn't the right word but that's at least what's been observed.
 

GKJ

Global Moderator
Feb 27, 2002
185,669
37,463
The problem with these is that they're so new, people don't really grasp how to use them.

I don't claim to be an expert, but I've seen more than a handful of people in this thread explaining a concept wrong, declaring hypothesis as facts, and being out-ad-out 100% wrong about something they're calling a fact.

Discussing advanced stats, their implications, weaknesses and strengths, predictive powers, etc on HFBoards is an exercise in futility.

That's the other thing, and to fully buy in may mean to compromise your own way of thinking. Not an easy thing to convince someone to do (and when ASA people can't do it, that's when some of them dismiss people as stupid, dumb, etc.)

It doesn't make anyone dumb if they don't get it or immediately buy in. It's just like learning anything else, learning one thing before the next.

Or start with James Mirtle:

http://www.theglobeandmail.com/sports/hockey/the-moneypuck-revolution/article2178766/

http://www.theglobeandmail.com/spor...ed-stats-caught-on-in-the-nhl/article2178777/

http://www.theglobeandmail.com/sports/hockey/hockeys-new-numbers/article2178781/
 

eklunds source

Registered User
Jul 23, 2008
8,323
0
Ed Snider's basement
Addendum: the best stat (and easy to understand) is PDO.

The opposing goaltenders save percentage at 5on5 (aka your team's shooting percentage at 5on5) plus your own goalie's save percentage at 5on5.

For example: the Bruins through the first 10 games were 3-7 and their PDO was 970ish - aka they definitely were playing much better than a 3-7 team but were unlucky.

After their insane 23-3-1 run, their PDO had climbed to 1050... They were (are) clearly a very good team, but definitely not THAT good..

The Leafs had a pretty high PDO for a long time - which is why they were very likely to drop off in their performance. Of course, their recent run of (correct me if I'm wrong) 2-14-2 saw their PDO drop off a cliff - they weren't as good as their record was, but they're also not nearly as bad as that recent string indicates.

Remember when Kopitar was leading the league in scoring in November a while back? High PDO. Kessel leading the league earlier this year? High PDO. The Sedins stinking it up recently? Low PDO. Vermette having a brutal year in Columbus this year? Low PDO. All of these trends have, or will, almost surely reverse at some point.
 

eklunds source

Registered User
Jul 23, 2008
8,323
0
Ed Snider's basement
I don't know if this is directed at me but the only "fact" I've alluded to is that team PDO largely tends to regress over a sufficient sample size. Perhaps fact isn't the right word but that's at least what's been observed.
Wasn't directing it at anyone specifically, but a few posts I read made me facepalm.. I don't think one of them was yours.

People don't realize that brilliant hockey minds have spent countless hours working on studies, analyzing data for predictive value, AND watching the games to see how and what it correlates with - these aren't just wild concepts thrown together by a bunch of mouth breathing yahoos with neckbeards. These are concepts proven by people who have done contract work with NHL teams.
 

Les Wynan*

Guest
The Sharks' recent collapse for which everyone in the organization deserves to be fired or traded is also entirely because of PDO.
 

Keaver

Registered User
Mar 4, 2012
444
0
Philadelphia, PA
Can someone post the advanced stats of the top 5 scorers? I expect Giroux to have highest qual comp and most defensive draw percentage.

I believe this is the post you're referring too.


Malkin no, but Stamkos? Here's a post i made that was ignored a bit the other day:

Obviously Malkin is streets ahead. Let's look at the supposed miles gap between Giroux and Stamkos then shall we? This comparison runs between the start of last season and today and all stats are taken from NHL.com

TOI in ES or PP opportunities:
Stamkos - 2994:30.
Giroux - 2560:25.

Points Scored:
Stamkos - 171.
Giroux - 151.

Minutes Per Point:
Stamkos - 17:30
Giroux - 16.55

Remember that the points totals for Giroux are all accrued whilst also shouldering the extra workload of 3 full minutes more PK time per game than Stamkos, and also doesn't just take into account the 'one season' of Giroux dominance. Also bear in mind that Giroux's linemates for half of this recorded time wasn't close to the quality of Stamkos'.

In the time since both players have started playing first line minutes the gap in this stat widens further, with Giroux's total of 15:15 per point being considerably lower than Stamkos' 17:05. Giroux's taking nearly 2 minutes of ice time less to score each point.

If this wasn't enough, here's a collection of extra stats too.

Face-Offs:
Stamkos - 46.2%
Giroux - 52.8%

Offensive Zone Start (Finish):
Stamkos - 54.5% (51.0%)
Giroux - 47.6% (51.4%)

On-Ice CORSI:
Stamkos - 1.09
Giroux - 8.76

Penalties Drawn Per Game (Taken):
Stamkos - 1.0 (1.0)
Giroux - 1.3 (0.4)


With these stats in mind, is the gap between the two players as big as you'd think? Aside from goals (and subsequently SH%) and hits, Stamkos doesn't really beat Giroux in anything. Giroux has a better face-off percentage, he starts far less shifts in the offensive zone but ends more, his CORSI absolutely crucifies Stamkos' and he draws more penalties whilst taking far less. Also, Stamkos has 29 points against his own (terrible) division. Malkin and Giroux play 6 games against 4 of the Islanders, Rangers, Devils, Penguins and Flyers. Stamkos plays 6 games against the Caps, Jets, Panthers and Canes......

So why is Stamkos perceived to be so much better than Giroux?

Also Stamkos QoC is .031 where as Giroux's is .069 which is far superior. Stamkos Corsi Rel QOC is .339 where as Giroux's is .877. Giroux is the better 2-way player by far. And for that guy too say that goalscoring is the only thing important for forwards is extremely naive.
 

Les Wynan*

Guest
Wasn't directing it at anyone specifically, but a few posts I read made me facepalm.. I don't think one of them was yours.

People don't realize that brilliant hockey minds have spent countless hours working on studies, analyzing data for predictive value, AND watching the games to see how and what it correlates with - these aren't just wild concepts thrown together by a bunch of mouth breathing yahoos with neckbeards. These are concepts proven by people who have done contract work with NHL teams.

For sure. If anyone's interested in digging deeper into this stuff, I'd recommend they go through the archives of Vic Ferrari's old site. If there's a Bill James of hockey, it's him and his ability to explain the concepts is more or less unmatched.
 

GKJ

Global Moderator
Feb 27, 2002
185,669
37,463
Addendum: the best stat (and easy to understand) is PDO.

The opposing goaltenders save percentage at 5on5 (aka your team's shooting percentage at 5on5) plus your own goalie's save percentage at 5on5.

For example: the Bruins through the first 10 games were 3-7 and their PDO was 970ish - aka they definitely were playing much better than a 3-7 team but were unlucky.

After their insane 23-3-1 run, their PDO had climbed to 1050... They were (are) clearly a very good team, but definitely not THAT good..

The Leafs had a pretty high PDO for a long time - which is why they were very likely to drop off in their performance. Of course, their recent run of (correct me if I'm wrong) 2-14-2 saw their PDO drop off a cliff - they weren't as good as their record was, but they're also not nearly as bad as that recent string indicates.

Remember when Kopitar was leading the league in scoring in November a while back? High PDO. Kessel leading the league earlier this year? High PDO. The Sedins stinking it up recently? Low PDO. Vermette having a brutal year in Columbus this year? Low PDO. All of these trends have, or will, almost surely reverse at some point.

I don't know if you're being facetious, but regression to mean is one of the core values of advanced stats.

The Wild were the worst team 5 on 5 in Corsi and Fenwick (I think) when they were tops in the league in points, and their injuries more or less insulated their downfall.
 

Sasso09

Registered User
Jan 2, 2009
12,409
1,892
Chicago
I believe this is the post you're referring too.




Also Stamkos QoC is .031 where as Giroux's is .069 which is far superior. Stamkos Corsi Rel QOC is .339 where as Giroux's is .877. Giroux is the better 2-way player by far. And for that guy too say that goalscoring is the only thing important for forwards is extremely naive.

Eye peeling compairison.. Like he said though. What about the other top scorers, could you do top 5?
 

eklunds source

Registered User
Jul 23, 2008
8,323
0
Ed Snider's basement
I have an hour to kill, what the hell...

People use them around here like they are gospel, when are they are really telling you is about shots for/shots against.

They can add a dimension of analysis, but a lot of people mistakenly rely on them as a proxy for a player's overall effectiveness.
Corsi and Fenwick are roughly 90% accurate to possession of the puck. Possession of the puck is what wins games. Babcock himself said "possession is everything". If you have possession of the puck for 60 minutes of a hockey game, it's almost impossible to lose - if you have the puck, the other team doesn't, and if the other team doesn't have the puck, the other team doesn't shoot it.

If a player has a Corsi ratio of 45%, that means, about 45% of the time, the opposing team has the puck in your defensive zone. How is that NOT measuring a player's effectiveness?

Obviously sample sizes play into it; there will be shifts occasionally where a player has very little impact on what happens on the ice, and there will be bad games here and there, but over a large enough sample size, Corsi is an incredible tool for measuring a player's effectiveness.

The thing about hockey is it's a team sport, and team stats are a better predictor for wins than individual stats. Stats in baseball are very useful because baseball is really an individual sport. It's always 1 vs 1.

A good example is: Let's say you have your number 1 D-man playing 30 min a night and he's awesome with those minutes. You also have a #3 playing 20 min a night and is also awesome. Good point totals, good +/- etc. If that top D gets injured and then the #3 has to cover the extra minutes, quite often you'll see his point totals dip along with the rest of the stats because he cant handle the extra minutes.

Jay Bouwmeester (I think) is a great example of this. I think if he played less minutes you'd see better point totals out of him because playing the minutes he does he can only really play defensively, and can't really be physical due to fatigue.
I think you have the right concept, but you're applying it the wrong way. It's not that added minutes decreases a players' effectiveness; it's that a #3 dman typically doesn't face the top opposition. When a top defenseman goes down, SOMEBODY has to take over the role of shutting down the opposing top line, and there are very, very few teams in the league who have a 2nd line that is equally talented offensively.

Adding minutes to a player isn't really a problem (unless their conditioning is an issue); it's what kind of minutes they get. Most players in the league will perform better if they're given more even strength minutes on an offensive zone faceoff, or against 3rd/4th lines and 3rd defense pairings, or more powerplay minutes... The same way most players in the league will perform worse if they face tougher competition, start in the defensive zone more, or get more PK time.
True, something that measured scoring chances in the same fashion, rather then shots, would likely be better, or at very least a good compliment.
A lot of teams have their scoring chances tracked by a 3rd party; for more information check here but the results really aren't terribly surprising. Good teams have a high Corsi and get more scoring chances, bad teams have a poor Corsi and give up more scoring chances than they create.

The Rangers often don't pursue the puck hard in the defensive zone. They just clog the middle of the ice, letting the opponents cycle the puck along the perimeter while they clog the shooting lanes, to ensure the shots are either blocked or wide. It's not surprising they have a poor Corsi, it doesn't, in my opinion, indicate they are any poorer of a team, it will just take 5 years of Dan Girardi's life.

Tying lack of injuries to Corsi is a point that needs further explanation.
The system a team plays certainly impacts their Corsi/Fenwick numbers... but at the same time, the more the Rangers allow the puck in their zone to float around, the less time they have to put the puck in the net at the other end.

In a game where a lot of goals happen because of lucky bounces, the more time the puck spends in the offensive zone, the more likely it's YOUR team that benefits from the bounce.

That's the same reason people don't think Preds are good. Rangers and Preds are all about preventing good scoring chances and allowing shots they know their goaltenders will have a good chance stopping.

Any system that treats every shot the same will be mostly blind to why these teams are succeeding. Not all shots are created equal.
Nashville is in the same boat as what I said above; they are riding an extremely good goaltender and good defensive system, along with an extremely high shooting percentage.

It's not possible to consistently create better scoring chances on your own without giving them up, which makes sense if you apply it to the game. How often will you see a risky pinch from a Nashville defender? It doesn't happen; that's why they don't give up many odd-man rushes and that's why Rinne is a fantasy goldmine as long as Trotz is around. Compare that to a team like Vancouver, who sends a defenseman as a 4th forward frequently, and instead rely on above average goaltending to bail them out from odd-man rushes.

New York and Nashville are high in the standings, yes, but they're also 4rd and 3rd (respectively) in 5v5 shooting percentage. If (and when) that shooting percentage drops out from under them, their record is going to crater. If you don't think shooting percentages can crater, look at Toronto and Boston's seasons...

They have a strategy, and so far, it's working. It's not a coincidence that both are getting unbelievable performances from workhorse goaltenders too... I'm just saying that their strategy is working, but probably not as reliable. Could it get them a cup? Certainly; stranger things have happened... but as Holland said, the best you can do is a build a team that will compete for the cup, and hope you win it every 5 years or so. You need a lot of things to go right in the playoffs to win a cup - you need your stars to perform, your role players to exceed expectations, solid to great goaltending, and good special teams. In any given playoffs, at least one of the top-8 teams will have all those things going for them and they'll be very, VERY tough to beat.

Relying on something uncontrollable like bounces (a.k.a. shooting percentage) is just one more thing that can go wrong, and it's why I'll bet against Nashville and the Rangers (despite the fact I like both teams. Nashville moreso.)

Corsi may tell us something at the team level but it's near useless when considering individuals. Problem is that you need to be a fanatic in order to use it for individual players. The results often aren't pretty. These are some posts from a Scott Gomez thread made by a person who shall go unnamed.

In my experience at least, this kind of attitude is the norm rather than the exception when discussing something with the advanced stats crowd. Primarily through Outscoring Champion of the World Shawn Horcoff.

The best way to learn about a player is to watch him play. Stats can be useful, but they completely skew reality when used as a primary method of analysis. No pun intended.
Whomever posted that stuff about Gomez has the right idea, but like I said previously, probably did more damage to his idea because he fumbled it from the start.

Cam Charron explains it better than I could:
Consider puck possession. One of the best stats for evaluating team performance is score-tied Team Fenwick rate (all shots at the net minus blocked shots for and again), a good measure of puck possession and an excellent predictor of future team success. Using data obtained via timeonice.com here, here, here, here and here, I filtered out the stretches of games that Scott Gomez missed this season.

I found that with Gomez in the lineup, the Canadiens have had a team possession rate with the score tied of 53.5%. Without him, it was 44.6%. Why is this important? Well, here's the overall team rank with Gomez in the lineup:

Rank|Team|Fenwick Rate
3|Pittsburgh|54.9%
4|Chicago|53.9%
5|Montreal|53.5%
6|San Jose|52.6%
7|Boson|52.4%
...and without...
Rank|Team|Fenwick Rate
26|Buffalo|46.7%
27|Anaheim|46.0%
28|Montreal|44.6%
29|Nashville|44.3%
30|Minnesota|43.7%
The rest of the article is worth a read. Gomez is, in a lot of ways, like Mason Raymond - a guy taking a ****-kicking from fans, but in reality, contributes a lot more than people notice.

Another great read regarding Gomez can be found at Habs Eyes On The Prize.

Advanced Stats aren't meant to be used as absolute gospel, and people who write about/using advanced numbers will tell you that.



That's why it evolves to Fenwick.

Raw Corsi is also good to understand for beginners, but Corsi Rel and Corsi Rel QoC are better indicators because they used other factors.

The reason this stuff gets used as a proxy is because no one can possibly watch all 1230 NHL games let alone watch all 1230 NHL games AND properly evaluate players.





This is the problem Nashville runs into, because they've been proven to be successful at it for years, but it wouldn't translate into the playoffs since it's not the regular season.

http://blogs.thescore.com/nhl/2012/...edicting-nashvilles-regression-five-theories/



http://www.arcticicehockey.com/2011/11/28/2591407/fun-with-pdo

Unless you honestly think Kent Huskins that good, Shot Quality is roughly 11% of the game. Advanced Stats covers the other 89%.
This needs more attention. Great post.
Devil fans use them a lot... but I ignore them...

I've played enough hockey, watched enough games at EVERY level... Peewee, midget, bantam, juniors, echl, ahl, swedish elite league, khl, nhl, international, and so on that I don't have to look at stupid corsi numbers to find out if a player is a good player, playing well, and so on.
No matter how much hockey you've watched, you haven't watched as much as Don Cherry. He said advanced stats were freaking useless too, and was pissed off when people used them to declare Ryan Johnson a terrible player..

That was almost exactly two years ago. Now? Ryan Johnson is out of the league, released from training camp after dressing for just 34 games in Chicago the year before.

People don't always know what to look for in a hockey player... For example, someone might see a player with excellent hustle to negate a scoring chance and think "hey, that was a great play" but they don't realize that the scoring chance wouldn't have happened if the player hadn't been out of position in the first place.

If you think you've got that mastered, then that's fine, but I don't think it's a coincidence that the GMs who have been more or less confirmed as using advanced statistical analysis are:

Chiarelli (Boston, recent cup winner)
Shero (Pittsburgh, recent cup winner)
Gillis (Vancouver, recent cup finalist / presidents trophy)
Holland (Detroit, perennial contender)
Wilson (San Jose, back to back WCFs)
Maloney (Phoenix, consistently outperforms his teams' perceived talent level)
...versus guys who are more or less confirmed to think they're baloney...
Burke (Toronto)
Howson (Columbus)
Tambellini/Lowe (Edmonton)

I don't think advanced stats are everything - certainly you also need a coach who can properly evaluate how to deploy players:
In Buffalo, Hodgson is playing an expanded role, facing superior competition, and starting equally as many shifts in the defensive-zone as he's starting in the offensive-zone. As a result, Hodgson's Corsi numbers have run off a cliff.

With Hodgson on the ice, Buffalo has controlled 38.1% of on-ice shots, and with the score tied that numbers gets even uglier (34%). That's down significantly from 48.9% of shots that the Canucks managed to control with Hodgson on the ice.

...you also need to look at chemistry (getting more from the sum than the individual parts), systems, etc. You can't just grab the best Relative Corsi players in free agency and start winning.

That said, they're a powerful tool, and ignoring them when their results are becoming more and more proven is a bit silly in my opinion.
Unlike advanced stats in baseball, there are too many factors that these stats don't take into account which make them all but useless.
What makes them useless?

Corsi has been proven to correlate highly with possession of the puck - if a team takes 60% of the shots when <player> is on the ice, then it's reasonable to assert that the team has possession of the puck about 60% of the time.

If you look at what kind of competition he faces (the average Rel Corsi of the opposing players while he's on the ice), where he starts his shifts (does his coach put him in defensive or offensive situations), how lucky/unlucky he's been (his teams shooting percentage and save percentage with him on the ice)... How does that not paint a picture of what kind of results he produces?

Hockey is a fluid game and it doesn't lend itself nearly as easily to stats as baseball does. Mircrostats are interesting and sometimes illuminating but I don't give them much weight. On our forum we've had people defending Gomez for two years because his CORSI is good. It's like CORSI has become the standard by which a player is successful or not. Way too much weight is given to them by some people and actual production takes a back seat.

If you've ever watched Gomez, he takes the puck brings it into the opposing zone very well but then... nothing. He's pretty ineffective and his point totals are a reflection of his uselessness. However, his CORSI is strong and so some folks assume he's playing great hockey. CORSI and other microstats are interesting but they shouldn't replace actual totals when analyzing a player's offensive performance.

People who point to Moneyball and baseball are missing the boat. With every at bat in baseball there is an empirical result. It is a statistically driven game that lends itself to that kind of statistical analysis.

Hockey is a much more fluid game and it's difficult to nail down with stats. Look at hits for example. All hits are measured the same. But does a hit from Brian Gionta feel the same as one from Chara? No. But there's no way to measure this. There's no way to measure the impact of what that hit has except via maybe... wins or +/- but even then there are tons of variables like goaltending, opponent ect...

As another example I"m not sure how CORSI is supposed to meaningfully show us how effective a player really is. It certainly doesn't always match up with production from players. Gomez is exhibit A. As I mentioned above, he was terrible despite those CORSI numbers.

I would expect that Brett Hull would be exhibit B. That guy didn't hold onto the puck too much. He didn't carry it into the zone and he didn't dipsy doodle around defenseman. But he was one of the best offensive threats of his generation. Puck would be on his stick and then in the net. I would guess that his CORSI wouldn't be all that strong but... so what? The guy was a beast. Again, CORSI might tell you that he wasn't a puck possession guy but I don't think that really means all that much in the grand scheme of things. It's certainly not an accurate predictor of how effective a player he was.

Moreover, there's the problem of the accuracy of the microstats themselves. Look at giveaways/takeaways... it's an official NHL stat and there's huge variances all over the league. If the NHL can't get it right on something that seems so simple then I'm not sure how we can really trust a bunch of bloggers with microstats that aren't officially kept by the league itself.
These aren't just bloggers creating fancy concepts out of a basement. These are concepts created by people who actually do contract work for NHL teams. Corsi is named after someone who worked with the Buffalo Sabres organization. Successful teams track Nielson numbers (basically scoring chance +/-), which is named after a successful NHL coach.

You used the word Corsi a lot, but if advanced statistics are a painting, Corsi is the type of paint used. It's important, but absolutely USELESS on it's own. If you don't understand how successful players contribute to their high Corsi, or why poor players contribute to a bad Corsi... If you don't understand (again) how a coach is deploying a player - there's a HUGE difference between facing the 3rd/4th lines and getting offensive zone starts, and facing top competition starting in your own end... If you don't look at underrated factors such as faceoff %, penalty-drawing (which is a proven talent)... Looking at Corsi alone easily is more harmful than advantageous.



Can someone post the advanced stats of the top 5 scorers? I expect Giroux to have highest qual comp and most defensive draw percentage.
The thing is... the top-5 NHL scorers aren't the best offensive players in the game. They're probably top-15, and CERTAINLY top-50, but there's so much "noise" in point totals, because points happen so rarely.

Here's what I mean: When Ryan Getzlaf has been on the ice this year (and I mean at 5v5 only), the Ducks have shot 7.12%. That's below league average, and 8th of 10 Ducks forwards with 40+ games played. Not coincidentally, he's having a very "poor" year with 48 points in 71 games (55 point pace).

Last year, with Getzlaf on the ice, the Ducks shot 11.97%. That's significantly higher than league average - and the abnormally high shooting percentage is what the Ducks rode to a 4th seed (despite a mediocre-at-best goal differential). Also not coincidentally, Getzlaf had a "great" year with 76 points in 67 games (93 point pace).

Looking at his other 'metrics' comparing the two years:
Year|5v5 TOI (rank)|Quality of Competition (rank)|Offensive zone starts|Relative Corsi|Team shooting percentage with Getzlaf on the ice
10-11|17.06 (1st)|0.709 (4th)|46.6%|+12.5|11.97%
11-12|16.43 (1st)|0.812 (3rd)|47.8%|+13.0|7.12%
Comparing the last two seasons for Getzlaf... He's gotten slightly less ice time, though both years he led the team. He's facing slightly tougher competition, ranking him roughly the same on the team though. He's starting in the offensive zone just a little bit more often. He's pushing play forward extremely well despite all that, both years, with a very good Corsi - considering he's facing the toughest opposition, starting in the defensive zone more often than not, and getting a lot of ice time, he's actually a fantastic two-way player. Does a great job of getting the puck from the D to the O zone.

Pretty much the only significant difference? Last year he was incredibly, incredibly lucky (as was Perry, who rode the same luck train to a Richard and Hart trophy) and this year, despite facing similar competition, despite starting his shifts in the same ratio of d-zone to o-zone, despite getting a similar amount of ice time, his offensive production has dropped off a cliff.

So.. you can see how much the percentages play into points... and how the percentages are unsustainable and mostly 'luck'.

That's why the Art Ross has become so difficult to win and we've seen so many new winners.. There are so many great offensive players that you need to be great offensively AND get some lucky bounces to win.

Corsi is the stupidest stat ever I'm pretty sure
That was the stupidest post ever I'm pretty sure.
That's the other thing, and to fully buy in may mean to compromise your own way of thinking. Not an easy thing to convince someone to do (and when ASA people can't do it, that's when some of them dismiss people as stupid, dumb, etc.)

It doesn't make anyone dumb if they don't get it or immediately buy in. It's just like learning anything else, learning one thing before the next.

Or start with James Mirtle:

http://www.theglobeandmail.com/sports/hockey/the-moneypuck-revolution/article2178766/

http://www.theglobeandmail.com/spor...ed-stats-caught-on-in-the-nhl/article2178777/

http://www.theglobeandmail.com/sports/hockey/hockeys-new-numbers/article2178781/
Another beauty. GKJ is nailing it.

Cheerio, as you were...
 

The Bored Man

5-14-6-1
Jul 2, 2009
7,009
1,150
Edmonton
And you're arrogant enough to believe that by watching the game you can make more accurate assessments of player value and more reliable predictions about player performance than through statistical analysis. It cuts both ways.

I'm not the smartest hockey fan in the world. I've been wrong many times about players and teams. But I put more stock in my opinion than a stat that tells me that relative to competition, Olli Jokinen(ranked #2) is better at driving the play than Steven Samkos (not in the top 30). Tom Gilbert leads MIN defencemen in CorsiRel QoC. Ask Wild fans what they think of him. I see Horcoff, Hemsky, and Smyth -- players that I watch a ton -- being valued so highly by a statistic and can't bring myself to put much stock in it.

Is it just a matter of the stats being abused? Maybe. But every discussion I've had about players that involves Corsi / Fenwick / PDO / sex offenders per square kilometre* makes me want to tear my hair out.

I'm not a fan of appealing to authority but when you see how much success Vancouver, Boston, San Jose and Pittsburgh have had using similar (albeit more "advanced" - hey it's one of your favorite words!) metrics that emphasize team possession and a player's contributions to team possession it's hard to argue that "watching the gamez!!!!1" is the be-all end-all or anywhere close. Not to mention that for most people, perhaps not you, "watch the gamez!!!1" as an anti-stats argument means "read the box score and see who had the most points."

Those teams are using Corsi and Fenwick? Link please. The closest I've ever heard of these stats being used in the NHL is from Brian Burke, who called them bull****, and Ron Mclean, who called them garbage. Brian Burke is a bit of an idiot and McLean is just a reporter so I'd be very interested in seeing if these statistics are being used by teams in the same manner as on the internet. On a similar note, I'd also like to see academic studies that support these statistics. Not blogs, y'know? Peer-reviewed university publications.

Call it a philosophical difference. This sport doesn't lend itself to this kind of analysis. I can't value an Alex Ovechkin rocket the same as a Ladislav Smid muffin. I appreciate that supporters of statistics put such effort into finding new ways to understand hockey. But everything I've seen so far doesn't come close to what you can learn by "watching the gamez!!!"

------------------------------------------
*http://www.battleofcali.com/2012/3/15/2869085/sharks-gameday-staccuracy
 

GKJ

Global Moderator
Feb 27, 2002
185,669
37,463
If you think you've got that mastered, then that's fine, but I don't think it's a coincidence that the GMs who have been more or less confirmed as using advanced statistical analysis are:

Chiarelli (Boston, recent cup winner)
Shero (Pittsburgh, recent cup winner)
Gillis (Vancouver, recent cup finalist / presidents trophy)
Holland (Detroit, perennial contender)
Wilson (San Jose, back to back WCFs)
Maloney (Phoenix, consistently outperforms his teams' perceived talent level)
...versus guys who are more or less confirmed to think they're baloney...
Burke (Toronto)
Howson (Columbus)
Tambellini/Lowe (Edmonton)

Eh, don't really think Brian Burke has bought in:

http://www.arcticicehockey.com/2012...and-mike-milbury-on-advanced-stats-in-the-nhl
 

The Bored Man

5-14-6-1
Jul 2, 2009
7,009
1,150
Edmonton
If you think you've got that mastered, then that's fine, but I don't think it's a coincidence that the GMs who have been more or less confirmed as using advanced statistical analysis are:

Chiarelli (Boston, recent cup winner)
Shero (Pittsburgh, recent cup winner)
Gillis (Vancouver, recent cup finalist / presidents trophy)
Holland (Detroit, perennial contender)
Wilson (San Jose, back to back WCFs)
Maloney (Phoenix, consistently outperforms his teams' perceived talent level)
...versus guys who are more or less confirmed to think they're baloney...
Burke (Toronto)
Howson (Columbus)
Tambellini/Lowe (Edmonton)

I mentioned this to Les Wynan but you might have the information handy: how do you know that they're using these statistics? More importantly, using the statistics you'd find on behindthenet and PDO etc. And using them in the manner that's being discussed here?
 

Les Wynan*

Guest
I'm not the smartest hockey fan in the world. I've been wrong many times about players and teams. But I put more stock in my opinion than a stat that tells me that relative to competition, Olli Jokinen(ranked #2) is better at driving the play than Steven Samkos (not in the top 30). Tom Gilbert leads MIN defencemen in CorsiRel QoC. Ask Wild fans what they think of him. I see Horcoff, Hemsky, and Smyth -- players that I watch a ton -- being valued so highly by a statistic and can't bring myself to put much stock in it.

Is it just a matter of the stats being abused? Maybe. But every discussion I've had about players that involves Corsi / Fenwick / PDO / sex offenders per square kilometre* makes me want to tear my hair out.

What is Jokinen ranked #2 in? I assume you're referring to Corsi Rel QoC - that's a stat that shows how Jokinen has been used this season by Sutter and the Flames coaching staff. The fact that he ranks so high in the metric league-wide and first among Calgary forwards indicates he's being deployed against the other teams' best players every night more frequently than any of his teammates. Assuming you've seen a lot of Calgary does this jive with what you've seen (legitimate question, not snark)? If it doesn't, you can take a look at head-to-head ice time charts to check who Sutter is using Jokinen against when he has last change and for how many minutes relative to his teammates.

By Corsi, though, Jokinen has been killed. Which is more or less what we'd expect since he's playing a role he has no real history of success in. His Corsi Rel of -7.0/60 tells us the puck is in the Flames' end of the rink far more frequently when Jokinen is on the ice compared to when he's on the bench. More important context to consider is that he's only starting 47% of his even strength shifts in the offensive zone, making it even more difficult for him to post good possession numbers and rendering his shifts even more of an uphill battle. So while Jokinen isn't terrible by any means just because he ranks among the bottom of the barrel on Calgary in possession (he's playing against excellent opposition and starting in his own zone more often than not - unless he's truly elite, which he obviously isn't, we would expect him to be struggling) he's not great either.

Meanwhile, Stamkos, from your example, plays against much lesser competition and starts about 55% of his shifts in the offensive zone - again, this makes sense as Boucher is trying to put his offensive star in a position to succeed - and has a high Corsi Rel of 6.9/60 as a result. So just looking at raw Corsi or Corsi Rel alone, you could draw the conclusion that Stamkos is much better at driving the play at Jokinen. Considering the role that they're used in closes the gap in their value a bit but Stamkos is pretty clearly ahead of Jokinen - I don't know why you thought the stats said otherwise.

Gilbert is first in Minnesota in Corsi Rel QoC because he was Edmonton's #1 shutdown defenseman. Again, Corsi Rel QoC speaks to the quality of the opposing players a skater faced over the course of the season. Renney always trusted Gilbert to take on the Sedins or Iginla or Duchene or whoever and he performed very, very well in that role. Trading him was another idiot move by the morons in charge in Edmonton.



Those teams are using Corsi and Fenwick? Link please. The closest I've ever heard of these stats being used in the NHL is from Brian Burke, who called them bull****, and Ron Mclean, who called them garbage. Brian Burke is a bit of an idiot and McLean is just a reporter so I'd be very interested in seeing if these statistics are being used by teams in the same manner as on the internet. On a similar note, I'd also like to see academic studies that support these statistics. Not blogs, y'know? Peer-reviewed university publications.

Call it a philosophical difference. This sport doesn't lend itself to this kind of analysis. I can't value an Alex Ovechkin rocket the same as a Ladislav Smid muffin. I appreciate that supporters of statistics put such effort into finding new ways to understand hockey. But everything I've seen so far doesn't come close to what you can learn by "watching the gamez!!!"

------------------------------------------
*http://www.battleofcali.com/2012/3/15/2869085/sharks-gameday-staccuracy

I don't have a link - these aren't things organizations publicize. I've had several conversations with people who've either consulted or been contracted by NHL teams to utilize this data and more advanced forms of it. I understand if you don't want to believe me, that's fine.

No one is saying individual players don't differ in shooting talent - with years of shooting percentage data, we can make educated guesses about a shooter's talent level and predict future performance by gauging how far away from their established clip they're currently finishing their chances. The main takeaway from all of this is that teams don't vary significantly in their overall shooting talent.
 

MessierII

Registered User
Aug 10, 2011
27,646
16,191
Corsi is useless. Corsi combined with zone starts, quality of competition etc is somewhat usefull but if you combine good old +\- with all of that you get a more accurate reflection of an individual's performance. It fails to take into account the scoring chance which is an absolutely massive omission. How was Sidney Crosby the 87th most effective player in the league last year and 8th most effective on his team. It's a usefull TEAM puck possession stat but when it comes to evaluating individual players it completely falls apart.
 

Ad

Upcoming events

Ad

Ad

-->