Corsi is Not a Proxy for Possession
Posted 1:06 PM ET | Comments 5
There was a lot of debate regarding advanced analytics this past offseason. The debate was filled with passionate supporters on one side and adversaries who vehemently rejected the statistic as relevant on the other. This blog is not an extension of that debate and does not challenge the relevance or significance of advanced analytics such as Corsi; however, it does challenge the theory that Corsi is an accurate proxy for possession.
People often interpreted Corsi as a statistic that demonstrates which teams or players drive possession and it appears they may be wrong.
Scott Cullen, the analytics guy for TSN and a strong believer in Corsi, tweeted today that based on a report, Colorado’s timed offensive zone possession for the 2013-14 season ranked in the top 5 of the NHL along with Chicago, Los Angeles, Boston, and St. Louis. Despite their high ranking, they ranked 27th in the league in Corsi during the 2013-14 season. This directly contradicts the idea that Corsi is a possession proxy.
To be clear, Scott Cullen cited Mike Kelly, a host on the NHL Network, as the source of the report. Kelly hasn’t given any further information regarding the report, but for the sake of this blog, we’re going to assume his information is accurate (advanced analytics supporters love assumptions).
Corsi is calculated by taking total shot attempts for and dividing them by total shot attempts against. The logic is that if a team possesses the puck more, it’s safe to assume they likely take more shots. It started as a team metric but was soon applied to individual players as well. It’s easy to see why that concept makes sense; however, at the end of the day, it’s simply an assumption. Additionally, Corsi uses imperfect and inconsistent statistics in its calculations: shots are recorded differently in every NHL arena. Again, pro-analytics supporters argue that it’s safe to assume that over the course of an 82 game season, everything balances out. Another assumption.
It’s difficult to argue that high Corsi numbers don’t directly correlate to wins. There has been a lot of research done in this regard and generally speaking, successful teams rank towards the top of the NHL in Corsi. Of course, there are a few exceptions to the rule (New Jersey ranks high but missed the playoffs, Colorado ranks low but made the playoffs) but regardless, high Corsi means wins. And it makes sense that if you outshoot your opponent consistently, then you have a higher chance of winning. But if Kelly’s information that Cullen cited is correct, it means that Corsi’s correlation to possession is greatly exaggerated and potentially non-existent.
People often interpreted Corsi as a statistic that demonstrates which teams or players drive possession and it appears they may be wrong.
Scott Cullen, the analytics guy for TSN and a strong believer in Corsi, tweeted today that based on a report, Colorado’s timed offensive zone possession for the 2013-14 season ranked in the top 5 of the NHL along with Chicago, Los Angeles, Boston, and St. Louis. Despite their high ranking, they ranked 27th in the league in Corsi during the 2013-14 season. This directly contradicts the idea that Corsi is a possession proxy.
To be clear, Scott Cullen cited Mike Kelly, a host on the NHL Network, as the source of the report. Kelly hasn’t given any further information regarding the report, but for the sake of this blog, we’re going to assume his information is accurate (advanced analytics supporters love assumptions).
Corsi is calculated by taking total shot attempts for and dividing them by total shot attempts against. The logic is that if a team possesses the puck more, it’s safe to assume they likely take more shots. It started as a team metric but was soon applied to individual players as well. It’s easy to see why that concept makes sense; however, at the end of the day, it’s simply an assumption. Additionally, Corsi uses imperfect and inconsistent statistics in its calculations: shots are recorded differently in every NHL arena. Again, pro-analytics supporters argue that it’s safe to assume that over the course of an 82 game season, everything balances out. Another assumption.
It’s difficult to argue that high Corsi numbers don’t directly correlate to wins. There has been a lot of research done in this regard and generally speaking, successful teams rank towards the top of the NHL in Corsi. Of course, there are a few exceptions to the rule (New Jersey ranks high but missed the playoffs, Colorado ranks low but made the playoffs) but regardless, high Corsi means wins. And it makes sense that if you outshoot your opponent consistently, then you have a higher chance of winning. But if Kelly’s information that Cullen cited is correct, it means that Corsi’s correlation to possession is greatly exaggerated and potentially non-existent.
Players that seem to exemplify this are the Sedin twins. They control the puck for minutes at a time and yet only take one or two shots. Those shots are often goals but it isnt reflected well with corsi. Its not a perfect stat, its just one of several important ones.
But how many shot attempts do the opposition generate while the Sedins have the puck?
Just to clarify: Corsi is not calculated by taking total shot attempts for and dividing them by total shot attempts against. You divide them with total shot attempts for AND against, leaving you with a shot attempt percentage for each team (or player). I agree that it is an assumption to use Corsi as a proxy for possession. But it is an assumption that aligns pretty well with common sense. In order to generate shot attempts for, you have to possess the puck. And when you possess the puck, you will deter shot attempts against. The same is true for your opponent.
And if this is true, you would also expect shot attempts to correlate with actual puck possession to a fairly high degree. From what I understand you have exactly one observation to back up your claim that this is false (though the details are a bit fuzzy). Do you trust that single observation more than others? I mean, you also mentioned some other teams: Blackhawks, Kings, Bruins and Blues. Do they fit with your alternative hypothesis? If not, why not? Also, have you thought about the time that the Avs spend in their own zone, without puck? If that time is significant in relation to OZ time, then you have no contradiction at all. Colorado might just be a bit unusual in the amount of time that is spent in both ends. A bold prediction: the rate of neutral zone face offs is probably pretty low when Colorado is one of the teams.
The Sedins are consistently great Corsi players.