I want to preface this by saying I am far from an analytics expert. I've never studied math or stats beyond the high school level, but I've been a fan of hockey stats since I was a kid, and have continued to keep a very rudimentary following of the advances in analytics over the years.
Earlier this year, in the process of trying to rank the best defensemen in the league, I decided I needed a lot more numbers to go off. I'm of the belief that corsi isn't nearly as useful for evaluating defensemen as it is for forwards. Corsi also only measures 5on5 stats, while special teams are such an important, and specific, part of the game.
So I went searching for some good stats for evaluating defensemen, and found three that I've ended up using a lot lately.
RelPPGF/60 – This stat measures how many more or less goals a team scores on the powerplay when a given player is on the ice than when he's not, per 60 minutes of powerplay time. I set it only for 5-4 PP, to keep things consistent.
When devising a new stat (I’m certain I’m not the first person to invent it, but I haven’t seen it in common use, and it wasn’t a stat that some site had ready-made, I just had the idea and put together some filters), you always want to check and make sure it passes the common sense test, and indeed, the top guys (minimum 50 PP mins) are basically the guys you’d expect to help a powerplay – Draisaitl, Marchand, Point, McDavid, Pastrnak are the top five. As a funny aside, I remember one article on advanced stats years ago talked about the “Pavel Datsyuk test”, as in if Pavel Datsyuk doesn’t rank well in it then you know it’s not a good stat.
The first surprise comes with Alex Chiasson at number ten, but this can largely be explained by him mostly playing with McDavid and Draisaitl on the PP; clearly McDavid and Draisaitl themselves are so high because the rest of the powerplay without them is utter garbage. So yeah, it is partly team dependent in a variety of ways, but overall I’m pretty happy with this stat.
On the other end of the spectrum, there are some surprises. Common sense test is satisfied to see guys like Lars Eller, Sven Andrighetto, Mikael Backlund and Dmitry Orlov in the bottom ten. But I wasn't expecting Charlie McAvoy, Kevin Shattenkirk and Tyler Johnson, although we can imagine some reasons they might be there.
RelSHGA/60 – Same idea as above. This stat measures how many more or less goals a team give up on the penalty kill when a given player is on the ice, per 60 minutes of PK time, specific to 4-5 PK.
This one is a little fuzzier on the common sense test. Penalty killing is a very specific skill that we don’t really have any mainstream stats to quantify besides plain SHTOI, and as much as I glorify TOI, of course coaches can get it wrong sometimes. Iterestingly, the teams who got this wrong and kept playing guys who struggle on the PK were almost universally non-playoff teams.
I would like to hope that this could become the first mainstream stat for evaluating individual performance on the penalty kill.
First looking at some of the worst ranked players (min 50 mins SH): Brian Gibbon ranked the worst; a career AHLer got stuck killing penalties on an awful team because most of the forwards were under 23 after the deadline. Darren Helm and Trocheck ranked 2nd and 9th worst. Maybe the coaches can be excused somewhat here because both guys are there for faceoff abilities and maybe to try to back off opponents with their speed or skill. Kesler ranked 3rd worst, he is there for faceoffs but is old and broken and the coach refused to make a change.
As for the top players in this stat, Charlie McAvoy redeems himself for his powerplay woes, ranking first in the league here. Next we have Ryan Johansen, Erik Cernak, Leo Komarov, Brandon Sutter, Valteri Filppula and Evander Kane. Again, penalty killing is a very specific skill, and until now we haven’t really had any stats to assess individual PK performance, so it is hard to know what to expect, but that all sounds reasonable to me.
Note that this stat is goals against, so a more negative number is better, meaning the player allowed less goals against compared to team average.
RelGF% - This stat measures what percent more or less of the goals a team scores at 5 on 5 when a given player is on the ice. So basically it is adjusted +/-. If a team, on average, scores 48% of the goals at even strength (and their opponents score 52%) and when a given player is on the ice, the team scores 50% of the goals, then that player is +2%. While this is a hell of a lot more useful than +/-, it still isn’t the greatest. Generally speaking, corsi is much more valuable, as luck plays more of a role in being on the ice for a goal scored for or against, as well as the quality of teammates and opponents; whereas an individual player can have more of an effect on driving possession. But as I said, I dont love corsi for valuing defenseman, and adjusted +/- can tell you a lot.
How does RelGF% do on the common sense test? Radulov ranks 1st, Tyler Seguin 4th, and Jamie Benn 14thin the league. When you have such a top-heavy team that relies so much on one elite line, and the rest of the team struggled through injuries and lack of depth, that top line is going to have to come through, and I think the fact that this team managed to make the playoffs is evidence enough that they did indeed come through. Also Teravainen and Aho are 5th and 12th - again, a hard-working, well-structured team that relies heavily on a couple of high-end skilled players. Demers and Copp ranked 2nd and 3rd, and I’m not entirely sure what to make of that. Call them underrated?
This is my least favorite of these stats I’m considering here, but I think it does have some value. Sidney Crosby ranks 7th, so that’s always a good sign.
Then if we look at the bottom ranking players (min 500 mins), everything looks in order. Dubinsky and Riley Nash rank 1st and 4th worst. The other side of the spectrum from Dallas - when you have such a stacked team as Columbus, and you have a couple players having some struggles, of course they’re going to rank poorly here, and Torts rightly played them on the 4th line, 12 and 10 mins a game respectively. Chandler Stephenson ranked 2nd worst, again a mediocre player on a top team. Jack Johnson ranked 11th worst, and any stat that makes Jack Johnson look bad gets bonus points in my book.
One issue with goal-based stats is they require pretty substantial samples to be very useful, or else Lady Luck will rear her ugly head. Corsi, on the other extreme, provides large samples, but I don't find terribly useful for special teams - simply taking shots (without accounting for quality of shot selection) on the PP can lead to loss of possession and/or zone.
So for supplemental stats I've been using relative PP scoring chances for and high danger chances for (RelPPSCF/60 and RelPPHDCF/60) for the powerplay, and relative SH scoring chances against and high danger chances against (RelSHSCA/60 and RelSHHDCA/60) for penalty killing.