Chu: Using Corsi, zone starts to group defensemen

Watching a defenseman, determining his playing style, and perhaps, even making a comparison to another defenseman are part of the player evaluation process. Player comparisons are usually made based on their styles of play: “stay-at-home defenseman”, “mobile defenseman”, “physical defenseman”, etc.

How would player comparisons be made if we were to look only at the data? And I don’t mean points or hits, not the obvious things that we can see watching the game. Instead, let’s take Corsi Quality of Teammates per 60 minutes, Corsi On-Ice per 60 minutes, Offensive Zone Start %, and penalty differential (penalties drawn subtracted by penalties taken) variables in five-on-five situations and treat them as inputs in an agglomerative hierarchical clustering model. The model groups players with similar numbers together.

Defensemen who have played at least 40 games this season are analyzed. Data from this season is collected from behindthenet.ca on February 10th, 2015 and are standardized or normalized with z-scores, so that each variable is on the same scale. Clustering models are highly influenced by large differences in scales.

 

Here are some of the interesting groups of blueliners from the algorithm. (Note the variables used in the clustering model are Corsi/60, QOT, OZS, PenDiff/60):

Anton Stralman, Matt Carle, Jason Garrison

Name TOI/60 Corsi/60 mins Corsi QoT OZS% PenDiff/60 P/60
Anton Stralman 16.1 12.0 8.7 54.0 -0.2 0.9
Matt Carle 17.2 10.1 7.8 53.2 -0.3 0.9
Jason Garrison 15.6 7.1 7.1 54.7 -0.3 1.0

The clustering model finds three Tampa defensemen to be much alike. (The only other group of defensemen from the same team clustered together are Mark Giordano and T.J. Brodie as you’ll see later.)

Carle has played about 46.3% of his even-strength time with Stralman. Garrison shares much less ice time with them, pairing up with either Carle or Stralman about 8% or 9 % of his even-strength time respectively. The three Bolts defenders haven’t played large proportion of even-strength situations together, so one is likely not carrying another, stats-wise.

Tampa has had approximately 56% of non-neutral zone faceoffs in the offensive zone. The three defensemen have actually played less in the offensive zone than the team average.

Their offensive efficiency, points per hour, a variable not included in the algorithm, is not much different between the three blueliners.

The only drawback with having them on the ice is that they tend to take more penalties than they draw, but that problem seems to be minor given that the Bolts are one of the top teams in the Eastern Conference.

Shea Weber and Brooks Orpik

Name TOI/60 Corsi/60 mins Corsi QoT OZS% PenDiff/60 P/60
Shea Weber 19.9 -1.1 3.5 45.0 -0.6 1.3
Brooks Orpik 18.2 -3.7 3.0 47.6 -0.7 0.6

Similarities between Shea Weber and Brooks Orpik go beyond hitting and blocking shots. They both start in the defensive zone between 52-55%, on average, of all their end zone face-offs and take a fair amount more penalties than they draw. While their teammates possess positive Corsi numbers, the two physical defensemen have negative Corsi values, which may not be unexpected because they block a high number of shots.

Johnny Boychuk and Jake Muzzin

Name TOI/60 Corsi/60* QoT* OZS%* PenDiff/60* P/60
Johnny Boychuk 16.1 16.3 10.7 56.7 0.1 0.9
Jake Muzzin 18.7 15.6 9.8 57.5 0.1 1.0

Jake Muzzin and Johnny Boychuck’s stats are nearly identical. Based on these positive numbers, they both appear to be great additions to any team. Muzzin’s 7.9 and Boychuk’s 8.8 Corsi Relative values are excellent considering the Kings are first and the Islanders are fifth in Corsi.

Both Boychuk and Muzzin’s offensive zone starts are about three to four percentage points higher than their teams’ averages, which could partially explain their high Corsi rate and Corsi relative.

Dion Phaneuf, Mark Giordano, T.J. Brodie, Andy Greene

Name TOI/60 Corsi/60* QoT* OZS%* PenDiff/60* P/60
TJ Brodie 19.4 -5.2 -8.6 40.7 0.1 0.8
Dion Phaneuf 16.5 -13.4 -8.7 42 0.1 0.6
Mark Giordano 18.7 -3.2 -8.8 41.2 0.1 1.4
Andy Greene 18 -3.8 -3.6 40.1 0 0.6

Brodie has been paired with Giordano for about 91% of his even-strength time. Like his partner, Brodie blocks a lot of shots and has a very similar penalty differential rate. They appear to be a great pairing despite what their raw Corsi numbers indicate.

Andy Greene may be an odd name in the cluster. Greene isn’t on the same team with any of them nor is he known for being a “tough” defenseman, but he does block a lot of shots. The Devils defender’s numbers are more similar to Brodie and Giordano’s than to Phaneuf’s.

Leafs fans may ask, “how in the world is Phaneuf in the same group as a potential Norris trophy candidate?” The data suggests Phaneuf is the worst defenseman of the group – his even-strength ice-time is the lowest of the group – but like the others, he makes most of his non-neutral zone starts in the defensive zone and is not on the negative side of the penalty differential.

Grouping defensemen based on shot attempt differentials, penalty differential, and zone starts can reveal some patterns and make us look at certain players from a different perspective.

Comments are closed.