Around 85% of the game is played at even strength, with most of that being 5 versus 5 situations. This is why that special teams, while important, are not very crippling and 5v5 play is the strongest predictor of future success currently known. We can see this in last season with Chicago Blackhawks being one of the worst special teams performers, yet the second best 5v5 team in the league.
Corsi, Production... and Arby?
The best measurement we currently have for creating team success by an individual is Corsi, which is simply a differential between shots directed towards the opponents net versus shots directed towards your own while said player is on the ice. Usually it is then either divided by icetime or be placed as a percentage to even the icetime difference between top players and fourth liners. The strength of Corsi is not due to every shot attempt being crucially important, but rather that shot volume is a side-effect of players seeking shot quality. Shot differentials actually become a greater predictor of future goal differentials than the traditional goal +/- itself. Many tests have been conducted (yes actual experiments using hypothesis, not just data-mining) where Corsi has been shown to very closely correlate to attack-zone time, possession time, scoring chance differentials, and ultimately scoring.
Mike Babcock once said "puck possession is everything" and in reality he is right. However, there is something to be said about players who consistently produce points better than another player with the similar possession metrics.
I wanted to show how players do in both scoring AND in possession. So, I invented my own statistic, even though really it's nothing but a very simple algorithm of the other two statistics.
Method: We will take the players 2011-13 Corsi differential per 60 minutes of play but only in close game situations (game within two points), removing most of the effects of teams going into a defensive shell when leading. We then remove the first 10 seconds after a puck drop, eliminating most of the effects of faceoff wins and zone start usage. Thus far, nothing revolutionary or not done at Arctic Ice Hockey before. We then take a player's 2011-13 5v5 points per 60 minutes of play and combine them with one fifth of their Corsi differential.
Method to the madness: The reason for the Corsi differential being only a fifth is partially arbitrary but there is some soft reasoning to it. In 5v5 situations there are five players for each team on the ice, so in a way there are five players responsible for the team's success. Now it isn't true that all five are equally responsible, as certain positions have been shown to affect Corsi differentials more so than others. Also, it has been shown how you can have drivers, anchors and riders of possession. However, the trends are real and it becomes a decent estimate.
Naming the stat: Because the line in possession I created is kind of arbitrary, I originally named it "arbitrary". I noticed thought when placing the short form "arb" in my spreadsheets that it looks like fellow AIH writer's "Arby" handle... so I changed it to Arby... Doesn't seem to weird since most underlying stats in hockey are named after people anyways (Corsi, Fenwick, PDO).
Without further ado, I present to you Arby 1.0:
* To quote Tim (AKA: Truck) when he first saw these: weird but it doesn't suck. For the most part it tracks along the lines of what I think about the players and their usefulness and my eye-test.
* Wellwood and Burmistrov, as usual, are shown to be players that are highly underrated for their accomplishments for the team.
* Frolik is a pretty solid acquisition, and while Setoguchi isn't the best, he beats the crap out of Miettinen for a 2RW.
* I wouldn't be surprised to see Frolik's against and Setoguchi's for numbers jump up a bit based off of who they will likely be mostly playing with and change of teams.
*Halischuk's has arguably had some of the worst possession linemates in the league... that will likely change a bit.
* Slater's numbers will always be bottom barrel, since face off specialists have two choices: get pinned in their own zone or line change.
* Wright, Miettinen and Thorburn are not very good players at hockey. Their numbers are similar to Slater's without the specialist excuse.
Not considering context for statistics is usually what causes people to misrepresent or not believe in what they are explaining to a person. This image coms from a great Tyler Dellow article showing how a players results should aways be taken in consideration of their usage (like teammates, zone starts and competition) and also which line they are on. Here is an image from that wonderful article (which you can read here):
Other than arbitrarily assigning a value in the balance between Corsi and scoring, the biggest factors missing are quality of linemates and competition. We know that competition is an important factor, but due to the fluid motion of the game and coaches only getting last line change in 50% of games, competition effects get minimized throughout the span of a season. Teammate effects however are a big deal, and highly controllable by coaching.
We can't really compare teammate effects between different teams that well, other than say Frolik was on an elite team, Setoguchi's team was similarly strong as the Winnipeg Jets and Halischuk was on a terrible team. However, we can compare within the Jets.
Here is the 2012-13 QoT between Jets' players (in same order as above):
|Name||QoT||Δ std dev from mean|
* This displays one of the faults in #NoeLogic, with giving prime-time to Miettinen and Jokinen over better players like Burmistrov, Antropov and Wellwood.
* The Burmi-Noel rift is no excuse to, as Antropov and Wellwood were also better options.
* I don't know if I should feel sorry for Wright being with Thorburn all the time, or Thorburn being with Wright all the time... how about neither?
* I do know I feel sorry for Kane carrying 1-2 anchors all season
While it's kind of silly, I think in the future I'll run some regressions with P/60 and Corsi, to try and find a real algorithm that works. In the future we can then evaluate a players combined contributions to scoring and possession, while currently we're only looking at a rough estimate. I'd most likely use Adjusted Corsi by Michael Pakartti once he improve past their current infancy. He is using an algorithm based off of regressions to adjust for quality of teammate and competition.
This could be a step towards truly finding the best players in the league relative to their ice time. It kind of sucks that the Jets have lost two of their top 5 guys, but in honest truth, if they weren't to be used optimally used anyways their loss is minimized.