I'm finally continuing the posts that I began a few months ago on an adjusted plus-minus (APM) statistic for hockey. Part 1 is here and briefly describes the regression methods I used. The methods are described in full detail here and here.
For the next couple posts, we'll mainly discuss the results, not the methods. Also, from now on, I will include adjusted plus-minus results based on goals (G), like I did in the first post, and also adjusted plus-minus results based on shots (S), Fenwick (F), and Corsi (C).
In this post, we'll focus on offense at even strength...
Let's start with the top 10 players in the league in the offensive component of adjusted plus-minus (APM) during even strength situations, in terms of goals (G) per season, using data from the last 4 full NHL seasons. The column headings refer to goals (G), shots (S), Fenwick (F) and Corsi (C) per season, along with the per 60 minutes versions of those statistics. A player's G result, for example, can be thought of as his goals contributed per season, independent of the strength of his teammates, the strength of his opponents, and his zone starts. The S, F, and C results can be thought of similarly. I should mention that S, F, and C are rescaled using league average shooting percentages at even strength, so that they are roughly the same range as G.
Just missed: Gaborik, H Sedin, Spezza, Getzlaf, Iginla.
Crosby is the man, and leads the league in G (even despite the injuries during 2010-2011 and 2008-2009) and also leads the league in G60. Datsyuk, Ovechkin, and D. Sedin are the next 3, which is no surprise.
Horton at #5 might be a bit of a surprise. However, Horton (teammates' GF/60: 2.43), along with Thornton (2.39) and Parise (2.23), had the worst strength of teammates on this list. Horton's raw numbers aren't top ten level, but when the regression adjusts his numbers, he moves into the top 10 probably because of the strength of his teammates. Of course, if we want to consider his S, F, and C results also, we might bump him down a little bit, since others have superior numbers in those categories. I'd be interested in know what Boston and Florida fans, or anyone else who has watched Horton alot, think about him as an offensive player at even strength.
Ryan might be another surprise on this list, especially given the absence of Getzlaf and Perry. However, Ryan has better raw GF/60 at even strength than those two, and also has worse linemates at even strength, so the rating seems reasonable. According to APM, a lot of Getzlaf's offensive value has been on the power play, where he's one of the top players in the league, as we'll see in another post.
Most of the other players are considered among the league's best offensive players. No defensemen showed up, so here are the top 3 defensemen:
Carle surprised me a bit, but looking at his raw numbers at even strength, and strength of his teammates, this seems reasonable. His S, F, and C results are solid across the board also. Streit isn't a surprise, but the margin that he leads by might be. However, the strength of Streit's teammates (2.15 GF/60) is one of the worst in the league, and is by far the worst among the top 30 or so offensive defensemen, so the rating seems reasonable. Streit, Alex Goligoski, and Andrei Markov had the best per 60 minutes stats, but Goligoski and Markov didn't have enough minutes to crack the top 3 here.
Here's one more list, this time giving the best offensive players at even strength using shots (S).
No huge surprises here. All these guys generate a ton of shots. Overall, I think for the APM results based on multiple years of data, the results based on goals are probably best, at least for offense. The results based on shots, Fenwick, and Corsi have more value when less data is used. For defense, a player's G results depend alot on the goalie playing behind him, so the S, F, or C results might be preferred. We'll cover defense in the next post.
One final note: apologies to Iain Fyffe and Rob Vollman for using the same name Adjusted Plus-Minus (APM) as they used for one of their stats. Rob mentioned in the comments of my last APM post that they wrote an article here about a stat they also called Adjusted Plus-Minus. I didn't know about the article, and would have at least mentioned it in earlier papers and posts had I known. I only chose the term ``Adjusted Plus-Minus'' because that is how the basketball folks refer to this kind of regression-based method.
As always, any comments, questions and suggestions are welcomed.