Marek Zidlicky, Jhonas Enroth, and a Corsi Event. Not pictured: a high-quality scoring chance. (Jim McIsaac, Getty Images)

Marek Zidlicky, Jhonas Enroth, and a Corsi Event. Not pictured: a high-quality scoring chance. (Jim McIsaac, Getty Images)

Last week, Cam Charron wrote about the NHL’s counting problem here at Backhand Shelf, bemoaning the secrecy of NHL teams when it comes to advanced statistics. One part in particular, however, caught my eye when he talked about hockey’s “Aha!” moment when it comes to statistics.

There’s a reference in the Friedman piece to Craig MacTavish walking around looking for the “Aha!” moment when it comes to hockey analytics. I don’t think MacTavish has realized that half the hockey world is a step ahead of him in that regard. The “Aha!” moment comes when you realize that shots are a hell of a lot more predictive than goals for determining future events. As soon as you realize that hockey is a game between two teams trying to take shots on goal, I think the rest of it falls into place.

Cam isn’t really wrong, but this is also one of the biggest problems that people seem to have with so-called advanced statistics: they’re almost entirely reliant on counting shots. Corsi and Fenwick are both shot-based statistics that are pretty much the opposite of “advanced.” All they are is adding and subtracting shots. The more shots for your team and the fewer shots against, the better. Outshoot your opponent enough, particularly at the right time of the game (such as when the score is tied or within one goal), and you’ll win a lot more games than you lose.

If this seems like an incredibly simplistic view of hockey, that’s because it is. It’s also a completely inaccurate view of hockey. That isn’t to say that Corsi and Fenwick aren’t useful, because they certainly are. As Cam points out, shot-based analytics have impressive predictive power. But they also are coming at hockey from the completely wrong end.

I believe this is part of the reason why so many people are resistant to shot-based statistics. What matters is winning, winning requires goals, and a high volume of shots does not, strictly speaking, create goals. Shots are a by-product and not a cause.

When we talk about Corsi, we’re talking about a statistic that acts as a proxy for puck possession. Corsi is essentially a plus/minus statistic that includes all shots on goal, missed shots and blocked shots, using the simple logic that if your team is taking a shot, then they have the puck. The more shots of all types that your team takes, the better. Fenwick is essentially the same, but removes blocked shots from the equation.

These two statistics have proven to be very useful for analyzing teams more accurately. A team’s Fenwick percentage when the score is close or tied, for instance, is generally far more predictive of a team’s future results than their current results. The Los Angeles Kings seemed to surprise everyone but the advanced statistics crowd with their dominance in the playoffs: they had the fourth highest Fenwick Close in the league. Then there was the well-predicted slide by the Minnesota Wild from top of the Western Conference to out of the playoffs.

But there’s still something about saying that more shots is better that rankles. When you watch a game, low percentage shots with no chance of going in certainly don’t look like they’re helping a team win. A defenceman that constantly misses the net is infuriating. I’ve even had discussions with people who suggest that players could start gaming the system if they knew their coach or GM used Corsi by intentionally taking more low-quality shots instead of looking for a better scoring chance. When you watch a hockey game, you can see that not all shots are equal, but Corsi and Fenwick treat them as if they are.

You don’t go to a hockey game to see your team take a lot of shots; you go to see your team score goals.

So, does a higher volume of shots lead to more goals? Sort of. Shots ride a fine line between correlation and causation. After all, as the Great One said, “You miss 100% of the shots you don’t take.” You can’t score if you don’t shoot. But a missed or blocked shot, by definition, cannot directly cause a goal, though they may lead to one indirectly.

All of those shots, whether on goal, missed, or blocked, are a by-product of what a team is actually attempting to do. The result a team is looking for is goals and, to that end, a coach puts in place systems, tactics, and game plans in order to prevent scoring chances for their opponents and create scoring chances for themselves that will produce that result. The players then put those into action and, as a by-product, produce shots on goal, missed shots, and blocked shots.

We count shots not because they’re the result of the process, but because they’re the most numerous by-product of the process.

There simply aren’t enough actual goals scored to create a large enough sample size to analyse. Fortunately, shots end up correlating nicely with goals over time. Fenwick correlates incredibly closely to scoring chances, but this isn’t because shooting the puck a lot is the same as creating a scoring chance. It’s just that the same processes that aim to produce shot quality will also produce shot quantity.

A forward might take a low-percentage shot from the outside, hoping for a rebound that will create a better scoring chance. A defenceman will sometimes throw the puck at the net while under pressure simply to keep the play alive. A fourth liner will shoot the puck as soon as he crosses the opponent’s blue line in order to create an offensive zone faceoff for his team’s top line. A top-six forward will optimistically shoot from a bad angle at the end of his shift after not being able to create a better scoring chance. The pursuit of shot quality will inevitably lead to shot quantity.

It’s not enough to simply outshoot your opponent: every NHL team knows this. It’s why you’ll never see players trying to game their Corsi by taking low-quality chances. What matters is goals and you score goals by doing things that hockey traditionalists love — making a good first pass out of the defensive zone, winning puck battles, creating turnovers in the neutral zone, going hard to the net, cycling the puck down low, etc.

All of the little things that lead to goals and wins over the long-term, lead to shots in the short-term. It’s not that shots matter more than goals — they absolutely don’t — but the things that make a team good at scoring and preventing goals also happen to make them good at taking and preventing shots.

As a by-product of trying to score goals, you’ll produce all sorts of shots — on goal, missed, and blocked. And we’ll add them, subtract them, and use them to analyse how well you did in the aggregate at all those things hockey traditionalists love.

Comments (39)

  1. This is, hands down, one of the best articles I have ever read on Advanced Stats. it should almost be required reading for anyone who argues against the merits of Advanced Stats in hockey.

    As someone who has resisted pure reliance on Adv. Metrics (for the reasons you’ve stated above), you have defintely put the whole thing in perspecctive.

    With all sincerity, thank you for this article Daniel.

    • Also required reading for those who argue FOR the merits of Advanced Stats.

      They’re vague approximations of scoring plays, at best, over the long run.

  2. Couldn’t Corsi or Fenwick be tweaked to account for where a shot originates? You could assign factors to origin points so that, for example, a shot from the slot has a factor of 1 (or whatever) while a shot from the boards has a factor of 0.25? You’d have to know shot percentages from those origins in order to assign accurate factors, but once you calculated those (if that’s possible), then you might end up with a more reflective stat.

  3. Thanks, like Derrick said, this article was phenomenal. As a fairly nerdy, numbers orientated hockey fan I really would have expected myself to take to advanced hockey stats quite easily. However, I’ve always had issues with it… until now. This line really puts it in perspective for me:

    “We count shots not because they’re the result of the process, but because they’re the most numerous by-product of the process.”

    As long as teams aren’t being coached to out-Corsi the other team, these stats do a good job of describing how well they’re going about trying to out score the other team.

    It does make me wonder that, if we had reliable posession statistic, if they would paint a more accurate picture. You do have to consider the case where a team is strong posessionally but elects to be more judicious with which shots to take. That would artifically deflate the team’s advanced statistics despite still driving posession.

    Zone entry stats are intruiging too but they’re yet to become prevelant enough for me to really see much about them.

    • the puck could be programmed to interact with players’ sticks, which also have to be programmed with unique identifiers. the results of puck possession could be transmitted
      wirelessly to a database.

      just thinking out loud, but it has many other possibilities besides puck possession.

      John DiToro

  4. Got to agree with Derrick, that was an awesome read.

  5. Thanks, everyone, I really appreciate the positive feedback.

    • My takeaway from this: Advanced stats tell you if a team/line/player is performing well, but not why.

      The lack of the “why,” is probably what keeps many people from embracing the advanced stats. That, and there’s also an elitism, whether actual or just perceived, among the advanced stats proponents (the Staterati) that is off-putting to many.

  6. Fascinating article. I do like and appreciate what advanced stats do for our understanding of the game, but I can see how sometimes they don’t tell the complete story.

    This article reminded me a little bit of this one, entitled, “The suicide shot,” from 2005 by a Ranger fan blogger: http://hockeyrodent.com/R1410.HTM

  7. Corsi this, Fenwick that. People need to learn about PDO.

  8. “I’ve even had discussions with people who suggest that players could start gaming the system if they knew their coach or GM used Corsi by intentionally taking more low-quality shots instead of looking for a better scoring chance.”

    Exhibit A

    http://www.behindthenet.ca/nhl_statistics.php?c=0+1+3+5+4+6+7+8+13+14+29+30+32+33+34+45+46+63+67&ds=1&f3=TYLERKENNEDY&f1=2012_s+2011_s+2011_p+2010_s+2010_p+2009_s+2009_p+2008_s+2008_p+2007_s+2007_p

    Of course, the idea behind corsi and fenwick isn’t that shots mean your team’s good, it’s that shots come with the prerequisite that your team has the puck, which, in turn, means your team is good.

    But you’re right about not all shots being created equal. Advanced stats types love to blindly decree that shot quality doesn’t exist on aggregate, despite battle-tested coaching strategies that are designed almost exclusively to manipulate shot quality.

    The idea of giving up a lot of shots but keeping them to the outside and capitalizing on high percentage counterattacks is the highest percentage strat that can be deployed against a superior possession team. The examples of this strategy having success are countless over the last several decades.

    Anyways, excellent column.

    • What you say is true if you are correct about these players. What is your evidence that Glencross and Kennedy take lower quality shots than the average player?

      • That they hit the goaltender in the crest. Kennedy, in particular, has more shot attempts per minute than any player since this started being tracked. If they were good shots, it stands to reason a lot more of them would go in.

  9. This is a really interesting companion piece to this article about the declining usefulness of Fenwick and Corsi as a predictive tool over time: http://hockeyanalysis.com/2013/02/27/the-declining-value-of-fenwickcorsi-with-increased-sample-size/

    All these new stats are very useful and insightful. But just like the “old fashioned” stats, they are also limited because they measure exactly what they measure, and when we start trying to use that to extrapolate other meanings or predict the future, we run into error. Like everything else, they should be considered as a part of a larger whole of a hockey game/season/career.

  10. I’m sorry to break this into pieces and dissect, but I’m not sure how else to approach my different thoughts on this piece.

    First, I think you’re taking Cam’s quotes out of context. He was referencing a remark made regarding player evaluation and advanced statistics from a sports analytics conference. For the purposes of what MacTavish was talking about, Cam is pretty well on the money.

    You write that “If this seems like an incredibly simplistic view of hockey, that’s because it is.” Isn’t that the basic concept behind statistical analysis? Getting things down to one number or a few numbers in order to make talent evaluation as easy and clear as possible? The more variables you add to any situation, the more complicated it becomes.

    “I believe this is part of the reason why so many people are resistant to shot-based statistics.” I don’t know. It always seemed to me that people who derided or disliked advanced metrics (I too find it funny that we refer to these as ‘advanced’) were those who a) didn’t care enough to take the time to learn what things like corsi/fenwick can tell us or b) don’t understand the basic value of statistical analysis.

    You say “When we talk about Corsi, we’re talking about a statistic that acts as a proxy for puck possession” and that’s obviously true. But then you say “What matters is goals and you score goals by doing things that hockey traditionalists love — making a good first pass out of the defensive zone, winning puck battles, creating turnovers in the neutral zone, going hard to the net, cycling the puck down low, etc.” Corsi/fenwick are proxies for this stuff too. All of the things you describe result in shots for or against. So if the goal is to put something into terms that most people can grasp, then I think using corsi as a proxy for those different types of events is a GOOD thing.

    I guess what I’m seeing in this piece is you saying that looking at hockey the way Cam described it and looking at hockey in the “traditional sense” are mutually exclusive. And I’ve never viewed it that way. I’ve played hockey myself for 20+ years and I can certainly appreciate the “traditional things” you describe. But I also find a ton of value in stepping back and looking at things from a numbers perspective. And there’s nothing wrong or backwards about it.

  11. I have a couple of serious issues with this article. First is the inaccurate premise that the entire article is based on. Somehow the author thinks “a team can artificially increase their overall Corsi/Fenwick numbers by taking poorer quality shots.” That is simply NOT true.

    In fact, one of the counter-intuitive things we have found is that the “ratio of quality shots to overall shots” is self-leveling.

    You would think better teams would have a higher ratio of quality shots to overall shots. However, better teams tend to be the ones that can get the puck back after a shot and/or enter the offensive zone easier. So they tend to be more willing to give up the puck by taking a shot, even lower quality shots. This tends to counterbalance their superior ability to generate quality chances. So higher Corsi/Fenwick numbers instead of a better ratio of quality to quantity is the result.

    And you can’t artificially impact Corsi/Fenwick by shot selection because a poorer quality shot is LESS likely to be “retained” by the shooting team. So a poor shot will generate less overall chances not more. I mean the “light bulb” should be going on here. If a shot could generate more shots it wouldn’t be considered a low quality shot to begin with.

    And my second issue is that this article acts like a team that uses Corsi would somehow be blind to all other stats. Even if a player COULD increase his Corsi/Fenwick numbers by reducing his shot quality, would he? High Corsi guys with low shooting percentages are on the 4th line in this league. Take a look at Michael Frolik of the Chicago Blackhawks if you want an example.

    Its true Corsi shows the results of what we really want to track but so what. We don’t have “possession time” so we use Corsi as an approximation. As long as you understand that then Corsi and Fenwick numbers are pretty useful information.

  12. “Somehow the author thinks “a team can artificially increase their overall Corsi/Fenwick numbers by taking poorer quality shots.” That is simply NOT true.

    In fact, one of the counter-intuitive things we have found is that the “ratio of quality shots to overall shots” is self-leveling. ”

    “We.”

    Anyways, advanced stats types like to say what you’ve said, but it’s demonstrably wrong the second anybody takes the time to plot out a heat map. And players like Tyler Kennedy and Curtis Glencross post outsized “possession metrics” compared to the amount of possession they drive for the simple reason that they take a lot of bad shots from harmless areas. Lots of shots from the wall do not correlate well with winning at all.

    If a guy, or even a bunch of guys on a team do that, corsi and fenwick are no longer proxies for possession. The two stats only work as proxies for possession as long as the underlying assumption that the two teams have identical or similar goals of what to do with the puck on every shift.

    Guy like Tyler Kennedy vs. a guy like Jordan Eberle…team like the Red Wings vs a team like the Rangers…corsi doesn’t tell you anything about them because they’re following completely different strategies.

    • How does taking a “bad shot” increase a player’s Corsi? Taking a shot loses possession. Corsi measures the ability to get the puck back. How does taking a poor shot improve ones chances of getting the puck back? It doesn’t. If anything it makes it harder.

      And “scoring chances” was a big deal in 2009/10. Bloggers were measuring what were considered good shots compared to overall shots. So people were making your heat maps back then looking at the ratio of quality to quantity. And people have mostly abandoned the idea. And the reason they have given for why is what I just stated.

      • Corsi literally measures shots, as I’m sure you’re aware. I think it’s a reach to say it’s measuring puck recovery. You don’t just get to hand-wave that one away.

        • It is not a reach at all. Corsi measures the result of puck recovery. Every Corsi event gives up the puck. So you can’t get another Corsi event until you recover the puck. So teams that win Corsi are the teams that win the puck recovery and puck possession battle.

          • The only shot that can not possibly be recovered is one that goes onto the net.

            the problem have with Corsi and Fenwick numbers are that a team that breaks out of their zone cleanly ,makes two passes , takes one shot and scores, show “worse” stats than a team that takes 5 shots, all missing the net ,or stopped by the goalie.

            It seems to me that there needs to be a “weighting” applied, that is a goal scored must be counted as something more than equal to a shot attempt.

            regards,

            Steve O.

          • You’re aware you can take a shot, not recover it, and still win the shot battle on a shift, right?

    • “The two stats only work as proxies for possession as long as the underlying assumption that the two teams have identical or similar goals of what to do with the puck on every shift.”

      They do have identical goals. Both want to put the puck in the other team’s net and keep it out of their own. There are lots of ways to accomplish that, but you can’t do it without maintaining puck possession.

      The stats guys have put a lot of quantifiable work into showing that Corsi correlates will with puck possession. If you want to disprove that, you’re going to have to do a lot more than suggest hypothetical counter-examples.

      • I think both of you need to watch a rebroadcast of the Penguins Bruins game that’s going on as I’m typing this (watch the Bruins lose though).

        As I’m typing this, the Penguins have doubled the Bruins shot totals and surely more than doubled their shot attempts. But the Bruins are completely in control of this game.

        How is this possible?

        Simple. The Bruins’ strategy is to allow a high number of awful shots as a trade-off for keeping Penguins’ shooters out of high percentage areas, while taking their chances with a handful of point blank shots on the counterattack (the Bs scored one goal from the paint and another due to a screen in the paint).

        Corsi will tell you luck decided the first 2 periods of this game. It didn’t. Julien successfully manipulating shooting percentage did.

        • Yeah, even if I were to accept a single game as a sufficient counter-example, you’d want to choose a better one. The Bruins were completely in control of a game in which they scored a single even-strength goal?

          Corsi doesn’t count shots on special teams, btw. But if you want another proxy for possession and “control”, note that the Bruins also drew a single penalty all game (the other was for too many men).

        • Let me get this straight. Your counter argument is to show one game? A game that you think counteracts what we are talking about using Boston as your example? Boston who got outshot? Boston who gave up around 14 shots in the house compared to 5. Boston who gave up three goals all on quality shots and Boston who LOST that game. They lost the quantity and quality battle in that game. Pretty funny example.

          And the concept of the ratio of shot quality to shot quantity leveling out is over a season, NOT just one game. Again provide some real data, otherwise you are just spouting an opinion. And remember, people spent a lot of time documenting quality chances over the last few years. They wouldn’t have abandoned that work if it had supported your position.

  13. Neal’s got it right. A common misunderstanding of Corsi skeptics is that it implies that every shot on goal is a good play. This isn’t true. You can’t “game the system” by taking lousy shots, because you are giving up possession and the other team is going to skate down the ice and take several shots.

    The significance of Corsi is that over a large number of plays, a team that is better at hockey – at ALL the little things, tangible and intangible, that help a team win – will end up with more shots on goal than their opponents. Measuring Corsi has more predictive value than counting goals, because goals are infrequent and require luck: off the post and in is millimetres away from off the post and out.

  14. @ larry dallas

    Kennedy and Glencross take a lot of bad shots from harmless areas? Says who?

    Not the NHL, who tracks shot locations:

    Kennedy: http://somekindofninja.com/nhl/index.php?season=Regular&year=2011-2012&shots=For&team=Pittsburgh+Penguins&ice_player_name=Tyler+Kennedy&withPlayer=On+Ice&player_name=&goalie_name=&event=Shots+and+Goals&game=Home+and+Away&strength=Even&time=Regulation&search=Search

    Glencross: http://somekindofninja.com/nhl/index.php?season=Regular&year=2011-2012&shots=For&team=Calgary+Flames&ice_player_name=Curtis+Glencross&withPlayer=On+Ice&player_name=&goalie_name=&event=Shots+and+Goals&game=Home+and+Away&strength=Even&time=Regulation&search=Search

    Here’s Claude Giroux’s “heat map” for comparison: http://somekindofninja.com/nhl/index.php?season=Regular&year=2011-2012&shots=For&team=Philadelphia+Flyers&ice_player_name=Claude+Giroux&withPlayer=On+Ice&player_name=&goalie_name=&event=Shots+and+Goals&game=Home+and+Away&strength=Even&time=Regulation&search=Search

    Players tend to shoot from all over the ice.

    Also, this blurb seems a little off: “If a guy, or even a bunch of guys on a team do that, corsi and fenwick are no longer proxies for possession. The two stats only work as proxies for possession as long as the underlying assumption that the two teams have identical or similar goals of what to do with the puck on every shift.”

    Corsi/fenwick are proxies for puck possession because the team has to have the puck in order to take a shot in the first place. It has little or nothing to do with where guys shoot the puck from because teams generally have the same goal of what to do with the puck: get a shot on net. Now you can argue if a team’s system is successful or not in generating/limiting shots, but the idea that two teams can’t be compared because they play different systems is ridiculous and unfounded.

    • Giroux’s not a goal-scorer, but I can certainly work with this.

      A far greater cluster of Claude Giroux’s shots are from below the dots and between the circles. Kennedy’s and Glencross’s are scattered all over the place, most of which are at considerably greater distance than the most common Giroux shot.

      Thanks for demonstrating for me that Kennedy and Glencross take ill-advised shots from everywhere and other high corsi players do not.

      • You said “Thanks for demonstrating for me that Kennedy and Glencross take ill-advised shots from everywhere and other high corsi players do not.”

        That is ridiculous. Glencross has a higher shooting % than Giroux. Remember that old stat? The one that tells you how many shots go in compared to how many you take. The shooting % shows that Glencross puts in more of hs shots than Giroux, so how could the shots be more ill-advised?

        Could it be that players are individuals with different skill sets and not machines you can accurately evaluate based solely on stats?

    • ” It has little or nothing to do with where guys shoot the puck from because teams generally have the same goal of what to do with the puck: get a shot on net.”

      Oh yeah. This isn’t correct. Not all coaches use possession systems. Many will happily trade a bazillion shots from low-percentage areas as long as their own (far fewer) shots are high percentage scoring chances and the other teams’ are not.

      • Larry, why don’t you do some actual work and show us just who these teams are. Quick tip though: don’t look at Claude Julien’s Bruins for an example of a team that trades a “bazillion” shots in exchange for high quality chances of their own.

  15. Just my two cents:
    I’ve always taken the approach that stats (simple or complex) should simply serve to confirm or disprove what we think we’re seeing.

    Statistical trends can be a starting point for speculation, but the inherent fluidity and dynamic nature of the game means that variables will almost always crop up and so a true “prediction” is nothing more than voodoo.

    PDO and Corsi are absolutely useful tools in player evaluations. That said, these metrics are far from the only method and a reliance soley based on them is foolish at best.

    It doesn’t have to be “traditionalist” vs “sabrmetrics”.
    Scouting and stats are complimentary and its about time we all go behind this line of thinking.

  16. The biggest problem right now in the use of new metrics is the inability to use them in their proper context or to explain WHY one is seeing the results that they are seeing. Same issues that often plaque baseball, in baseball the biggest mistake is understanding the proper sample sizes you need for each stat. Advantage of baseball is that their is so much data being tracked that you can often find reasoning behind the results quite easily. Hockey does not have that required data, and this is why we have these arguments. The hockey blogosphere is overwhelmed by guys who are providing “analysis” and scouting by numbers with out ever attempting or having the ability to explain why for example a certain player or team is deficient in a particular area. This combined with traditionalists refusal to even try to understand some new metrics even ones as simple and unbelievably important as zone starts, is causing usless bitterness. I go back to a point Keith Law once made about what he learned being in a front office. He said most important revelation was finding the balance between the numbers and the eyes, and using both together. Just not enough people in hockey community are doing that right now.

  17. is there a tweak that can be applied to count for when blocked or missed shots lead to the opponent scoring a goal? or is it even necessary? from a spectator’s point, it seems like these two things lead to a number or goals going the other way, but maybe crunching the numbers shows that it isn’t really a big deal as we fans perceive it.

  18. I love your blog.. very nice colors & theme. Did you make this website yourself or did you hire someone to do it for you?

    Plz respond as I’m looking to construct my own blog and
    would like to know where u got this from. thank you

Leave a Reply

Your email address will not be published. Required fields are marked *