© 2013 Michael Parkatti BOTB12-5

Rating NHL Defencemen

I’m a huge user of Gabe Desjardins’ Behind the Net site, which offers a vast array of advanced hockey statistics for nerds like me to pore over and tinker around with.  One area I’ve been tinkering around with is the concept of shot and goal differentials and how you can use them to rate a player.  This piece deals with defencemen, which I intend to extend to forwards at some point.

So what can we learn about defencemen from shot or goal differentials to form an opinion about a player?  Well, one traditional stat is plus/minus, or how many of his team’s goals a player is on the ice for versus how many goals the opposing team scores while he is on the ice.  This stat has been looked at suspiciously for years by the advanced stat community, as it does not account for variables like team quality, linemate quality, competition quality, etc.  Each little second of ice time in the NHL is not created equal, and the game is much more complicated than +/- accounts for.

I had an idea for a quick and dirty way to rate defencemen based on goal differential, but making it relative to team performance.  I used the following formula:

Goal Differential vs Team = (goals for on-ice/60 mins – goals against on-ice/60 mins) – (goals for off-ice/60 mins – goals against off-ice/60 mins)

I’m taking their goal differential while they’re on the ice, and then subtracting their team’s goal differential while that player is off the ice.  If a player’s goal differential while he’s on the ice is equal to his team’s goal differential while he’s off the ice, this statistic ends up being zero.  If a player has a good goal differential while his team has a bad goal differential when he’s off the ice, this stat will be high.  If a player has a low goal differential while his team has a good goal differential when he’s off the ice, this stat will be low.

Here are the top 20 players by this metric using last year’s data.   There were 174 dmen in my sample, including those playing more than 30 games and averaging more than 14 minutes of ES icetime:


To come up with the stat in the last column, you simply take the difference of the first two numbers, and then minus the difference of the last two.  Kris Letang had an amazing year last year with this stat, and you can see why: when he was on the ice, his team outscored the oppostion 4.14 to 2.56 every 60 minutes of even strength ice time, while when he was off the ice the Penguins were actually outscored 2.68 to 2.48 for every 60 minutes of ice time.  The Pens are a different (and much better) team when Letang is on the ice.  Stated another way, if Letang could somehow play an entire 60 minute game of even strength play, the pens would outscore their opponents by 1.78 more goals than if Letang did not play at all in this theoretical game.

There are some usual suspects here for players commonly considered the best defencemen in the league (Karlsson, Weber, Subban) but also some oddballs like MA Bergeron and Andy Sutton.  I’ll admit right off the top that this does not control for quality of team or quality of competition, which I hope to work on in the coming days.

Here’s a list of the 20 worst NHL defencemen by this metric:


The only real surprises on this list are perhaps Edler, Staal, and Kronwall, whereas defencemen like Derek Morris or Ryan Whitney (for us Oiler fans) really do match the perception of their play.  The Canucks actually get outscored when Edler is on the ice at even strength, but they killed teams when he was sitting on the bench. What the hell?

A common critique of plus minus is that it uses fairly extraordinary events (goals) to judge a player, and that there is another basis (shots) to rate players using similar logic but a much larger sample size.  This is called Corsi (when you include shots, missed shots, and blocked shots) or Fenwick (just shots and missed shots).  I will use Fenwick here.

Using similar logic as above, I took the shot differential for each player and subtracted the team’s shot differential when that player was off the ice.  Here are the top 20 players by this metric:


Some similar players are visible from the goal differential chart above, with some new faces.  All 3 Norris finalists are here in Karlsson, Weber, and Chara, along with a top 5 dman of all time in Lidstrom.  But have a look at TJ Brodie at the top of this list — with him on the ice, the Flames out-Fenwicked opponents 37.3 to 36.7, but when he was off the ice, the Flames were outshot 46.9 to 35.1!  Again, I haven’t controlled for quality of competition or teammates, but there are players on this list that I’m sure play against top competition (Chara, Weber, and Lidstrom).

Here’s a list of the 20 worst NHL defencemen by the shooting metric:


Again, some similar names from above, but some new laugh-worthy entrants in Jack Johnson and Luke Schenn (among others).  This shows that with Marc Staal on the ice the Rangers were outshot 39.8 to 34.3, but when Staal was off the ice the Rangers wiped the floor with their competition 41.2 to 35.5.

I do intend to see how we can standardize this a bit more, but I’d love it if any of the analytical community would comment here or send comments to parkatti@gmail.com.

One Comment

  1. Bruce McCurdy
    Posted January 24, 2013 at 5:58 pm | #

    Very interesting line of thought, Michael, in fact to a degree we share parallel tracks. This morning I made this customized list of BtN 4v5 stats for Kings d-men, to support a comment I was making over at Lowetide’s site about Matt Greene.


    I deleted a bunch of the Corsi columns as so much noise, but left together paired ON/OFF rates for GA and SA, with the thought that both shed a little light on each guy’s success as a PKer. One of those things where if the same guy is at or near the top of both categories, it’s less likely to be a fluke than one category alone. Or so goes my logic.

    That said, T.J. Brodie? The smell test for your lists above is how reasonably do any of them do at casting the best 20/worst 20 d-men in the circuit. Especially the guys on the bottom, you may find a lot of role-play playing a role. :) If the guy starts out in his own zone against Crosby, chances are he’ll have poor #s even if he’s pretty darn good.

    The other thing to be wary of in all such within-single-team comps is the balance or otherwise on the roster. Used to be you could do a comp stat like this that “showed” that the Triple Crown Line was better than the Gretzky-Kurri-Firehydrant line, when the real issue was LA’s 2nd, 3rd & 4th lines sucked when comnpared to their equivalents on the Oilers. Not saying that invalidates it, just that by limiting to within a single team you are in effect limiting the sample size of your comparables.

    I need to think about this some more, but let’s start there.

Post a Comment

Your email is never published nor shared. Required fields are marked *


You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>