Monday, May 30, 2016

What's Wrong With Corsi?


In recent years the NHL has been undergoing a "statistical revolution" with the introduction of advanced analytics. Among the most heralded of these statistics is Corsi, which attempts to estimate puck possession with a form of "shots at goal plus/minus". While I agree that these analytics can be useful when put in the proper context, they can lead to poor decisions when trusted too much. Sometimes Shots at Goal Plus Minus can give you a better understanding of a player's contribution to team success, but not always.

If you look at the leaderboard of best and worst Corsi, something doesn't pass the eyeball test. I am a Detroit Red Wings fan who has been starved for a true #1 defenseman ever since the retirement of Niklas Lidstrom. A player you can play against the other team's best player and win the game. They have still not replaced that void, making them mediocre at best despite a talented group of forwards. Niklas Kronwall and Danny DeKeyser are often forced to face tougher assignments than they should, and they have a bad Corsi. Brendan Smith and Jakub Kindl are their 6th and 7th defensemen, and they have great Corsi. At least until they dumped Kindl for next to nothing.

Anyone who watches the games knows that if you put Smith and Kindl on the top line to play against the other team's best players, the results would be catastrophic at best. But I'm sure some analytics analyst out there has written a post somewhere that Detroit could be a Cup contender if they played Kindl and Smith 20+ minutes a night.  It's ridiculous, but without the proper context, how do you know?

I think team Corsi might be a better indicator than individual Corsi (even though not all shots are created equal). Individual Corsi is too much swayed by the other 9 skaters on the ice to say this is what this player is responsible for. A writer for TSN recently wrote that Florida won the Gudbranson trade by a wide margin because he had a bad relative Corsi against, meaning his team gives up more shots than usual when he's on the ice. If we used relative Corsi against to measure the quality of a defenseman, then 3 of the worst D in the NHL are Brent Seabrook, Dan Girardi, and Roman Josi (who most experts will tell you are pretty good). It also has the two best defensemen as Josh Manson and Fedor Tyutin. Don't panic just yet Canucks fans.

All these stats are significantly affected by who else is on the ice. If you play a great D with a terrible D, one guy gets his numbers dragged down, the other dragged up. You need to be able to identify the false positives and false negatives, because there are many. Too many. To those NHL teams out there who are going down this rabbit hole, be warned if you can't properly distinguish between the right and the wrong, it will lead to some very costly mistakes and comical blunders. Proceed at your own peril.

No comments:

Post a Comment