Tuesday, 23 June 2009

Double-Dip Tuesday

I've been intrigued with an idea lately thanks to a blog called The Phoenix Pub. Basically, the writer, known as First Derivative, wanted to find a way to evaluate NHL defencemen in a similar vein to baseball's Sabermetric system. At first, I was skeptical that this may work based upon the number of variables that hockey has - giveaways, takeaways, hits, and odd-man rushes as a few examples. However, it occurred to me that defencemen are paid primarily to keep pucks out of their nets. If a defenceman contributes offensively, that's a bonus for the team as they are responsible less for offence as they are for defence. Of course, there are lots of offensive defencemen in the NHL, but I also had to consider the defensive defencemen - the guys who don't appear on the scoresheet often, but play a big role in their teams' successes.

In trying to figure this out, I had to determine what statistical evidence I could use. After all, Sabermetrics is based upon information available, and not some arbitrary statistical analysis. I went to the NHL stats page for analysis.

First off, defencemen are tracked according to points and plus/minus primarily. Points is fairly self-explanatory as to what that means to a team, but plus/minus is a little more difficult. Some people have discredited plus/minus as a "useless stat" since it's more of a reflection of the people on the ice than the player himself. I fully disagree with that since it is a team game, and not an individual effort. Players are sent over the boards in waves, and, if a line doesn't work for whatever reason, everyone pays for the mistakes made. Therefore, plus/minus would be an important part of this equation. After all, teams give up goals, not individuals. Good teams don't give up many goals, and good lines usually have a solid plus/minus for every player on that line.

Secondly, there had to be a way to level the playing field based on the number of games-played per season. Players get injured, can be scratched, or sent to the minors, and all of these affect the total number of NHL games played by the player. Instead, I went with shifts-per-game. A player is counted on to be out on the ice for his team in certain situations, and these situations don't change very often without some sort of major event. It also reduces the variation in games-played between players since the number of shifts per game is relevant to the player's role on the team.

With these major factors out of the way, the next thing needed was a formula to evaluate the players. I decided that all points scored by a defenceman would be offset by his even-strength play. In that regard, points scored could go up or down based upon the addition of that player's plus/minus rating. Coaches talk about the importance of strong five-on-five play, so if a player can keep their opponents off the board while playing even-strength, he would be more valuable to his team than someone who wasn't very good in even-strength situations.

Since all defencemen are paid to keep pucks out of their net while contributing offensively, I also factored in the team's total goals-for and goals-against for each player. This takes into account all situational play: shorthanded, powerplays, and even-strength. If a player doesn't get out on the penalty kill, he is still part of the team if it allows a goal, and he is relied upon to equal the score or prevent it from happening again. Therefore, all goals are factored in.

Here is my formula. I am open to suggestions on this, so don't think this is written in stone. After all, Einstein needed some time to perfect his Theory of Relativity, and I'm not Einstein by any means.

Now let me be clear here. The "Player Value" is not a hard statistic. It will change over time based upon the statistics associated with that player. If, for example, a player has a stretch of two weeks where losses are piling up, his value will decrease. However, if a player has a career year, his value will go up accordingly.

With this in mind, let's take a look at some of the examples I'm working with to give you a better idea of how representative this formula is.

  • Brendan Witt, long considered the worst defenceman in the NHL this past season due to his lack of production and terrible plus/minus of -34, is valued at -0.759. In other words, he didn't have a very good year. In fact, only Thomas Pock of the New York Rangers valued lower than Witt with -0.777.
  • The top player I have found thus far in my examples is Boston Bruins' defenceman Dennis Wideman. Wideman had a phenomenal season with Boston, but was overshadowed by defensive partner Zdeno Chara who won the Norris Trophy. However, Wideman's value is 4.284 as compared to Chara's 3.772 according to the formula.
  • The three finalists for the Norris Trophy - Chara, Nick Lidstrom, and Mike Green - rank as follows: Green (3.982), Chara (3.772), and Lidstrom (3.763). Pretty good values on all three players meaning they were excellent players on both sides of the puck. And isn't that what the Norris Trophy is about?
  • Some of the more intriguing free agents this off-season may not deserve the truckloads of money being discussed for their services. Scott Niedermayer's value last season was 1.920, Jay Bouwmeester's value was 1.345, yet Rob Scuderi's value was a solid 1.878, and Rob Blake was valued at 2.824. Out of those four, I'll assume Scuderi will get the smallest payday.
  • Mark Streit, defenceman for the Islanders and seventh in NHL scoring for defencemen, was valued at 1.621 this past season, yet he came in eighth in Norris Trophy voting. Andrei Markov of the Canadiens was valued at 2.217, and came in seventh in the voting. Duncan Keith of the Blackhawks was valued at 3.292, and came in sixth. And Shea Weber of the Predators was valued at 1.809, yet he finished fourth in voting. Who do you think was the best all-around defenceman this season out of those four players?
  • What's worse about the voting is that Boston's Dennis Wideman finished 11th. Do the Professional Hockey Writers not watch Boston games?
  • Some of the more surprising values are as follows: (1) John-Michael Liles, who finished 30th in defencemen scoring, valued in at 0.603 due to his poor defensive play; (2) Mike Komisarek's poor offensive production valued him at a measly 0.419; (3) Kevin Bieksa, praised for his defensive ability and toughness all season long, only scored a 1.666 value based upon his negative plus/minus. He was slightly better than Mark Streit, even though Streit patrolled the blueline for the worst team in the NHL last season.
  • It is almost a given that Brian Rafalski is important to Detroit's system, but his value of 3.304 shows how important. Lidstrom and Rafalski's average of 3.534 put their defensive pairing right behind Chara and Wideman's average of 4.028 as the best pairings in the NHL.
  • For all the complaining about Brian Campbell's lack of defensive responsibility, Campbell's value was 2.803. That is a better value than Scott Niedermayer, Sheldon Souray, Shea Weber, Dion Phaneuf, and Andrei Markov this season.
So there's a little math work today for anyone who wants to have some fun. Granted, it's not a scientific way of proving one's value, but the idea is that it takes into account both sides of the puck when attempting to give a numerical value for rating players based on their statistics.

Let me know what you think in the comments!

Until next time, keep your sticks on the ice!


FirstDerivative said...

I'm printing this out, and will get back to you later, either with comments or email. Let's work together on this.

TheMetalChick said...

"Brendan Witt, long considered the worst defenceman in the NHL this past season due to his lack of production and terrible plus/minus of -34, is valued at -0.759. In other words, he didn't have a very good year."

No he did NOT have a good year at all but "worst defenseman in the NHL"???????????? He was in no way even the worst defenseman on his TEAM, let alone the NHL. Come on.

Teebz said...

Sorry, MetalChick. That's just how the puck bounces when you're a league-low -34. I'm not saying he was the worst, just the worst of the examples I picked. I'm still looking for other players who were worse. ;o)

Teebz said...

Found one! Thomas Pock of the NYR actually ranks lower. His value of -0.777 was .018 lower than Witt. :o)

windowmaker said...

The problem that I see with your statistic is that it doesn't take into account matchups within a game. Take, for example, Andrei Markov and Brian Campbell. Markov is usually out against the opposition's top line whereas Campbell wouldn't be. This will impact their defensive ratings. Brian Campbell may have a good rating according to your statistic, but the fact that he would rarely be playing against Crosby or Ovechkin whereas Markov would be has to be considered.

I'm not sure how one would go about it, but in order for a more precise statistic you need some sort of 'quality of opposition' measure.

Teebz said...

I fail to see how quality of the opposition would matter when the opposing team is looking for that matchup. If Campbell isn't playing against Ovechkin in Chicago, he would be in Washington.

However, since a player like Duncan Keith would play against Ovechkin and his ranking was HIGHER than Campbell's ranking, my system already accounts for the discrepancy. Keith is paid to keep pucks out of the net, whereas Campbell is paid to generate offence. The results, as proven, make Keith more valuable as a defenceman than Campbell.

Anonymous said...

That seems like you just tossed an arbitrary group of stats together and called it a "value."

Proof of this, is Brian Campbell, who was dreadfully below average for the Hawks this year is ranks above Shea Weber, Sheldon Souray and Andrei Markov.

You're reasoning with this value is COMPLETELY flawed.

Teebz said...

Explain to me how it is arbitrary when it takes goals and assists that the player scored + their plus/minus stat (goals they are on the ice for) multiplied by the total amount of goals scored by the team divisible by the number of times he is on the ice multiplied by the number of times his team was scored on.

Brian Campbell may have been worse offensively than those three defencemen, but his lack of production is reflected in his defensive play and his line's defensive play.

And if it is just arbitrary, how do stolen bases and walks related to slugging percentage in Sabermetrics?

Again, this provides a value based upon the player's PLAY, not his worth. Subjectively, Campbell may not have played as well as the other three, but the teammates around him played well and he was a part of that. Therefore, his value for last season, in which the 'Hawks were successful, is reflected by that.

Again, it's not a scientific evaluation. It's a value based upon his play within his team's structure based upon his stats.

And that's entirely what we're looking here: stats.

JB said...

This is fantastic! Perhaps you could come up with a similar formula for forwards (might help in the pools).