Relative Deployment Z-Score

Hockey players have roles. Some players are asked to score, some are asked to be physical, some are asked to shut down top competition. With that said, how do you go about comparing players between roles? It’s easy to look at point totals and say “Player A is better than Player B because he has more points,” but that doesn’t mean it’s correct. What if Player A is getting more playing time? What if player A is facing weak competition and aided by strong teammates? What if Player B is facing top competition with weak teammates? Advanced stats like CorsiRel QoC and QoT along with Offensive Zone Start % paint the picture of how the player is being used---but that still doesn’t solve the problem of how to compare players in different situations. I am attempting to solve this problem by creating a stat called RDZ: Relative Deployment Z-Score. It’s simple in theory: How does a player fare against other players who were deployed in similar situations.

To start I’ve compiled data for the past five years (06-07 to 10-11) for every forward who played at least 40 games in a season (If a player played 40 games or more in each of the five years, there will be five different results---one for each season). The total sample size is 1,514 players broken down like so:

In order for this to be broken down I copied and pasted every team (individually, 30 teams per season, 5 seasons, 150 teams) and figured out the Comp/Team stats for each player for ever year (Yes, this took a while and the spreadsheets contain tons of data). Once I had it broken down I then separated it into 16 different spreadsheets, one for each deployment situation (1v1, 1v2, 1v3, 1v4, 2v1, etc). Over the last five years 83 players went up against top competition with top teammates over the course of one full season. The next step was to set up a graph (I am not good enough with the SB Nation Fan Post to copy/paste graphs, if someone can help me with this I think it would be easier to explain) comparing the OZS% to each of G+A1/60 and CRel for every player (Two different graphs). I was looking for some sort of regression line, ideally sloping upward (As OZS% goes up, so should both G+A1 and CRel) and after the data was entered, I found the results I was looking for. When isolating each of the 16 scenarios we could then isolate other stats within each individual scenario looking for correlation and causation. The results were promising, with the G+A1 having a positive slope and an R^2 value of .117 for G+A1 and .0832 for CRel (not great, but when you look at the actual graphs you can see that there is an obvious correlation with OZS% and both G+A1 and CRel).

Once I had the expected G+A1 and CRel for a given OZS%/QoC/QoT combination, we can then enter the individual’s OZS% and find the expected G+A1 and CRel for the player. Once we have that we can find the variance between actual and expected. In the data for the 83 players who were “1v1” the average variance between actual and expected G+A1 was “-7E-04” or “-.000746987…”, either way, it’s basically zero (average variance for CRel was 0). The standard deviation of G+A1 was .3295 and 4.88 for CRel. If you have the actual variance, the expected variance, and the standard deviation of variance, we can then calculate the “Z-Score,” or a formula for explaining how many “standard deviations” each individual result is away from the average. 66.6% of a population falls between the Z-Scores of -1 and 1. 95% between -2 and 2, and 99.7% between -3 and 3. So when looking at a Z score think of it like this: 0 is average (better than 50%). 1 is better than 84.1%, 2 is better than 97.7%, 3 is better than 99.9%. -1 is better than only 15%, -2 is better than only 2.3% and -3 is better than .1%. The whole theory of the stat is that if you know a player’s Z-Score for 1v3, or 2v4, or 1v1 etc... you can then compare it to anyone of any other situation.

We calculate the Z-Score for both GA1 and CRel and call them “GA1Z” and “CRelZ”; we then take a 50/50 average of the two scores to give us our Relative Deployment Z-Score. I use the 50/50 because I want to put equal value on individual scoring and the ability to drive the play forward. Before we look at the players I want to look at the average G+A1/60 and CorsiRel's of each situation given a 50% OZS:

There have been other attempts at stats similar to this one, but nothing takes into account all three major situational factors: O-Zone Start %, competition AND teammates. You can clearly see in the above table that for each level of competition, as your teammates get better, so do both your individual scoring and your ability to drive the play forward---not to say that wasn't expected. It is important because there is an individual formula for every scenario.

Lets take a look at five players of all different situations:

Lets take another look at those same players:

Datsuk is far and away the best player (his +2 rating means he’s in the top 97.5 percentile of his situation) of the above. Top competition with only 3rd line help, he’s expected to have a GA1 of 1.3 and CRel of -1.26 and he’s blowing it away with a 2.16 and 9.1. Staal’s not quite scoring as much, his Corsi is slightly lower, he’s playing similar situations, but playing second line competition opposed to top line and he’s in the offensive zone 4% more. This bumps his RDZ to 1 (85th percentile). Drury is the one performing as expected with his scoring, but he’s a -17.6 opposed to a -6.01 expected, but he’s doing it while only starting in the offensive zone 38% of the time. Cooke’s with second teammates vs. top competition, scoring well below expected and being driven back while he’s expected to go forward, resulting him in a -1 (15th percentile). Powsie’s ’09 campaign was horrible. Top teammates vs. bottom competition aided by a 52% OZS, his GA1 is almost less than half of everyone else on the chart, and his CRel is 10 points lower than expected. Now that we see how the stat works, lets check out the Flyers this current season.

Let me start off by saying that Sean Couturier’s 2.535 RDZ is the second best of every player I’ve compiled data for, behind only Sidney Crosby’s 2011 season: 2.817. As a matter of fact, only 7 times ever has a player recorded a 2 or better (1500+ players evaluated): Crosby (‘11, ‘10, ‘08), Ovechkin (‘10), D. Sedin (‘10), Datsuk (‘11), and Guillaume Letendress’s fantastic ’09 season in which he, like Scooter, destroyed 4v4 situations with a 48.8 OZS% to the tune of a 1.53 GA1 and 6.2 CorsiRel when only a .8 GA1 and -9.4 CRel expected. So, yeah, Couturier is playing amazing. Talbot is another surprise behind Scooter, but not when you really look at how good he’s playing in his tough situations. Giroux, Voracek, Read, Jagr, and Harry Z all playing significantly better than an average player would given the situations. Before you freak out and say Hartnell is way underrated---don’t forget this is an even strength only stat, but he’s still above average. Simmer is the last one that can say that. A few of the Flyers at the bottom of this list shouldn’t surprise anyone. Shelley is horrible, in the bottom 10th percentile. Schenner is also in the bottom 15th but that’s been moving up quickly. Rinaldo is surprisingly average and so is JVR, but Danny B has had a rough even strength season, in the bottom 20th percentile. His scoring is almost on pace (1.52 actual, 1.7 expected) but his CorsiRel is just not.

And here are all Flyers for the past 5 years:

And a closer look at the best and worst seasons over the past five years:

The first thing I want to point out is that in 2009 the Flyers were led by 46 goal scorer Jeff Carter, 30/30 Scott Hartnell, and now All-Star Joffrey Lupul. All 3 are in the top 12 in the past 5 years, all 3 had a negative CorsiRel, however they lit the lamp enough to more than compensate for the very mediocre CorsiRels. Aaron Asham makes the list. twice. Giroux’s rookie year and his 2011 make the top 12. Upshall’s first year as a Flyer and Mike Richard’s breakout 2008 season were all memorable. Nicolay Zherdev and his ridiculous Corsi from last year put him up here, and Briere and Jeff Carter’s 2011 rounds out the memorable seasons. As for the bad ones… Powe’s rookie year, Kapanen’s final year, Leino’s half Det/half Philly season, and Danny Carcillo’s first season’s were all ones to forget. A surprising appearance by a dominant PK duo: Lappy’s 2010 and Bettsy’s 2011. A not surprising appearance by king argument starter (and not-so-good hockey player): Jody Shelley.

Questions?