The Eye-Test Vs. Corsi: Canucks Game 1

[JamesB]

The Canuck game 1 provides a very good illustration of the contrast between using the "eye-test" to evaluate player performance and using advanced stats, particular shot metrics such as CORSI and FENWICK. Personally, I like some advanced stats but I do not like these shot metrics. Game 1 against Edmonton illustrates the problem. First, for anyone interested, the shot metrics are available here: https://www.hockey-reference.com/teams/VAN/2018.html#all_stats_adv_rs

 

The eye-test summary of the game runs as follows.

 

1. First, Travis Green,  the reporters (even Botchford), the players, and most of CDC thought that the Canucks played a good game against a good team. After all, the Canucks won the game and a lot of guys looked good doing it.

2. The first star of the game was Horvat, who scored two impressive goals and looked great -- fast, strong, high intensity.

3. The "second star" of the game was the Sutter line (with Dorsett and Granlund) who shut down arguably the best line in the NHL (with McDavid and Draisaitl) and scored a goal.

4. Markstrom let in a questionable first goal but was otherwise very good.

5. If there was a disappointment in the game for the Canucks it was probably the Sedin line (with Vanek) who looked slow out there and did not do much on the PP despite lots of PP time.

 

The CORSI (shot attempt) summary is completely different and illustrates the weakness of CORSI numbers. Here is the even strength CORSI summary.

 

1. The Canucks had a bad game, getting crushed by Edmonton on shot metrics.

2. The best Canuck line by far was the Sedin line. The top 3 forwards in CORSI were Daniel (61.5%), Vanek (57.1%), and Henrik (50%).

3. The worst Canuck line was the Sutter line. Their shot metrics were Granlund (22%), Dorsett (25%) and Sutter (26%).

4. The Horvat line was not much better: Horvat (32%), Baertschi (32%), Eriksson (33%).

 

This comparison illustrates my pet peeves about CORSI.

 

1. Most importantly, CORSI does not adjust for strength of competition, which is a huge factor. The difference between going up against the McDavid line and going up against the Edmonton bottom 6 is huge in terms of expected CORSI. The Sutter line was able to keep the McDavid line from scoring, forcing them into a lot of missed shots, blocked shots, and  low percentage shots, all of which count in CORSI. They did a great job. The Sedin line played the Edmonton bottom 6 about even -- which is not such a great accomplishment. There are some attempts to adjust for quality of competition but they are woefully inadequate. The explanation is long, but it is very hard to adjust properly for quality of competition. 

 

2. CORSI is usually reported for even strength play. But special teams are obviously very important. Factoring special teams in correctly is difficult. Once again, in this game, Sutter, Dorsett and Granlund were good on the PK. The Sedins did not play PK but they led the forwards in PP time and did not do much in that time.

 

3. Shot quality is obviously important. Generating low quality shots, missed shots, and blocked shots is obviously less valuable than driving the net (as Horvat did on his first goal) or getting close-in chances (as on Horvat's second goal). There are attempts to adjust for shot quality, but it is hard to do well. Horvat and Baertschi go not generate a lot of low quality chances but play hard to get high quality chances. This hurts their CORSI numbers.

 

The point of this post is not to criticize the Sedins. They have had great careers and I hope they have a good year this year. They were okay in the first game. And my point is not to criticize advanced stats. However, looking just at even strength CORSI is often very misleading. Getting good advanced stats is difficult. I am, for example, also not a fan of GAR (goals above replacement) or WAR (wins above replacement), largely because they fail to account for quality of competition (and quality of teammates). 

 

Stats I like include actual zone time metrics, defensive zone exit stats, and metrics based on high quality scoring chances. However, all of these should be adjusted for quality of competition and quality of teammates.

 

Any comments would be appreciated.

 

 

[-AJ-]

CORSI, like most stats, lacks a lot of context. In any statistical analysis, a higher number of metrics should result in better accuracy. CORSI is just the new +/- stat to me, except it's a little less obvious. I do like to look at it, but I try to look at other statistics a bit to complement it when doing some analysis (admittedly, I don't look at super advanced statistics most of the time though).

[chon derry]

some of these stats you don't really see as the game is unfolding ,I guess when you've noticed a trend in the stats from game to game then its helpful ,but how many times have you seen the blackhawks   in the game but getting out shot and out scored and then they seem to ,or used to be able to score at will, but they were an exception to that. stats go out the window when something like that happens.

[Rick Blight]

I will supplement the eye test with advanced stats but rely primarily on the eye test.

If I wanted to rely primarily on advanced stats I would probably become an Arizona fan and that ain't going to happen!

[bloodycanuckleheads]

Stats like Corsi are completely useless when looking at a small sample sizes.  But, they are extremely valuable when looking at larger sample sizes.

 

They also need to be taken with a grain of salt.  NHL coaches sit back on leads.  So, if you have a team (like our's last year) who tend to down a goal or two in the early-going, that means that the opposing team isn't going to be doing very much for the next 55 minutes, other than playing defense.  So, after the game, the terrible team that was out of the game right from the get-go will actually have decent Corsi numbers.  Because, they controlled all the shot attempts.  Not because they were any good, but because they were so bad that the other team stopped trying.

 

If you look at our Corsi numbers from the last couple seasons, they aren't too bad.  We look like a decent team.  Even though weren't a decent team.  At all.

 

So, yeah, looking at Corsi for single-games is pointless.  But, looking at them over the course of a couple seasons on the other hand...

 

[Fantomex]

Advanced stats can be skewed and often misrepresent the actual game. They're great to supplement the eye-test, but I wouldn't judge any game/player on advanced stats alone. Sure the corsi says the Granlund-Sutter-Dorsett line wasn't controlling the play, but anyone that watched the game would tell you that they were highly impactful in the role they were given.

[oldnews]

18 minutes ago, JamesB said:

The Canuck game 1 provides a very good illustration of the contrast between using the "eye-test" to evaluate player performance and using advanced stats, particular shot metrics such as CORSI and FENWICK. Personally, I like some advanced stats but I do not like these shot metrics. Game 1 against Edmonton illustrates the problem. First, for anyone interested, the shot metrics are available here: https://www.hockey-reference.com/teams/VAN/2018.html#all_stats_adv_rs

 

The eye-test summary of the game runs as follows.

 

1. First, Travis Green,  the reporters (even Botchford), the players, and most of CDC thought that the Canucks played a good game against a good team. After all, the Canucks won the game and a lot of guys looked good doing it.

2. The first star of the game was Horvat, who scored two impressive goals and looked great -- fast, strong, high intensity.

3. The "second star" of the game was the Sutter line (with Dorsett and Granlund) who shut down arguably the best line in the NHL (with McDavid and Draisaitl) and scored a goal.

4. Markstrom let in a questionable first goal but was otherwise very good.

5. If there was a disappointment in the game for the Canucks it was probably the Sedin line (with Vanek) who looked slow out there and did not do much on the PP despite lots of PP time.

 

The CORSI (shot attempt) summary is completely different and illustrates the weakness of CORSI numbers. Here is the even strength CORSI summary.

 

1. The Canucks had a bad game, getting crushed by Edmonton on shot metrics.

2. The best Canuck line by far was the Sedin line. The top 3 forwards in CORSI for were Daniel (61.5%), Vanek (57.1%), and Henrik (50%).

3. The worst Canuck line was the Sutter line. Their shot metrics were Granlund (22%), Dorsett (25%) and Sutter (26%).

4. The Horvat line was not much better: Horvat (32%), Baertschi (32%), Eriksson (33%).

 

This comparison illustrates my pet peeves about CORSI.

 

1. Most importantly, CORSI does not adjust for strength of competition, which is a huge factor. The difference between going up against the McDavid line and going up against the Edmonton bottom 6 is huge in terms of expected CORSI. The Sutter line was able to keep the McDavid line from scoring, forcing them into a lot of missed shots, blocked shots, and  low percentage shots, all of which count in CORSI. They did a great job. The Sedin line played the Edmonton bottom 6 about even -- which is not such a great accomplishment. There are some attempts to adjust for quality of competition but they are woefully inadequate. The explanation is long, but it is very hard to adjust properly for quality of competition. 

 

2. CORSI is usually reported for even strength play. But special teams are obviously very important. Factoring special teams in correctly is difficult. Once again, in this game, Sutter, Dorsett and Granlund were good on the PK. The Sedins did not play PK but they led the forwards in PP time and did not do much in that time.

 

3. Shot quality is obviously important. Generating low quality shots, missed shots, and blocked shots is obviously less valuable than driving the net (as Horvat did on his first goal) or getting close-in chances (as on Horvat's second goal). There are attempts to adjust for shot quality, but it is hard to do well. Horvat and Baertschi go not generate a lot of low quality chances but play hard to get high quality chances. This hurts their CORSI numbers.

 

The point of this post is not to criticize the Sedins. They have had great careers and I hope they have a good year this year. They were okay in the first game. And my point is not to criticize advanced stats. However, looking just at even strength CORSI is often very misleading. Getting good advanced stats is difficult. I am, for example, also not a fan of GAR (goals above replacement) or WAR (wins above replacement), largely because they fail to account for quality of competition (and quality of teammates). 

 

Stats I like include actual zone time metrics, defensive zone exit stats, and metrics based on high quality scoring chances. However, all of these should be adjusted for quality of competition and quality of teammates.

 

Any comments would be appreciated.

 

 

Good observations - and recognition of some of the fairly obvious limits of what are called "possession" metrics.

They don't really measure possession, however - they measure shot attempt differentials.

And as you've pointed out - if you watched that game - you'd realize that the Canucks certainly didn't have inferior scoring chances to Edmonton - in fact, certainly throughout the heart of that game, Vancouver produced a whole lot of grade A chances, and Laurent Brossoit was simply outstanding and stoned them repeatedly.

 

So - this is clear evidence that context is always crucial - always fundamental to understanding what is actually taking place - and the more you can gleen, the better idea you'll have regarding the limits of 'advanced stats'.  It's still a very nascent field, with serious limitations that a good analyst sees quite readily.

 

Some folks grasp on a single metric like corsi - attempt to suggest that it tells us a great deal about a player - and erroneously come to far too many conclusions based on it.

 

Players like Larsson (Edmonton), Gudbranson, Sbisa etc have been ripped repeatedly by people employing highly simplistic 'analytics' - and unfortunately, it creates a lot of negative value in the perceptions of casual observers.

 

That game has to be broken down much deeper than shot attempt metrics, or a person utterly whiffs on what actually took place.

I'm not arguing that corsi/fenwich are useless - they simply need to be kept in context and their limitations - which are quite substantial - realized.

 

Anyhow - enjoyed your post and love to see people digging in to get better understandings.    Far too much 'analytics' is satisfied with relatively worthless analysis - I'd point out the stuff that people like Travis Yost, CA, etc post on a regular basis, that gets parroted far too often and doesn't really serve an 'advanced' understanding of the game.

Analytics are only as good as the analyst - a truism I take from the brilliant point made by Toni Morrison - "love is only as good as the lover."

[JM_]

@JamesB nice evaluation. 

 

I've been perusing the analytics sites for about a year now, and i do have an R&D background that includes stats training. So what I see lines up with what you are describing here, a significant over-emphasis on corsi because its easy to measure, but also a lot of NHL fans tend to focus on scoring much more than anything else (just my 2cents but that seems to be the overriding emphasis in the media as well).

 

The big failure of hockey analysts imo is a lack of ability to describe a game like Saturday. Where i'd add to your post is zone starts. By looking at that and corsi we should have been brutalized by the Oilers. So.... whats missing from his picture? No ability yet to adequately describe two things: checking lines, and pure defensive play. I haven't seen anything really to describe that well at all, so we're left with noobs like Travis Yost telling us Gudbranson is a hot turd but anyone that watched the game knows how important he was in the win. 

 

There needs to be a statistical analysis that can look at pure defensive play before we can really start to take hockey analytics seriously, otherwise its not really more insightful than basic stats.  

[canucksnihilist]

Interesting thread, thanks for posting!

 

Corsi gives an overall long term view - which may not play out in any game.  If the players never get any shots, they won't score much, even if a few games they have some great goals.

 

What Corsi doesn't do account for talent (quality of shots, as you have indicated).  If McDavid plays like a super-star for 3 shifts in the game, and gets 2 goals - and the rest of the game he is average - his Corsi might not be that great.  But his ability to pick his game up when the opportunity is there should give him a better assesment than Corsi could give him.  I am NOT saying that McDavid is like that - just using him as an example.  IMHO one player who really exemplified this was Forsberg.  He would be invisible for most of a game, then score the game tying goal and set up the winner.  In a game where his Corsi would suck.  Superstars can't dominate every shift, they pick their opportunities - and sometimes have off nights.  But over the course of a season a superstar's Corsi is going to be good, as they have lots of opportunities that an average player won't even see.

 

Anyways - another stat - that measures effectiveness in the top 10 or 20% of the game for that player - that would be a really good number that I would pay attention to.  Maybe it already exists or can be crunched anyways...  From that it would be obvious that Horvat had a fantastic game, and the stats would back it up.  Even if his Corsi for the whole game didn't support it.   To look at "raw corsi" for one game is almost useless imho.  I want the top minutes of each player compared, not the whole game when they are biding their time looking for opportunities... I want to judge them on what they do with those opportunities.

 

Hope that makes sense!  

 

Peace

 

Hope that makes sense...

[Ghostsof1915]

Personally advanced stats are not much better than +/-, in other words without context its more of an indicator. Hockey is not like baseball. 

 

[aGENT]

As some of us have been preaching for a few years, a lack of context and in turn, poor analysis of advanced stats, (especially when combined with most of those stats not being terribly good at measuring defensive play), leads people to erroneous conclusions. Especially with players like say Gudbranson or Sutter...

[oldnews]

55 minutes ago, JamesB said:

The CORSI (shot attempt) summary is completely different and illustrates the weakness of CORSI numbers. Here is the even strength CORSI summary.

 

1. The Canucks had a bad game, getting crushed by Edmonton on shot metrics.

2. The best Canuck line by far was the Sedin line. The top 3 forwards in CORSI for were Daniel (61.5%), Vanek (57.1%), and Henrik (50%).

3. The worst Canuck line was the Sutter line. Their shot metrics were Granlund (22%), Dorsett (25%) and Sutter (26%).

4. The Horvat line was not much better: Horvat (32%), Baertschi (32%), Eriksson (33%).

 

This comparison illustrates my pet peeves about CORSI.

 

1. Most importantly, CORSI does not adjust for strength of competition, which is a huge factor. The difference between going up against the McDavid line and going up against the Edmonton bottom 6 is huge in terms of expected CORSI. The Sutter line was able to keep the McDavid line from scoring, forcing them into a lot of missed shots, blocked shots, and  low percentage shots, all of which count in CORSI. They did a great job. The Sedin line played the Edmonton bottom 6 about even -- which is not such a great accomplishment. There are some attempts to adjust for quality of competition but they are woefully inadequate. The explanation is long, but it is very hard to adjust properly for quality of competition. 

 

2.  In addition to quality of competition - equally important is another aspect of deployment - zone starts.

 

Sutter's line had 14.29% offensive zone starts - as did Edler (and Tanev even lower at 13.3%).

The Sedin line were at 50%.

 

This differential has to be looked at as well and factored into an understanding of those shot metrics.

 

Additionally, Edmonton took a lot of low quality shots - which meant that Markstrom was making saves and stopping play relatively frequently, leading to zone starts for the Coil in the Van ozone.  This is also a factor in ice time - as the Canucks deploy their Dzone start specialists - ie an elite faceoff and shutdown guy like Sutter - to handle these Dzone starts. 

 

Understanding the value of a player like Sutter must be kept in context - or 'analytics' utterly fail to realize how crucial he is to a team's success.  It was likewise with a player like Manny Malhotra, who also handled almost exclusively dzone starts, was an elite faceoff and shutdown guy, and produce much of the real territorial advantage that players like the Sedins could take advantage of.  It's an integrated effort.

 

The thing with Sutter however, is that he can provide those hard matchup and dzone minutes, but he's also an excellent counterpuncher - which makes him, and his linemates a sneaky threat in transition.   Add to that the fact he's a right handed draw and shot, has a good quick release on his shot, is strong on the boards, very effective in puck retrieval, etc - and he's also an effective powerplay presence for guys like the Sedins.  Versatile and valuable.  Which is why it's such a head scratcher to see purported "analysts" utterly whiff on his real value.

 

Sutter has traditionally played with a lot of tweener and borderline bottom six type players - particularly in Pittsburgh.  He generated his scoring very much on his speed, defensive abilties, takeaways and counterpunching transition game,  When you pair him with some talented two way players, he becomes a serious threat.  When you give him linemates like Steve Downie, Nik Spaling, Craig Adams, Joe Vitale, Chris Conner, etc, lock him into dzone starts (behind Crosbly and Malkin) - and then wonder why he didn't generate a great deal of primary assists - and making assumptions and conclusions about that to the effect that he's a horrible playmaker -  you're really failing to understand context and coming to negative value 'analytics'.

 

You could see the other night the frequency with which he, Granlund and Dorsett produces high quality scoring chances against the Coilers top line of elite scorers.

 

That is extremely frustrating for the opposition - and it's largely unaccounted for by casual observers - those guys weren't just shutting Edmonton down - they may have been outshot, but in the end, they outscored their matchps 5 on 5 - and that's despite the fact Granlund was stoned on numerous grade A chances, including a couple breakaways.

[Orcasfan]

Good post! Needed to be stated boldly. And that first game was a great example of the hollow reality of using stats like Corsi to assess any one game. Unfortunately, there is one website that is built on doing just that....Canucks Army. And JD Burke's analysis of Saturday's game on that website was so off-kilter from any other post-game perspectives, that you gotta wonder how long the media's love affair with that bunch of stats guys is going to last! They have lost a lot of regular visitors over the past couple of years because of their idealogical adherence to a skewed perspective about hockey.

 

Personally, I think the Canucks, like a lot of teams, utilize stats in a reasonable way. They are only one of the data tools that are used to flesh out a player's effectiveness in the role they are assigned. Certainly, like the infamous plus/minus stat, corsi stats would be only minimally helpful. This seems fair. 

[Angry Goose]

I know this gets said all the time but advanced stats can be a nice tool for noticing trends, identifying particular aspects of the game etc. but ultimately evaluating players yourself in the context of how they play is more important.  Look at the highlight of the Edler breaking up a pass with his feet (then subtly moving the puck) that could have led to a major scoring chance (if not a goal).   I dont know how advanced stats rates the 'quality' of a play like that but that was a significant defensive play.  

[Canuck Surfer]

Corsi is just a stat.  It does measure performance. But not in a complete way.

 

Apparently Florida's billionaire owner bought in? Buying a stock just because it is at a current low is also worrisome. Repeated often enough, trending can produce (slightly) above average winners. But knowing why a business is trending down, and having (knowing and being confident the company has in a simple stockholders case) the company has an effective action plan that will correct performance is better. An investment in new products that will increase its sales? A strategy to consolidate / save costs that will inversely increase profits. Stock should only go up on the real prospects of increased profits, or down because profits are shrinking. Buy stocks when you know the fundamentals. 

 

Corsi also should just point to fundamentals. Nothing more. If a player has poor Corsi, it may relate, as Old News has pointed out many times, to deployment in the defensive zone. Which is not a bad thing. Or it could relate to you constantly getting beat? Or having plays take place because you were out of position, or made the wrong read. The coach should know & understand the difference. The ability, and preparation, to compete in each situation a player will face is complex. It would be great if we could also have stats like winning contested pucks (in the corners). But really good coaching will have systems that players can execute fluidly. And breakdown the individual tasks to execute. Practice the tasks to win each battle the player will face, train to be the physically dominant (or faster, or more skilled) player in each situation.

 

And then take a quick peak at Corsi to see who might need help. 

 

 

[189lb enforcers?]

Meh, I like the eye test.

 

Look at Weber.

Watch him and notice the puck is never on his side of the ice because his opponents stay the &^@# away from him. No stat, just an eye test.

 

Look at Kes and Burr's coming out year.

That eye test was just having Ryp making their game and life expectancy go way up for obvious reasons. 

 

I like how Guddy kept the Coilers players in check infront of the net and after whistles. No stat for that.

 

My eye test noticed that the twins spent more time trying to get into the offensive end on the PP than the other lines did. Would like to actually look at that and compare it to an objective stat. 

[Baggins]

Nothing beats the eye test. The rest are tools to consider but need to be weighed with additional information not individually.

[oldnews]

5 minutes ago, Baggins said:

Nothing beats the eye test. The rest are tools to consider but need to be weighed with additional information not individually.

Depends on the eyes though.

 

A lot of 'eye tests' depend on eyes that need to be tested / a set of glasses and/or the crap wiped out of them lol.

[Nucksfollower1983]

analytics work for a slow predictable game like baseball where the play in any given situation is obvious but hockey is too fast and dynamic there are many situations when it comes to analytics that say the outcome of a game should have been different or a given player should have better stats. Look at Florida last year, they went all analytics happy and it totally blew up in their face. when it comes to hockey i'll go with the eye test as opposed to fancy stats.

[Baggins]

10 minutes ago, oldnews said:

Depends on the eyes though.

 

A lot of 'eye tests' depend on eyes that need to be tested / a set of glasses and/or the crap wiped out of them lol.

There's truth to that. Some watch with rose colored glasses while others watch with their blackened hate glasses.

 

I like to believe I'm pretty objective.