Friday, October 11, 2013
Adjusting NHL’s Plus/Minus, part 6 of n: Locutus and Borg
The 1986 Flyers were quite the team, lead by Mark Howe and Brad McCrimmon, two defensemen who would end up with very long NHL careers. Their goalies were also excellent, as their save percentage would indicate they saved 54 goals (44 Froese #2 in the league, 7 Jensen, 2 Lindbergh RIP, 1 Chico). During plus/minus situations, let's presume that was 40 goals saved.
The Flyers were +94 overall, but +74 in plus/minus situations. Since we figure the goalies were +40, that means the rest of the team was +34.
So, the first thing we need to do is reduce every player's plus/minus because of the goalies. If a player played 40% of the time, he'd get a benefit of that +40 of their goalies for a total of +16. If he played 35% of the time, we need to knock out +14 from their plus/minus.
Mark Howe was +85 and Brad McCrimmon were +83. They were defensemen. The next highest defenseman was a 0 (Marsh), -2 (Richter) -5 (Crossmam), -12 (Erickson). Clearly, Howe and McCrimmon were paired all season long.
First things first, we knock out the benefit they got from Froese et al. Assuming Howe, McCrimmon played 40% of the time and Crossman 35%, we get:
Plus/Minus after goalie adjustment:
+69 Howe
+67 McCrimmon
-19 Crossman
As we did with Pederson and Middleton, let's see what we get if we assume our two giants were randomly paired, or exactly paired. Going through the same math, we get:
Random pairing
+88 Howe
+85 McCrimmon
Constant pairing
+42 Howe
+40 McCrimmon
That is an enormous range! That kind of range, 45 goals(!) is enough to stop us dead in our tracks.
Even worse is poor Doug Crossman. If we assume he played 35% of the time, his unadjusted -5 would convert to -19 after the goalie adjustment, and further down to -42 on the idea that he was randomly paired with any of the other D, including Howe and McCrimmon.
What do we do next? Well, that all depends on balancing ease of adjustment and accuracy of results. While we'd love to get an approach that would lead us to Howe being a +42 on the 1986 Flyers, and we can't accept him being +88, how high a number can we live with in terms of being wrong, just so that we can implement something that is not so convoluted a black box that requires manual determination of who was paired with who, and for how much they were paired? Can we live with a process that makes Howe +52? +63? +75? Or just live with the +88 (meaning a process that assumes random pairing)? And what of poor Doug Crossman?
And then, we go back to Ray Bourque, where we presume he was randomly paired in 1984. If he really should be a +51 with the longer more correct process, can we live with +41? +30? That's on the idea that he was likely paired with better than average Bruins.
?
Real good stuff - don’t know where to comment, but I’ll pick here.
In adjusting for the goalie, shouldn’t you split out between defense and offense and assign the adjustment to the defense only portion?
If you have player A who plays lower minutes but is a high scoring player and player B who plays many more minutes, shouldn’t we adjust Player B more?
Why don’t you add back the goalie adjustment, and then get an offense and defense rating for the team? The goalid is only affecting defense -