HarrisonFord
President of the Drew Doughty Fan Club
Here's what I have a problem with when it comes to applying sabermetrics to hockey.
1. I know it applies well to baseball, but the dynamics of baseball and hockey are very different. Baseball in a nutshell is a series of one-on-one events. The pitcher throws the ball, and the batter tries to hit it. Very basic, very simple. One pitcher, one batter, one event, and primarily one defensive player.
2. What I have issue with, is that hockey has far more dynamics involved. Instead of three variables like in baseball (pitcher, hitter, defender) you have numerous others. If we are talking strictly about shots, then there are the following: shooter, goalie, opposing defenders (2), opposing forwards (3), available teammates (1-4), shot location, is there a screen?, is there a tip?, is there a rebound? Are the defenders boxing out the rebound-attacking forwards? All factors that are not considered by corsi.
3. What corsi does is essentially removes all the above factors that I listed above and labels them as insignificant variables. Which undermines it's accuracy. Essentially, all shots are treated equally; and they are not. A shot from the point is not as dangerous as a shot from in close. An unscreened shot is not as dangerous as a screened shot. A clean shot is not as dangerous as a tipped shot.
Advanced statistics in hockey is really just in it's infancy. It's not an exact science; not yet. Not close. If you want to be actually predictive with a stat/measure, then you need to have your indicator be much more accurate than it currently is. Generalization of variables makes your predictor less accurate.
If you look in economics and check the confidence level of predictors (i.e. in industries, etc.) they typically have 99% accuracy. If you look at this: http://www.habseyesontheprize.com/2013/4/4/4178716/why-possession-matters-a-visual-guide-to-fenwick it shows that teams with below .500 fenwick (not corsi, I know, I know) made it to the playoffs 31% of the time. That's a 69% predictive success rate for "bad puck possession teams". On the opposite end? "Good puck possession teams" only made the playoffs 76% of the time. Not really something you can put a whole lot of stock into. Clearly there is evidence here (76% vs. 31%) that better puck possession teams make the playoffs, but it is in no way a certainty.
1. I know it applies well to baseball, but the dynamics of baseball and hockey are very different. Baseball in a nutshell is a series of one-on-one events. The pitcher throws the ball, and the batter tries to hit it. Very basic, very simple. One pitcher, one batter, one event, and primarily one defensive player.
2. What I have issue with, is that hockey has far more dynamics involved. Instead of three variables like in baseball (pitcher, hitter, defender) you have numerous others. If we are talking strictly about shots, then there are the following: shooter, goalie, opposing defenders (2), opposing forwards (3), available teammates (1-4), shot location, is there a screen?, is there a tip?, is there a rebound? Are the defenders boxing out the rebound-attacking forwards? All factors that are not considered by corsi.
3. What corsi does is essentially removes all the above factors that I listed above and labels them as insignificant variables. Which undermines it's accuracy. Essentially, all shots are treated equally; and they are not. A shot from the point is not as dangerous as a shot from in close. An unscreened shot is not as dangerous as a screened shot. A clean shot is not as dangerous as a tipped shot.
Advanced statistics in hockey is really just in it's infancy. It's not an exact science; not yet. Not close. If you want to be actually predictive with a stat/measure, then you need to have your indicator be much more accurate than it currently is. Generalization of variables makes your predictor less accurate.
If you look in economics and check the confidence level of predictors (i.e. in industries, etc.) they typically have 99% accuracy. If you look at this: http://www.habseyesontheprize.com/2013/4/4/4178716/why-possession-matters-a-visual-guide-to-fenwick it shows that teams with below .500 fenwick (not corsi, I know, I know) made it to the playoffs 31% of the time. That's a 69% predictive success rate for "bad puck possession teams". On the opposite end? "Good puck possession teams" only made the playoffs 76% of the time. Not really something you can put a whole lot of stock into. Clearly there is evidence here (76% vs. 31%) that better puck possession teams make the playoffs, but it is in no way a certainty.