Corsi, shot quality, and the Toronto Maple Leafs

[Ho Borvat]

I agree - that level of tracking would be the Holy Grail of hockey analysis.

The thing that confuses me is that the NHL tracks things like hits, blocks, giveaways, and takeaways (which are quite subjective)

But they don't track things like
- Penalties Drawn
- Zone Entries
- Shot Differential (take away the fancy names, call it what it essentially is, dumb it down)
- Time of Possession

This stuff doesn't have to be advanced, but really the NHL is dropping the ball. Its left to 3rd party websites/bloggers to track this information which makes it easy to write-off as "Advanced Stats" when its really not that advanced.

Its kind of weird that the guys at TSN will reference Capgeek, but not BehindTheNet

 

[Nithoniniel]

Interesting article. Shot quality effects tend to be underestimated quite often in the stats community.

However, I do take issue with Birnbaum using raw Corsi (it looks like he's using raw Corsi because 44.1% was the Leafs' raw 5-on-5 Corsi last season) and comparing it with shooting percentage. High shooting percentages lead to more goals, which means the team is more frequently in the lead and score effects will come into play, lowering their Corsi. His conclusion is that Corsi is "too biased by situation and shot quality to be taken on its own". But this is something most analysts already know - and that's why stats like Score-Adjusted Fenwick, Corsi/Fenwick Tied, etc. exist - to limit score effects.



Against most teams, 3 quality shots. Against Toskala or Dubnyk, I'll take 20 easy shots.

But like the old cliché goes, good things happen when you throw the puck on net. If you could choose between 3 quality shots or 3 quality shots plus 17 easy shots, which one would you pick? Gaining control of the puck and putting it on goal is generally a good thing, even if it's not everything. Even Carlyle recognizes this - in his post-game interview last night he said "we're going to get our butts kicked most nights if we play like this".



The year that Carlyle's Ducks won the Cup, we don't have Corsi or Fenwick stats available, but they did outshoot their opponents heavily. The other time they had a good playoff run (2009), they had a 50.1% Fenwick Close and 50.9% score-adjusted Fenwick. The one time the Ducks made the playoffs with a low Fenwick (2011), they needed a brilliant performance from Hiller, and were knocked out by Nashville in the first round with Hiller sidelined with vertigo. So generally, even with Carlyle the pattern is that good things happen when you outshoot your opponents in the long run.



Corsi and Fenwick measure the effectiveness of a team's skaters at even strength. It does not cover goaltending or special teams (and Corsi doesn't even consider shot blocking a skill). Also, shootouts are ignored.

The 2009-10 Leafs were dead last in the league in both PP and PK, and had Toskala/Gustavsson in net to boot. So the reason they could come in 29th when they had a high Corsi is because they were downright awful at all the aspects of the game not covered by Corsi.

It's fairly well known that having great goaltending can "cheat" Corsi - see 2010 Habs, 2011 Bruins, 2012 Predators.



Shots definitely come in different qualities, and most statistical analysts have identified that:

A) shot quality is higher when a team is on the power play, because there are more cross-ice passes and back-door plays that make life difficult for goalies.
B) teams take better-quality shots when they are ahead (because the other team is taking more chances) and lower-quality shots when they are behind (because the other team is retreating into a defensive shell).

Analysts generally have decided to consider special teams a completely separate aspect of the game (limiting Corsi and Fenwick to measurements of even-strength play) to account for A, and have developed formulas to take score situations into account to account for B (for an example see here: http://www.broadstreethockey.com/2012/1/23/2722089/score-adjusted-fenwick).

But aside from those two relatively simple factors, shot quality is just considered too subjective to be factored into calculations. So essentially, we're just going to have to live with an imperfect system. Current Fenwick stats still give a pretty good picture of what is going on for most teams, despite the imperfections. (For example, the Hawks and Bruins showed up near the top of most shot-based stats last season, and the Sharks are currently #1 in Fenwick).



It's just difficult to figure out how "good" a shot is without having game tape in front of you. The only thing you can really do is work with shot locations, and even that is pretty tedious. So for the most part, people just ignore shot quality and accept that the system isn't perfect.

In the case of the Leafs, the blog Pension Plan Puppets actually did a series of posts over the off-season which showed that the Leafs weren't particularly good at keeping shots out of prime scoring areas last season. So in other words, even accounting for shot location wouldn't change the fact that the Leafs were getting outplayed on a regular basis.

You make a lot of good points, but I'd just like to address the bolded part. I agree that it is an imperfect but generally good system, but a lot of those that criticize Toronto's record doesn't seem to allow for some doubt due to the imperfection.

I think it's likely that Toronto will come back to earth if they keep this up, but there is also the chance that the imperfection makes Toronto specifically hard to judge with the tools we have.

 

[Master_Of_Districts]

That doesn't indicate anything.

No - you just don't understand what it indicates.

You see, you asserted that "[g]oing back to Carlyle's time with the Ducks they were similar as well." Meaning that the Carlyle coached Ducks were similar to the Carlyle coached Leafs.

Which is to say, both teams experienced success despite struggling with possession at even strength.

[Of course, I use the term 'success' in the most relative way imaginable in Toronto's case.]

In any event, the data directly contradicts your assertion. The Carlyle coached Ducks made the playoffs every year in which they had an above-average Fenwick/Shot percentage at even strength/5-on-5 (05-06, 06-07, 07-08 and 08-09).

In 2009-10, they posted a poor Fenwick and missed the playoffs.

In 2010-11, they made the playoffs while again having a poor Fenwick number. However, they were outscored by 20 goals when 5-on-5. They only made the playoffs by virtue of their power play. And even then, just barely, as their goal differential as a whole was also weak, compared to other playoff teams in the West.

In 2011-12, they continued to suck at even strength for the first quarter of the season (both in terms of possession and results), culminating in Carlyle losing his job.

So - once again - your assertion was wrong. The Carlyle coached Ducks were nothing like the Carlyle coached Leafs within the context of this discussion.

 

[JoelWarlord]

The thing that confuses me is that the NHL tracks things like hits, blocks, giveaways, and takeaways (which are quite subjective)

But they don't track things like
- Penalties Drawn
- Zone Entries
- Shot Differential (take away the fancy names, call it what it essentially is, dumb it down)
- Time of Possession

This stuff doesn't have to be advanced, but really the NHL is dropping the ball. Its left to 3rd party websites/bloggers to track this information which makes it easy to write-off as "Advanced Stats" when its really not that advanced.

Its kind of weird that the guys at TSN will reference Capgeek, but not BehindTheNet

Kadri was top 5 in the league in penalties drawn last year IIRC, I'm surprised that's not one that the NHL tracks more closely. You'd think that'd be more openly recorded, as it's certainly a part of what makes Kadri a great player.

As for time of possession, it doesn't really seem to be a meaningful stat because it's already kept track of by Corsi and Fenwick. PPP did a brief study last season and they found that time on attack correlates very closely with Corsi. The logic behind Corsi is that if you're shooting on net, you're driving the play, which correlates strongly with time on attack stats, and time of possession as a total doesn't mean a lot because it's measuring meaningless possession time (holding the puck behind your net waiting for a line change, etc). It'd be interesting to see though, perhaps with a larger sample size there would be more to learn from it.

http://www.pensionplanpuppets.com/2013/9/16/4727746/leafs-attack-time-at-the-halfway-mark

 

[Freudian]

Kadri was top 5 in the league in penalties drawn last year IIRC, I'm surprised that's not one that the NHL tracks more closely. You'd think that'd be more openly recorded, as it's certainly a part of what makes Kadri a great player.

As for time of possession, it doesn't really seem to be a meaningful stat because it's already kept track of by Corsi and Fenwick. PPP did a brief study last season and they found that time on attack correlates very closely with Corsi. The logic behind Corsi is that if you're shooting on net, you're driving the play, which correlates strongly with time on attack stats, and time of possession as a total doesn't mean a lot because it's measuring meaningless possession time (holding the puck behind your net waiting for a line change, etc). It'd be interesting to see though, perhaps with a larger sample size there would be more to learn from it.

http://www.pensionplanpuppets.com/2013/9/16/4727746/leafs-attack-time-at-the-halfway-mark

The problem with all these things that corsi/fenwick supposedly correlate closely with is that if you add a bunch of things with 0.8 correlation together you end up with something that's not all that reliable.

Corsi/fenwick correlate decently with shot quality/scoring chances but it's basically assumed it's a very good correlation. Same with corsi/fenwick and it's correlation with attack time/possession. The proof of this correlation was based on parts of 24 games (only using 5on5 tied situations). That's an insignificant sample and any correlation might be down to just randomness, something that the graph seems to indicate might very well be the case.

 

[Tyler Biggs*]

Everyone is talking about the Leafs corsi numbers and getting outshot. Maybe so but I remember when Stajan, Antropov and Blake were here and the team outshot everyone else and still lost! If they can offset the shot count with their PP and PK they should ride it out. Not to discount advanced stats.....but the more important stat is wins/losses. GO BUDS GO!

Everyone is saying 'Wait till the Leafs play a top team' Fun fact, including playoffs last year, the Leafs were 10-10-2 against Pittsburg, Boston, Montreal and Washington...Which for those keeping track were the top 4 ranked teams in the East last year...Which means the Leafs were above 500 against teams better than them in the Standings....They may lose against those 'top teams', but they might win some too!

 

[Ho Borvat]

Kadri was top 5 in the league in penalties drawn last year IIRC, I'm surprised that's not one that the NHL tracks more closely. You'd think that'd be more openly recorded, as it's certainly a part of what makes Kadri a great player.

As for time of possession, it doesn't really seem to be a meaningful stat because it's already kept track of by Corsi and Fenwick. PPP did a brief study last season and they found that time on attack correlates very closely with Corsi. The logic behind Corsi is that if you're shooting on net, you're driving the play, which correlates strongly with time on attack stats, and time of possession as a total doesn't mean a lot because it's measuring meaningless possession time (holding the puck behind your net waiting for a line change, etc). It'd be interesting to see though, perhaps with a larger sample size there would be more to learn from it.

http://www.pensionplanpuppets.com/2013/9/16/4727746/leafs-attack-time-at-the-halfway-mark

Assuming a team has a 20% PP, if a guy like Kadri theoretically drew a penalty every game, he would directly be responsible for an additional 20 goals over a season.

For the 2nd part, I was kind of more referring to a time on attack stat, kind of like the NHL video games do. Having the puck behind your own net really isn't advantageous, but having the puck inside the other teams zone is a threat.

Im just surprised that the NHL is willing to track things like giveaways/takeaways but not other things.

 

[Tyler Biggs*]

They don't give points in the NHL for a team playing to the best of its ability - they only give points for wins. I would trade 50 shots from the blue line against the boards for 10 shots in the slot. You might have less shots, but you would likely have more goals. Shots for and against are just another stat, and it does not mean as much as some people seem to believe. The reality? The Leafs have yet to play their best hockey this season and have 12 of 14 possible points - call it luck if you want, but great teams create their own luck.

No doubt this team needs to improve certain aspects of their game, but you can flip this argument around a couple of ways. First, look at the Leafs' opponents to date as a 'team' - they really need to be looking for answers... Dominant in possession and shots-on-goal but are 1-6 to start the season. Second, would it be a preferable situation to be 1-6 with questionable goaltending and special teams costing them games they should have won? Winning games you don't necessarily deserve to win is often cited a trademark of a great team - not that I'm suggesting this team is great, they're not...yet, but let's think about the big picture here. They score, have great special teams and goaltending. Build on that, don't tear it down.

 

[weems]

Every shot is not the same.
Until analytics people can accurately track and value every single shot taken advanced statistics need to be taken with a HUGE grain of salt. Even then just because a player gets a shot off in a certain area of the ice it doesnt automatically mean its identical to another one. Someone could curl out of the corner and and take a shot from the high slot with noone in front of the goalie = easy save. Then you could get someone come down on a 2 on 1 and take a shot from the exact same spot (Josh Leivo tonight) but its a totally different play and a much harder stop for the goalie.

Advanced statistics are usefull when using them in unisen with what your eyes tell you. Theres just so many variables and intangibles in hockey that really arent seen in other sports that will always make this sport very dificult to accurately judge by analytics. I like some of them but then theres also some that are way too much. To me the Leafs don't look like a "lucky" team but moreso a evenly balanced team that is really good with its special teams and have a TON of individual gamebreakers.

 

[Tyler Biggs*]

Outshot does not mean outplayed.

 

[habsfanatics*]

When they miss the playoffs I'd imagine this thread will die. Their play is unsustainable

 

[overlords]

Outshot does not mean outplayed.

It doesn't necessarily, but the leafs have still been outplayed in a fair amount of their wins.

 

[The Mighty Duck Man]

I made a point about this and the Ducks last year. I like the data CORSI and Fenwick track, but to me it's not overly meaningful. A lot of it is just more indicative of style than anything. You look at the Kings, who always rank highly in these categories, yet really don't score a lot of goals. A lot of people say it's luck, but really it's just terrible shot selection. They shoot from everywhere and hope for the best, and more often than not, it doesn't work.

Does that mean any team with a high CORSI is a fraud, or any team with a low one gets an undeserving bad rap? No. However, you really need context with stats like this.

 

[Master_Of_Districts]

I made a point about this and the Ducks last year. I like the data CORSI and Fenwick track, but to me it's not overly meaningful. A lot of it is just more indicative of style than anything. You look at the Kings, who always rank highly in these categories, yet really don't score a lot of goals. A lot of people say it's luck, but really it's just terrible shot selection. They shoot from everywhere and hope for the best, and more often than not, it doesn't work.

Does that mean any team with a high CORSI is a fraud, or any team with a low one gets an undeserving bad rap? No. However, you really need context with stats like this.

Making bald assertions is easy.

It becomes problematic, however, when others care enough to test those assertions against the evidence.

The Kings had a Corsi percentage of 56% during 5-on-5 play in 2012-13.

http://stats.hockeyanalysis.com/teamstats.php?disp=1&db=201213&sit=5v5&sort=FFPCT&sortdir=DESC

A blogger also took the time to record scoring chances for the Kings during the 2012-13 season. His data discloses that the Kings earned 56% of the scoring chances at even strength.

http://www.jewelsfromthecrown.com/2...g-chances-part-3-goaltending-and-team-numbers

In other words, there's no evidence that the Kings employed any kind of tradeoff between shot quantity and quality, like you're suggesting. The Kings' scoring chance percentage was identical to their corsi percentage.

 

[The Mighty Duck Man]

Making bald assertions is easy.

It becomes problematic, however, when others care enough to test those assertions against the evidence.

The Kings had a Corsi percentage of 56% during 5-on-5 play in 2012-13.

http://stats.hockeyanalysis.com/teamstats.php?disp=1&db=201213&sit=5v5&sort=FFPCT&sortdir=DESC

A blogger also took the time to record scoring chances for the Kings during the 2012-13 season. His data discloses that the Kings earned 56% of the scoring chances at even strength.

http://www.jewelsfromthecrown.com/2...g-chances-part-3-goaltending-and-team-numbers

In other words, there's no evidence that the Kings employed any kind of tradeoff between shot quantity and quality, like you're suggesting. The Kings' scoring chance percentage was identical to their corsi percentage.

Yes, let's just mark out an area in the offensive zone, and every shot attempt from there is a scoring chance. This is the exact same flaw as CORSI, that all shots are assumed equal. In this instance, it's that all shots within those boundaries are created equal. Again, you need context.

I'm not here to say CORSI and Fenwick are complete junk and you're all idiots, but context is needed to back up the statements of those stats. There will be exceptions, on both sides.

 

[Master_Of_Districts]

Yes, let's just mark out an area in the offensive zone, and every shot attempt from there is a scoring chance. This is the exact same flaw as CORSI, that all shots are assumed equal. In this instance, it's that all shots within those boundaries are created equal. Again, you need context.

I'm not here to say CORSI and Fenwick are complete junk and you're all idiots, but context is needed to back up the statements of those stats. There will be exceptions, on both sides.

Ah - but you claimed, in reference to the Kings specifically - and I'm quoting directly here - "[t]hey shoot from everywhere and hope for the best, and more often than not, it doesn't work."

Surely a team that employed such a strategy would be found to take a higher proportion of its shots from the non-dangerous areas of the offensive zone versus the more dangerous areas (i.e. the area of the ice in which shots taken are generally counted as scoring chances).

 

[Hammer Time]

The biggest issue with Corsi or Fenwick is that in small sample sizes, score effects often make a big impact. I don't think the Ducks were actually nearly as bad as their Fenwick Close of 48% made it seem last year - because of their high Sh% they were in the lead more often than not and conceded more possession as they retreated into a defensive shell (Anaheim was actually a respectable 50% in score-tied Fenwick).

The classic example is probably the 2011 Cup Final. Boston took 45% of the non-blocked shot attempts on goal (BOS 277, VAN 334) but had 54% of the scoring chances (BOS 103, VAN 87 as tracked by Derek Zona at The Copper&Blue).

(In addition, with Bruins goalies consistently posting high sv%s, I think Boston might actually be able to consistently suppress shot quality, although I'm not sure if the evidence shows this).

 

[Tyler Biggs*]

'Shots on net' is a statistic to give the loser something to brag about.

 

[Bear of Bad News]

'Shots on net' is a statistic to give the loser something to brag about.

Are you here to instigate, or to contribute meaningfully?

If it's the latter, then OBVIOUSLY the only meaningful results from a game are goals scored and goals allowed (and even then, the only thing that matters is whether goals scored is greater than goals allowed - you don't even need the numbers themselves).

However, the question is whether or not other totals (such as shots on goal totals) predict future performance better than goals scored and goals allowed.

You seem to be attempting to distract from that task. Why?

 

[GuineaPig]

Because being outshot 40-20 had nothing to do with the Leafs' loss yesterday, obviously.

 

[TieClark]

Because being outshot 40-20 had nothing to do with the Leafs' loss yesterday, obviously.

Not really. The Leafs are 6-3 while being outshot I believe every single game this year. The Leafs were lucky to only lose 3-1 last night but that's because they didn't show up to play. Quality of shots is far, far more important than quantity of shots and the shooter has significant implications as well. Colton Orr in the slot isn't nearly as dangerous as Phil Kessel in the slot.

 

[The Mighty Duck Man]

Ah - but you claimed, in reference to the Kings specifically - and I'm quoting directly here - "[t]hey shoot from everywhere and hope for the best, and more often than not, it doesn't work."

Surely a team that employed such a strategy would be found to take a higher proportion of its shots from the non-dangerous areas of the offensive zone versus the more dangerous areas (i.e. the area of the ice in which shots taken are generally counted as scoring chances).

I'm actually surprised and a bit suspicious of those numbers, just given how many point shots they take(which are actually some of their better chances), but from what I've watched, even in an area described as a high opportunity area, they take bad shots.

Then again, of the full games of theirs I watched last year, many were against the Ducks, who did a good job last season of baiting teams into taking bad chances, and of course there's a good chance my memory is failing me somewhat(although I know I said pretty much the same things all last year). But, I definitely remember most of the time when I saw them I couldn't help but think how they don't generate quality chances much at all.

Anyways, the scoring chance project is ambitious, but fairly dumb. When it comes to shot quality, until there's better tracking data, it's just one thing you have to accept isn't ready to be accurately recorded yet. Something like the scoring chance project is short-sighted and is something I suspect that would bring out a lot of complacency amongst the analytics crowd.

 

[Beef Invictus]

The thing that confuses me is that the NHL tracks things like hits, blocks, giveaways, and takeaways (which are quite subjective)

But they don't track things like
- Penalties Drawn
- Zone Entries
- Shot Differential (take away the fancy names, call it what it essentially is, dumb it down)
- Time of Possession

This stuff doesn't have to be advanced, but really the NHL is dropping the ball. Its left to 3rd party websites/bloggers to track this information which makes it easy to write-off as "Advanced Stats" when its really not that advanced.

Its kind of weird that the guys at TSN will reference Capgeek, but not BehindTheNet

The media and teams track "scoring chances" on their own, but there is no "official" stat. That bothers me, as well as the lack of stats you listed.

I would like pass completion % as well; dividing by zone would be nice. Giveaways/takeaways by zone would be fantastic, too.

 

[Ho Borvat]

The media and teams track "scoring chances" on their own, but there is no "official" stat. That bothers me, as well as the lack of stats you listed.

I would like pass completion % as well; dividing by zone would be nice. Giveaways/takeaways by zone would be fantastic, too.

I kind of find it funny that the NHL Video Game tracks passing %, and time on attack yet the real life NHL doesn't...

 

[Feed Me A Stray Cat]

Here's where Leafs fans are missing the point: NO TEAM has been able to sustain the Leafs' shooting percentage from last season in modern history. You can hem and haw about how not all shots are created equal, however unless you think the Leafs are the best team the NHL has seen at creating scoring chances in a long time (hint: they're probably not), expect their shooting percentage to go down to more reasonable levels.

For some comparison, the Pittsburgh Penguins have consistently been the team with the best shooting percentage in the NHL (makes sense with Crosby and Malkin up front), and even they're at like a 9.5% average over the last five seasons.