From Kukla's Korner blogger PuckStopsHere:
Corsi Numbers Top 20
Kuklas Korner, July 10 2009
In this summer’s look at sabermetrics and hockey, I have introduced the Corsi Number as an alternative to +/- ratings.
Let’s take a look at some of the players with the best Corsi Numbers from the 2008/09 regular season. The best online source for this information is behind the net, but it treats Corsi Numbers as a rate stat (giving them per minute of ice time).
I think it is a much more logical starting point to look at the raw counting numbers. This is how we most frequently view +/- ratings and should be best comparable to Corsi Numbers. In the adjustment process, we may wish to treat them as rate stats, but not at this point.
Here are the top 20 Corsi Numbers from 2008/09:
The most obvious conclusion when looking at this list is that the Detroit Red Wings dominate it. Seven of the top eight players were Red Wings (Datsyuk, Rafalski, Kronwall, Lidstrom, Franzen, Hossa, Zetterberg) and an eighth Red Wing (Samuelsson) also appears.
In the same vein, no Boston Bruins appear despite many of them having high +/- ratings. Boston had the top two +/- ratings (David Krejci and Blake Wheeler) and three more players in the top 20 (Dennis Wideman, Michael Ryder and Marc Savard) and has none of the top 20 Corsi Numbers. This shows the influence of goaltending on +/-. Boston had the goalie with the best saves percentage in the league (Tim Thomas), while Detroit had the starting goalie with the worst saves percentage in the league (Chris Osgood). Boston did better in terms of +/-, where we are looking at goals scored, but Detroit did better in terms of Corsi Numbers, where we look at shots directed at the net. Boston was able to allow a relatively high number of shots at the net while not allowing many goals, due to their good goaltending. Detroit allowed a relatively high number of goals, given the low number of shots allowed on goal, due to their bad goaltending. There is a goalie influence on +/- and it can be seen by comparing to Corsi Numbers.
One obvious conclusion is that to compare players on different teams, we need to make a team adjustment. There are too few teams with players in the top 20 Corsi Numbers for it to be any kind of comparison between teams and Detroit dominates too much.
Looking beyond the Detroit players, There are seven other teams represented on this list (Chicago and Washington three times each, Anaheim twice and once each for Calgary, New Jersey, NY Rangers and San Jose). These are all playoff teams and are all among the better teams in the league.
NHL MVP Alexander Ovechkin is the only player who was good enough to break the Detroit stranglehold at the top of the charts. He places third.
Most of the players on the list are players one might expect on a list like this.
The two biggest surprises are David Moss of Calgary and Scott Gomez of the New York Rangers.
Moss is an underrated Calgary grinder, but is he really this valuable?
Scott Gomez is a good set up man, who was lost with the Rangers because they had nobody who could score on the chances he created. He was traded to Montreal because of his large contract, but I think he will do quite well in Montreal (better than many predict) if he finds a top sniper to play with.
This list is probably closer to a list of top players in the NHL than a top 20 +/- rating list, despite the over-representation of Detroit Red Wings. This shows, preliminarily, that Corsi Number may be more useful than +/- because of the increased number of events involved.
Here's further explanation of 'The Corsi Numbers', excerpted from PSH's blog:
Looking at +/- ratings and the results of a couple standard methods of their adjustment, +/- ratings are sometimes considered a problematic starting point.
One problem is “signal to noise” there may be slightly over 100 goals scored in even strength for a given player in a given season. Their +/- is the difference between the goals for and the goals against.
If there are a few fluke events they can carry the majority of the signal (i.e. if a player is on for a few fluke goals those goals can greatly influence their +/-). One way to try to get around this is to increase the number of events we are looking at.
Buffalo Sabres goaltender coach Jim Corsi has pioneered another metric. Why not keep track of all even strength shots directed at the net both for and against a team when a player is on the ice.
This includes goals scored, shots on goal, blocked shots and missed shots.
The NHL keeps enough data in its games online that this can be calculated for each game. The main benefit of this method is that it gives roughly sixteen times the number of events for a player that standard +/- does. This is intended to increase the signal to noise (in experimental physics any counting number has an experimental error that scales with its square root so in principle the Corsi Number could be four times better).
The problem is that it equates a goal scored with a blocked shot. Those events are clearly not equivalent. The hope is that over a season everything will wash out. The same team that allows a lot of goals also allows a lot of failed shots directed at the goal. The hope is that the distinction between players is a real effect and not one of choice (i.e. a team might choose to allow a lot of low percentage long shots while another team might not). This number can be adjusted and interpreted in the same (or slightly different) manner to +/-.
Looking forward to the discussion on this and the views of our GM Hockey experts.