Poor Crease/Poor Slot: It Doesn't Matter

Thursday on TSN.com, Analytics contributor Travis Yost launched an article titled: You don’t have to be Chara to be an effective NHL defenceman .  Within the article, Yost broke down defensemen using shot location data to determine whether or not the defensemen in the chart was a good crease clearing or slot clearing defensemen or a bad one. 

Yost's methodology makes sense in theory, for certain. And it is an innovative way to break it down. However, Yost leads the article off as such:

"There are 30 seconds left in the game, and your team is up a goal. You need to ensure that the final wave of attacking forwards are kept off of the scoreboard, deterring them from those prime scoring areas where goals are frequently scored."

He then embraces his shot location methodology to break defensemen down into four quadrant (the 1-4 ranking is something I did to apply a unique identifier to each quadrant in Yost's chart. This will come into play later in the post!):

  1. Good crease defending, good slot defending
  2. Good crease defending, poor slot defending
  3. Poor crease defending, good slot defending
  4. Poor crease defending, poor slot defending

The argument Yost continues down, is that you'd rather have players that fall into that first quadrant on the ice with thirty seconds to go, rather than guys in that fourth quadrant.

My hypothesis? This doesn't matter at all.

To test it out, I used the same data set as Yost, d-men who have played over 1600 minutes and 120 games over the last two seasons. In the data set, I applied Yost's quadrant to each d-man depending on where he landed in Yost's chart.

If what Yost believes is right, then the d-men who fall into quadrant four should have a very low relative goals for %. If they are giving up too many prime chances, it should surely be hindering their metrics. Any coach would want to stay away from these guys when there are thirty seconds left in the game, because they get scored on far more often, relative to their teammates, then players who may find themselves in quadrant one. 

(Using relative GF% instead of raw GF% to eliminate any goalie impact on the numbers. A player in quartile four like Dan Girardi who plays in front of Henrik Lundqvist might just have a better raw GF% than Jake Gardiner in quartile one, who plays in front of Jonathan Bernier or James Reimer). 

To get a clean look at the graph, I applied a random decimal to each player point so they would not be stacked on the vertical axis line. There's no difference between any of the data points within each quadrant on the X-axis. 1.9 or 1.1, they are both quadrant 1 players. 

As we can clearly see, there is no relation between what quadrant Yost has categorized these defensemen as, and performance on goals relative to their teammates.