Advanced Stats Crash Course: Possession (Part 1)

 

 

Although "new stats" are still considered the useless obsession of a small collection of dilettantes in some quarters, they are nevertheless growing in prominence and use across the game. Possession stats like corsi are the basis of this upheaval and I still get a lot of questions and requests for clarification when it comes to these sorts of metrics.

Instead of answer them individually via email I decided to put together a crash course of the most important measures. In this series, I’ll discuss what the measures are, what they mean, why they are useful and how they can easily referenced.

We’ll start with corsi.

What is it?

Corsi is named after Buffalo Sabres goalie coach Jim Corsi. It refers to all the shots directed on net (including blocked and missed shots). He created it as way to judge goalie performance.

In 2007, Vic Ferrari (Hockey’s version of Bill James) heard Corsi describe the measure during a radio interview and applied the measure, expressed as a differential (all shots for and against), to the Edmonton Oilers in this seminal article.

From there, corsi was adopted as a proxy for offensive zone puck possession at even strength (5on5 mostly). Later, when individuals started to count scoring chances, it was found that corsi strongly correlated to chance differential over time. Ferrari sums up the key initial insights in that final link:

Bottom Line: Players that drive possession at EV drive results at EV. Maybe not the next game the Oilers play, or the next dozen, but eventually. It’s unstoppable, ability trumps luck eventually, you just have to be patient.


Corsi begets scoring chances, scoring chances beget goals, and even over the course of a full season the hockey gods have a huge say in a player’s scoring (and outscoring) results, and a big say in the same results for the team.

And while corsi may not be a perfect measure of territorial advantage, it’s still terrific, moreso than any of us thought, I’m sure.

You can make a good estimate of a player’s scoring chance numbers by using his corsi. And I suspect that the difference is largely luck, scoring chances drifts toward the highly repeatable corsi.

Summary: corsi is all shots directed at the net for and against at even strength. It can be applied at the team level or individual skater level. Corsi stats are a proxy of offensive zone puck possession, meaning a high ratio or differential means a player or team is spending more time in the offensive zone. Conversely, a low ratio or negative differential means a team or player spends more time in the defensive zone.

Studies have shown that corsi highly correlates with scoring chance differential at even strength.

Why are Possession Stats Useful?

In simple terms, corsi stats tell you who is spending more or less time in the offensive zone. That also tends to tell you who has a good or bad scoring chance differential. Over time, outshooting (ie; good possession) correlates with outscoring.

In math terms, counting all of the shots on net gives possession stats more statistical "power" because they wash out a lot of randomness.

Most conventional analysis is goal based, which is superficially rational since scoring and preventing goals is the point of the game. The problem is, goals are relatively rare and relatively random events in hockey games, which makes them rather poor indicators of talent in small samples (a season or less). Not because goals aren’t important, but because of their rarity you need a lot of goals to separate the statistical noise (randomness) from the signal (talent).

There are 10-15 times more shots directed at the net in any given game than there are goals scored. Only about 8% of shots on net at 5on5 actually end up as goals and only about 15% of high quality shots (scoring chances) light the lamp.

To put it another way, there is much, much larger sample of shots than there is of goals at any given time, whether at a team or individual level. Goals are much more subject to swings due to randomness alone as a result. Possession rates, in contrast, are far more stable because the sample size is larger. Once corrected for moderating variables (quality of teammates, ice time and competition), corsi tends to be far more stable than other, more conventional player and team measures. Meaning it is repeatable ie; representative of talent. 

Summary: possession measures are more powerful because the number of shots vs goals is vastly larger. They are more stable across players and teams, particularly when certain factors are considered. They correlate with scoring chances and goal rates over time. They are therefore predictive of future success.

Practical Applications

Understanding that corsi = possession = chance differential is a huge step forward in analyzing and predicting outcomes in the NHL. Here is an excellent visual representation of the importance of possession and success in the NHL by Chris Boyle:

Chris explains:

I have charted every team’s fenwick close since 2007-2008. The rings of the graph represent each round of the Stanley Cup playoffs. The further the logo is away from the ring represents the distance from a playoff berth. I have also charted the percentage from .400 to .600. The further away from the .400 represents a stronger possession team. The ultimate on this index would be the 2008 Red Wings with a score of 59.39 located on the top portion of the Stanley Cup in the +.550 section. The 2008 Thrashers scrape the bottom of this index with a 41.23 and 28th position during the same season.

You can see the magic number of success is +.500. If you manage to crack this number you have a greater than 75% chance to qualify for the playoffs. If you break the +.550 mark you have a 25% probability of winning the Cup.

I’ll add to the discussion.

Since 2007-08 (not including last year), 75% (50/66) of teams with a possession ratio between 50-54.9% have made the playoffs, while 100% of clubs with a possession rate about 55% (8/8) have made the dance. On the other hand, just 31% of clubs with a corsi ratio below 50% made the playoffs over the same period (22/70), while no team below 44.9% made the dance. In very simple terms, then, every GM’s goal should be to build a +55% possession team (and to avoid building a sub-45% roster).

Controlling play is a big portion of success in the NHL. The last few seasons, the best teams in the league by this measure has been the LA Kings, Vancouver Canucks, Chicago Blackhawks, San Jose Sharks, Pittsburgh Penguins, Detroit Red Wings (though not this season) and the Boston Bruins. Recent entrants to the top of the field are the St. Louis Blues. If that looks like a list of the elite teams in the NHL, well…it is. While some clubs can experience spikes of success due to a spate of hot goaltending or an outburst of goal scoring, to be consistently successful in the NHL means to control the puck, spend more time in the offensive zone and out-chance the bad guys.

Having a great possession team means you aren’t wholly reliant on world class goaltending or unusually high shooting percentages to get by. Which is good, because those don’t tend to stick around forever (more on that in the future).

Summary: Take a look again at the graphic above and then glance at the current corsi ratios (CF%) for each team in the NHL. Given what you know, predict how each team will finish the year.

Next up – How to reference and use possession measures

  • I’ve found a great analogy that people can understand for the heading Why are possession stats useful is “possession stats change the focus from outcome orientation to process orientation” in terms of player evaluation.

    You always hear out of the sport psychology literature that coaches need to get their players or athletes to focus on the process not the outcome. Rather than worrying about the score or focusing on scoring a goal. Focus on the process, for eg. shooting the puck at the net more often and the outcome (goals) will come. Game broadcasters constantly refer to he’s grasping his stick too tight he just needs to get back to doing the simple things like just shooting the puck. Well in reality CORSI measures just this. It’s a measure of the process not the outcome. It’s an objective way of determining if a player is doing the right things in the processes of his game to get the results (outcome) down the road.

    Managers are always looking for a better way to predict success in the evaluation of their players. This really is that tool. It’s one statistical way to determine if a player is getting the process part of his game right and generally speaking in sports if you’re doing the fundamental things right it will lead to success in outcome.

    • ChinookArchYYC

      Yes, and the same goes for the well run businesses and good management practices. Good companies and managers measure events that can be controlled, while poor companies and managers attempt to “manage” outcomes.

      As an example, many sales managers point to young sales people and suggest they are not successful because the sales person can’t close properly (in hockey terms “they can’t bury the puck”). The manager than puts the sales person through a Closing course. Good companies concentrate in ensuring their sales professionals are seeing enough prospects and moving those prospects through a sales process (or getting the puck, keeping it and driving to the O Zone).

      • Well said. Business literature and sports literature are usually very similar with implications of course crossing over. It’s a big reason why businesses will hire ex athletes without any education or awareness of business sense. Because the processes of being a good athlete can translate to being a good employee.

  • Southern_Point

    This is why the Leafs suck. Their management has taken it upon themselves to try and disprove possession stats like Corsi and Fenwick. They made the playoffs last year largely due to the shortened season, they didn’t play enough games to regress fully towards the mean.

    In the off season they ran two of their best possession forwards in Macarthur and Grabovski out of town, in exchange for Bozak and Clarkson, and one of those contracts is an exceptionally terrible piece of business.

    This also why I think many are ambivalent at best towards former Leafs GM Brian Burke. He is a noted curmudgeon when it comes to possession stats, and his love for trucluence may move us farther away from that hallowed 55% number.

    • Southern_Point

      The Leafs are again in a playoff position. Most teams play a very similar style and system. The Leafs are trying to do something different/ creative a s they realize playing a similar style to Pittsburgh for example will not work in terma of defeating them. They don’t have the talent. Will their approach work? Probably not. But, you can’t fault them for thinking outside the box.

  • So let’s see how close this turns out according to CF% if it holds for the rest of the year.

    The Playoffs should look like:

    ATL: Boston Tampa Detroit
    Metro: PIT NJD NYR/Philly
    East wildCard: NYR/Philly & Ottawa

    CEN: CHI STL DAL/MIN
    PAC: LAK SJS VAN
    west wildcard: ANA & DAL/MIN

  • ChinookArchYYC

    Anyone that looks at the illustration above and dismisses Corsi as meaningless and unimportant will eventually be compared to believers of a flat earth.

    • Burnward

      True. However, we still have to remember that the game isn’t a math equation.

      CORSI is a great analytical tool, but not a science.

      Awesome stat to incorporate into our view of the game though.

    • EugeneV

      Context, context context.

      To win a Cup you play four best of 7’s.

      This doesn’t give much space for Corsi (averages) to even out.

      Oh and the Earth is flat. In some contexts.

      • Southern_Point

        A good team in the regular season is still a good team in the post season. So I’m not sure exactly what you mean.

        Yes over the course of a seven game series the outcome is more likely to rely on randomness and be far less predictable. But every year there is a minimum of 24 and a maximum of 56 games played in the first round of playoffs. As Kent has noted many times there is a significant relationship between positive possession stats and wins, while one series may be decided on a lucky bounce or unsustainable shooting percentages combined with unsustainable save percentages. Another series will be decided by the team which dictates possession.

        In the long term, the winners of more best of seven series provided there is a relationship between possession rates and wins- which has been demonstrated above- will be the team with better possession numbers, and teams with poorer possession numbers will win those series less.

      • ChinookArchYYC

        Corsi measures how effective players and teams are in puck possession over time it was never intended to predict who would win a 7 game series. KW provided a poker analogy last week which was very good in making the point. Put it this way, I may have a reasonable chance of winning 4 of 7 poker hands against Daniel Negreanu, but over an entire tournament my chances of winning dimmish (luck after all takes one only so far).

        Your assertion about context for a flat earth does not apply here. The earth can be perceived to be flat, but is nevertheless a sphere. Similarity, some perceived the Leafs to be a good team because they got to the playoffs in a short season, but the Leafs results continue to fail over the long run, just as Corsi results indicated.

  • Derzie

    Kent you explained Corsi well then put in a chart which focuses on Fenwick without any intro. Same thing? The chart makes sense (takes a while to digest at first) but does not refer to corsi.

    • Southern_Point

      Corsi numbers are all shots directed towards the net. This includes blocked shots.

      Fenwick removes block shots under the premise that a player shouldn’t be penalized for blocking a shot.

      I prefer corsi it provides for a larger sample size, and a blocked shot for you still means that your team didn’t have the puck while you were on the ice, but there are certainly reasonable arguments to use Fenwick.

      Basically both are interchangeable and measure roughly the same thing.

  • EugeneV

    I understand the whole “law of averages” thing, and can see how corsi shows up for the full season results.

    My issue with it is that it is the equivalent of watching the big bang theory in kurdish if I don’t speak Kurdish. I will miss the point.

    I’d like to see this chart for the playoffs or Finals etc… The law of averages doesn’t work when you reset everything every 7 games and have no second chance for the results to even out.

    Winning a Cup is about Beards after all.

  • Southern_Point

    One aspect to consider when using corsi and other possesion stats is the difference of hockey style between regular season and playoffs. The sedin sisters are effective in the regular season but fade in the playoffs. Using corsi numbers heavily weighted by regular season could get ya into trouble when the goal is hoisting the cup. Two cents.

  • Southern_Point

    I don’t think anyone is confused by how Corsi is measured. BUT, last year it was stated that Iginla needed to go because he failed to drive possession anymore. He is now one of the better right wingers in terms of Corsi. What has changed? He is simply on a better team with better players. Corsi isn’t something teams fret over. Good teams naturally have better numbers. Some poor players have inflated numbers while playing on good teams and in good systems. The opposite is true too.

    • ChinookArchYYC

      I think your right, a high tide raises all boats. Iginla’s numbers are inflated by a PDO of 103.5, which is inflated by a 95% save percentage. To Eugene’s comment context should be considered. More to the point no advanced stat should be considered on its own.

  • ChinookArchYYC

    Is there a way to determine how many goals a player is worth by looking at their Corsi? Sort of like a WAR, but for hockey? I know points is used to judge players, but while it is good, I’m not sure if that is the most reliable way.