Adjusting for Zone Starts
One of the biggest omissions in my player rankings is making adjustments for zone start differences. We know that Manny Malhotra has a significant bias towards starting his shifts in the defensive zone and that his teammates Daniel and Henrik Sedin have a significant bias towards starting their shifts in the offensive zone. The result is Malhotra will unfairly be penalized for giving up more shots and goals against simply because he starts more often in the defensive zone and the Sedins have a huge advantage in generating shots and goals because of how often they start their shifts in the offensive zone. The question is, how much of an effect does it have and how do we adjust for it?
Over the past couple of weeks I have been pondering these questions and I thought of two potential solutions to the problem. The first solution is to find some sort of adjustment factor based on zone start statistics. I briefly pondered a few ideas but wondered if a uniform adjustment factor can be fairly applied to all players who have varying skills and talents. I decided that I would take a look at my second idea first.
My second adjustment idea is really a simple idea and really isn’t an adjustment at all. The idea is to just ignore any play that occurs during some stretch of time after an offensive/defensive zone face off. After some length of time, any advantage (or disadvantage) one might get from starting in the offensive (or defensive) zone would be nullified. Worst case scenario is we have to eliminate ~45 seconds after every offensive or defensive zone face off which would essentially nullify the whole shift.
So, with that in mind I took a look at 3 year (2008-09, 2009-10 and 2010-11) 5v5 statistics and did a comparison of four different lengths of time to ignore after an offensive/defensive zone faceoff – 0, 10, 20 and 30 seconds. To evaluate what is going on I looked at each players fenwick for and against per 20 minutes and calculated the correlation between each time after faceoff adjustment. Here is what I found:
|5v5 vs F10||0.8639||0.8451|
|F10 vs F20||0.9882||0.9866|
|F20 vs F30||0.9870||0.9883|
|5v5 vs F20||0.8718||0.8368|
5v5 is no zone start adjustment, F10 is ignoring 10 seconds after an offensive/defensive zone faceoff, f20 is ignoring 20 seconds after and f30 is ignoring 30 seconds after. The numbers are r^2 for fenwick for per 20 minutes and fenwick against per 20 minutes.
As you can see, there is a somewhat sizeable difference between 5v5 and the F10 adjustment but there is very little difference between the F10 and F20 or F20 and F30 and there isn’t really any difference between 5v5 vs F10 and 5v5 vs F20. All of this tells me that any advantage (or disadvantage) a player gains because of their zone stars occurs during the first 10 seconds after an offensive or defensive face off. After that, only the players talent matters and there is no benefit to removing more data from our analysis.
Wanting to confirm this works for a single season of data I decide to take a look at Manny Malhotra and Henrik Sedin’s stats from last season.
|Malhotra FenA/20||Sedin FenF/20|
This confirms what we witnessed with the correlations using 3 years of data. By ignoring the first 10 seconds after an offensive/defensive zone faceoff we can eliminate any benefit/penalty a player may get because of his zone starts. When I finally get around to updating my stats site I intend to include F10 data as well and I think this is a simple enough solution to abandon any attempts at any other zone start adjustment technique.