Corsi and Moneyball

There is nothing “advanced” about “advanced stats”, and CORSI doesn’t stand for anything. “Corsi”, which is a simple shot differential metric, is named after the Buffalo Sabres goaltending coach that used it to measure the workload on Sabres goalies.

It’s simply counting up shot attempts in one end of the ice. Goals, shots, missed shots and blocked shots. The NHL has been keeping that data since 1998, and the early version of the NHL’s Event Summary isn’t too different to the sleek one we have today.

Jesse Spector of The Sporting News wrote up a feature on Corsi this season, and it’s well worth a read. There are several paragraphs I could quote that I could use as a jumping off point to write the rest of this post, but I’ll use these two:

Actually, one more is good:

“an engineer in California” of Gabe Desjardins of BehindTheNet.ca, was not the person that turned Corsi’s epiphany about goaltender workload and began using it as a player-performance metric. That was Vic Ferrari, the nom-de-plume of a software programmer somewhere out West, writing for an old Oilers blog Irreverent Oilers Fans.

Basically, Ferrari’s thesis was this: if Corsi found that looking at all shot attempts was a metric that signified more closely a goaltender’s workload, then should the players that make the other goaltender work the most not get credit for playing with the puck often in the offensive zone?

When a long shot is taken from the far point, that is not a shot that has a chance of going in. From that point on the ice, a lot of players would actually fail to put it on net. But the goaltender still has to drop to his knees and see the puck into the corner and the defence still has to recover the puck and protect against forechecking forwards. Forwards that are successful in recovering that puck will have a better chance of generating *another* shot. If the defence clears the puck, the damage is limited.

Ferrari’s next big step was using the “zone time” count that the NHL.com game sheet had at the time, and noting that it matched up quite closely with the “Corsi” tracker that a Buffalo Sabres goaltending coach had unknowingly developed. Ferrari and Desjardins began writing scripts to collect NHL play-by-play and game sheet data and put them into spreadsheets, and in 2007-2008, the NHL began publishing which players were on the ice for each recorded event in addition to the event itself.

So Ferrari could track the Corsi numbers of the Edmonton Oilers’ players.

That’s basically a brief history of Corsi, and it developed on a side of the Internet that treated hockey as a hobby and not as a job. A lot of metrics were generated in those early days on Ferrari and Desjardins’ respective websites, and they’re lots of fun to go back and read and see how these ideas developed.

The concept of “tough minutes” is now used on broadcasts, but it wasn’t five years ago, and the only place you could find that sort of phrasing was on Oilers blogs where they would check to see which defencemen played against the forwards that scored the more goals, or checked to see which centremen were forced to take faceoffs predominantly in the defensive end. They seem like simple qualifiers to take into account, but remembering how I thought about hockey before I ran into this stuff, I would never take those into consideration when watching games.

This isn’t intended to be a lecture on how your thoughts on hockey or Joffrey Lupul’s thoughts on hockey are completely wrong. Lupul actually isn’t wrong. I agree with him that generally, contracts are not awarded by this Corsi. It would be silly to take only one thing into consideration when awarding a contract for one, but for two, the teams and players that have the best Corsis aren’t going to be the best teams and players.

Corsi is best used as a performance metric to better guess future performance. Havoc was raised in the short season because it can take more than a hundred games for results to match performance. This isn’t a new concept in any way. Players often talk about getting the bounces, but over a smaller sample of games, Corsi does a better job at separating the luck from the actual performance. Recover enough pucks, play with the puck in the right spots in the ice, and make the right decisions, and you’ll come out ahead.

(Note here that ‘the bounces’ don’t just refer to deflections, but it can be a simple thing about a puck hitting an attacking defender’s stick the wrong way and having it bounce clear to a player that sets up a 2-on-1. That chance doesn’t manifest itself all the time, even when you go looking for it. Tyler Dellow wrote a great line a month ago “a penalty shot is one of the best chances in hockey. You’ve got a 30% shot at scoring a goal or so. That said, you’d be foolish to build an offence around generating penalty shots because there aren’t enough of them to live off of”. It’s just like trying to structure an offence based on creating 2-on-1s.)

The discrepancy between these shot indicators and results is generally called “PDO”, which is the addition of shot percentage and save percentage. If you note that Brian Campbell and Erik Gudbranson had the lowest +/- in the NHL last season, you have to take into account that the Florida Panthers had awful goaltending. While on the ice, the even strength save percentage of the Panthers’ goaltenders for Erik Gudbranson was .899. For Brian Campbell, it was .898.

Consider that over a six-year period, the lowest even strength save percentage for a regular player on the ice is Filip Kuba’s .906. If those .899 and .898 numbers for Gudbranson and Campbell were repeatable, you’d expect some players to have six-year totals that low. You don’t, and it becomes importance to separate the luck from the reality when measuring players.

Somebody who watched Campbell and Gudbranson and the Panthers closely all season could tell you that perhaps Campbell and Gudbranson gave up a lot of extra 2-on-1 opportunities or 3-on-2s, or there were more pucks in the slot with those players on the ice. But those are just the results of small bounces we don’t appreciate that result in opportunities, the bounces that strategy cannot account for.

Just as a frame of reference to account for a statistic that people are used to, we know that 50 goals for a season is a terrific achievement. While 17 players had 50-goal seasons in the last seven 82-game seasons, no player had 350 goals between 2005 and 2012 which would mean breaking 50 every year. Though 50 goals is a great thing to have, it is also not sustainable, and you can’t expect players to have 50 goals in any given season. This goes the same for shooting percentages and save percentages.

Sometimes we’re successful at doing that. James Mirtle wrote about Nazem Kadri’s high on-ice shooting percentage right before Kadri’s drought at the end of last season. A year ago, I wrote about the Minnesota Wild’s results imbalance to their shot metrics and the Wild went from first in the Western Conference to out of the playoffs. Sometimes, not so much. Even though it was a 48-game campaign, I still forecasted that the Leafs would not make the playoffs last year based on their PDO after 17 games. While their PDO dropped and their winning percentage dipped from .588 to .516, certain factors kept the Leafs in the race. For a year and a half now, I have been forecasting that Joffrey Lupul’s shooting percentage would regress to the mean, but the dice have no memory, as they say, and Lupul shot at a 26.2% clip last season to score 11 goals in 16 games.

But again, it takes years and years to get things right. Ville Leino was once a sought-after free agent and signed a $27-million contract and has so far scored 10 goals in the two years he has played with the Sabres. Many people have brought up David Clarkson’s success in his last two seasons as a reason the Leafs would sign him, but fail to take into account that the Leafs should not be paying him for his performance in the last two seasons.

The “certain attributes” discussed by Lupul can be used to justify Clarkson’s seven-year deal, but Brad Richardson, Matt Hendricks, Boyd Gordon or a number of other “gritty” players that signed in the free agency class that espouse those attributes still signed for less and a smaller number of years. What’s the difference between Clarkson and Brad Richardson. Ultimately, it’s the production that the Leafs expect out of Clarkson that separates him from the Max Lapierres of the world or even the Clarke MacArthurs.

Moneyball was never about “stats versus scouts” or drawing a line in the sand between some new smart way of thinking and an old dumb way of thinking. Those lines simply don’t exist. Many “numbers” guys will use screen shots or video evidence from game footage to look at why a discrepancy in a number showed up where it shouldn’t have. A lot of coaches who have grown up around rinks understand that they need more than “their eyes” to gauge how their team has been playing. In minor hockey rinks, a backup goalie will count shots and faceoffs. In junior hockey rinks, healthy scratches are dispatched to the press box to count shots, faceoffs, and turnovers.

Moneyball was a business book and not a sports statistics book. The cash-strapped Oakland A’s were one of the first teams to use the data that everybody else was already tracking and began partially using it to make player decisions because there was a chance they could find undervalued players if they focused on walks, home runs, and patient hitters. Just a couple of years after Moneyball came out, every team started hiring Baseball Prospectus writers and paying attention to OBP. General manager Billy Beane still runs the Oakland As, and they won a division title last season even though the edge that exists by understanding the importance of not making outs has been wasted. 30 teams know that now, so Beane moved onto other undervalued things like pitching, defence and youth.

As applied to hockey, “Moneyball” would not be about using Corsi to make player evaluation decisions. It’s a tool to make better forecasts, by using other things you know about a player’s scoring ability, playmaking ability, and how much character, commitment and drive he has to the game. All those things are important, but they all have to be looked at together to make a 

For the Toronto Maple Leafs, I would doubt that there are a handful of executives in the room that have read Moneyball and understand it in any significant way beyond “derp, baseball team uses stats to win games. That would never work in hockey” which is a simplistic way to view the book as a whole. It was about exploiting market inefficiencies. If store-brand sugared flakes tasted as good as Frosted Flakes™, who in the right mind would buy the more expensive cereal?

But that doesn’t have much to do with Lupul. Lupul plays the game, and plays it as best he can, and it’s up to people in the executive offices that determine whether it’s wise to bet on his future or the future of a potentially cheaper replacement. In this analogy, Nik Kulemin ($2.8-million) is my store-brand, while David Clarkson ($5.25-million) is the national brand.

And lastly, nobody that uses Corsi or QualComp or even strength points per 60 in any tangible capacity would suggest that statistics are the be-all and end-all. Statistics tell you what happened, but there are other things that help you determine what *will* happen, and when making bets with contracts, how a player performs relative to the money and years he is being paid is what is important.

Tools in the box. Anybody paid in any executive capacity should have to account for all the available information. And for the love of Science, stop calling anything I do on this blog “advanced”. It’s a way of evaluating how a player has performed in puck-possession over a small sample of games by using a statistic that the NHL already tracks, so that decisions you make aren’t biased by luck.