Out with the old
Thursday June 4th the Oilers announced that assistant coach Craig Ramsay would not be returning after only one season. At the time, I was elated. I was puzzled by the choices the Oilers had made regarding defensive deployment and I figured that Ramsay was responsible. I was disappointed in the way he used Martin Marincin. Out of the gate Marincin faced the toughest minutes of any of the Oilers defence and was given a low percentage of offensive-zone starts. As far as I could tell the Oilers were trying to sink him. It was patently obvious that they were pampering the likes of Justin Schultz and throwing Marincin to the dogs.
But what if I was wrong (no, that could never be)? What if things weren't how they initially appeared? What if there was a method behind Ramsay's madness or aspects of his coaching practice could be quantified? It's difficult to make decisions about how to deploy a group of defensemen that aren't having much success. As our fair editor Ryan Batty wrote in response to one of my comments:
Honest question: How do you effectively deploy a defence with Nikitin, Ference, and Schultz? Clone Petry?
A failing class
Coaching, in my estimation, is not unlike teaching and the Oilers' defence is a failing class. Their Corsi-Against is well above 50, and their For Percentages are below evens. Evaluating failing students can be tricky, but it's important to measure student progress in order to encourage learning.
There are three basic kinds of evaluation teachers can use to measure progress: evaluation based on external objective criteria, evaluation relative to the performance of a population, and evaluation relative to past performance. All three are equally important, and thanks to analytics all three are accessible in some way to the curious fan.
Corsi is effectively an observable or objective measure. It stands-in as a measure of puck possession. We can look at how Corsi fluctuates over time in order to evaluate players relative to past performance.
But what of the second category of evaluation? How do the Oilers perform relative to other players in the league? When it comes to Corsi, this is not an easy thing to get at. So many variables affect scores that direct comparison between players is often unfair.
dCorsi
Enter dCorsi, the brainchild of Stephen Burtch. Burtch is a well respected figure in the analytics community and has, among other accomplishments, a PhD in physics. He's also a Leaf's fan who's authored many important posts over at the Pension Plan Puppets. The culmination of years of statistical tinkering dCorsi is the difference between a players' actual (objective) Corsi-For percentage (CF%) and an expected CF%. He describes it as follows:
dCorsi stands for 'delta Corsi'… Here we are using delta to represent differential, specifically the differential between a skater's Expected Corsi and Observed Corsi… The differential between Expected Corsi (as determined by regression) and Observed Corsi depends on the "usage effects" explained by Expected Corsi - this is what you'd expect out of a perfectly average player in an average season if he was handed the same minutes with the same players against the same opposition… In an individual year the number will shift due to random factors and effects but over time (particularly multiple seasons) a lot of this washes away and you're left with a pretty solid measure of player skill relative to their usage.
In other words, dCorsi is a measure of a player's performance relative to an average performance constrained by similar variables. It measures a player's performance relative to a population of a similar age facing similar situations. Obviously the variables composing a given player's situation are many.
Expected Corsi-For (ECF) variables include: age, position, five-versus-five time on ice per 60 minutes, an average of teammate's corsi for scores per 20 minutes of play, percentage of zone starts in offensive and neutral zones, face-off win percentages in offensive and neutral zones, and a dummy variable "team".
Expected Corsi-Against (ECA) variables include: age, position, an average of quality of teammate's corsi against scores per 20 minutes of play, percentage of zone starts in the defensive zone,face-off win percentages in the defensive and neutral zones, and a dummy variable used to represent the team.
The formulas used to calculate ECF, ECA, and dCorsi are rather complex and may be found in this document.
Looking for ECF, ECA and dCorsi scores? Don't worry, you don't have to calculate them yourself. Not since this past winter when the kind geniuses at war-on-ice.com decided to start calculating them for you. Prefer a Fenwick to a Corsi? They've got that one covered too.
The above chart shows expected (horizontal) versus observed (vertical) Corsi-Against Per 60 Minutes for defenders over the past three seasons. By hovering over the bubbles you can see the relevant cartesian coordinates. Bubble size corresponds to the difference (dCA60) between ECA60 and CA60. The lower the bubble, the better the shot suppression from the defender. The father the bubble is to the right, the higher the expected score. If you imagine a diagonal line transecting identical values for ECA60 and CA60, bubbles above and to the left of that line correspond to performances below expectation. Bigger bubbles correspond to higher differentials, which in the case of Corsi Against is not a good thing.
The 2014-2015 Oilers (red) are both lower and father to the left than the Oilers of 2012-2013 (blue) and 2013-2014 (orange). Ramsay's defence outperformed Steve Smith's for both observed and expected values. I interpret this to mean that Ramsay's squad was better positionally and better at suppressing attempted shots.
Ramsay's squad is composed of two groups of defenders: smaller-lower bubbles and higher-larger ones. Jeff Petry was Edmonton's best defensive defender in a group that exceeded expectations consisting also of Fayne, Nikitin, Klefbom, and Marincin. Aulie, Davidson, Hunt, Ference and Schultz gave-up more Corsis than average defenders of their same age facing similar circumstances.
The best defensive defender the Oilers have had in the past three seasons as measured by dCA60 is Ladislav Smid. Other dearly departed defenders who exceeded expectations include Jeff Petry, Mark Fistric, Anton Belov, and (just by the hairs of his chinny-chin-chin) Nick Schultz.
You can learn something about Ramsay's defensive strategy from this chart. Players with lower expected values (farther to the left) were given easier rides. This past season those players were Justin Schultz, Oscar Kelfbom, Brad Hunt, and Nikita Nikitin. You'll notice that Ramsay's red bubbles are more spread-out than the other two seasons. I interpret this to mean that he really thought about deployment.
The above chart shows Expected (horizontal) versus Observed (vertical) Corsi-For Per 60 Minutes for defenders over the past three seasons. By hovering over the bubbles you can see the relevant cartesian coordinates. Bubble size corresponds to the difference between ECF60 and CF60 (dCF60). The higher the bubble, the better the shot generation from the defender. The father the bubble is to the right, the higher the expected score. High expected scores belong to the most offensive defenders. If you imagine a diagonal line transecting identical values for ECF60 and CF60, bubbles above and to the left of that line correspond to performances above expectation. Bigger bubbles correspond to higher differentials, which in the case of Corsi-For is a good thing.
Ramsay's defenders are generally better on the attack than Smith's for both expected and observed values. The old guy wins again (this gives me hope)! For the Ramsay Oilers there are (again) basically two groups of bubbles, a series of larger higher-bubbles with positive differentials and two clusters of defenders who did not meet expectations for generating attempted shots. Hunt, Petry, Nikitin and Marincin all outperformed expectations. Keith Aulie did too just by the hairs of his non-playoff beard.
Ramsay's attacking deployment strategy is somewhat visible. The farther the bubble is to the right, the easier the offensive ride for the defender. Schultz, Kelfbom, Nikitin, and Hunt were all put in good positions to score. Of course, the red bubbles are probably boosted by improved play from the Oilers' forwards too (we will leave a defence of Keith Acton for another day).
The above chart shows Expected (horizontal) versus Observed (vertical) Corsi-For Percentage Per 60 Minutes for defenders over the past three seasons. This chart is the combination of the previous two. By hovering over the bubbles you can see the relevant cartesian coordinates. Bubble size corresponds to the difference between ECF%60 and CF%60 (dCorsi). The higher the bubble, the better the observed possession of the defender. The father the bubble is to the right, the higher the expected possession. High expected scores belong to the players given the easier competition, best teammates, and zone starts. If you imagine a diagonal line transecting identical values for ECF%60 and CF%60, bubbles above and to the left of that line correspond to performances above expectation. Bigger bubbles correspond to higher dCorsi; these are players who's combined values exceed expectation. These players include Petry, Marincin, Mark Fayne, Hunt, and Nikitin.
Old Guys Rule
The above bar-chart shows dCorsi for the Oilers defence over the past three seasons. The 2014-2015 Ramsay Oilers were better deployed than the Smith teams of the previous two seasons. We know this because the variance is smaller between the lowest and highest dCorsi scores. Our old friend Ramsay used his players more effectively than Steve Smith.
A persistent negative dCorsi means that the defender is sinking in relation to Expected Corsi values, the context in which he's positioned. A persistent positive dCorsi means that the player is exceeding expectations and might need to face more difficult competition. A consistently positive dCorsi could also mean that the player in question is very good for his age given a particular context. Experienced coaches like Ramsay have in their minds an intuitive sense of this where to put players when. Ramsay achieved a more-balanced deployment than his predecessor. How he got there? Well, probably nobody knows exactly.
The curious case of Martin Marincin
Marincin has had the highest dCorsi of any of the Oilers defenders of the past two seasons. What this means is that he is one of Edmonton's most important defensive prospects. According to Burtch, players who keep producing high dCorsi are players who drive possession. Marincin's leaps and bounds above expectation given the tough toughs that he's faced and his age. He's exceed expectation more than Jeff Petry (cue collective sigh about trade). As long as Marincin can maintain positive spirits, maybe it's not a terrible idea to have him face such difficult competition.
Maybe, just Maybe Craig Ramsay wasn't a crazy old fool but rather was highly concerned about the development of the Oilers' most important defensive asset. Observing Marincin's excellent play relative to the challenges presented to him, perhaps Ramsay figured Marincin for an "A student" and wasn't about to throw him a bird course.