The last couple of years have been eventful for the hockey analytics community and hockey analytics has definitely gone main stream. The past year or so has seen some interesting developments including more investigation into shot quality (Scoring Chances by war on ice, expected goals from @DTMAboutHeart, etc.) , some interesting enhancements in goalie evaluation, and the passing project led by Ryan Stimson has shown some interesting results and could provide many more interesting insights into the game. All that said there is one area where I hope will get more attention from the analytics community in 2016.
Impact of Coaching and Player Roles
I have shown that coaching can have a huge impact on team statistics and as a result on player statistics. This is especially true for Corsi which I feel is significantly playing style driven, maybe more so than actual player talent (though talent is certainly a component as well). In player evaluation we can mitigate coaching impact somewhat by looking at relative statistics but by not fully understanding coaching impacts it still makes comparing players across teams difficult.
Furthering the complexity of player evaluation is that players are given different roles on their team and this affects how they play and what their statistics are. In my opinion the hockey analytics community has done a terrible job at understanding the impact of roles on a players statistics. One prime example of this has been how the analytics community has failed to understand the relationship between zone starts and possession numbers. If you do a correlation between zone starts and CF% you will find a very strong relationship between the two where players with more offensive zone starts generally have much better CF%. This led many in the community to conclude that zone starts are a major driver of CF%. The problem is while there is correlation there isn’t causation. It isn’t the zone starts that are driving the possession numbers it is a combination of player talent (specifically offensive ability) and role (defensive players -those who get defensive zone starts- have a primary goal of not allowing a goal against rather than generating a goal for so the offensive component of CF% suffers). Zone starts are an indicator of role and roles are a significant factor in driving CF% and thus zone starts correlate well with CF% but are not a direct driver of CF%.
None of this is a problem if you want to evaluate past performance but it is hugely important if you want to evaluate player talent. Let’s take James van Riemsdyk who went from having one of the worst CF% ratings in the league among forwards (315th, 44.6%) last season to one of the best in the league this season (19th, 56.4%). JVR isn’t more talented this season than last season, he is just playing under a different coach with different line mates with a different role.
I believe that coaching and roles can affect the percentage stats as well – shooting and save percentages. The Maple Leafs under Mike Babcock have significantly improved their possession statistics but their shooting percentage has dropped significantly. Under Randy Carlyle the Leafs had terrible possession numbers but their shooting percentage was among the league leaders. Under Mike Babcock (and Peter Horachek as well) the possession numbers have improved significantly but their shooting percentage has dropped down to the bottom third of the league.
At the individual level I believe roles drive the percentages as well, including save percentage. Players who are given offensive roles, especially if they are given offensive freedom and are less constrained by being requited to play a defensive role, typically have the best on-ice shooting percentages.
More controversial is that I believe players given more pure defensive roles often show an ability to boost their goalies save percentage. In the summer I identified Brandon Sutter as a defensive center who has shown the ability to boost his goalies save percentage. You want to know something? He is doing it again this year in Vancouver posting a Sv%RelTM of +2.3. This is an example of where the players role is to limit goals against which results in poor possession numbers (45.7% over last 4 years, 291st in the league) but far better goals against numbers (-0.49 GA60 RelTM, 17th in the league).
I believe coaching and roles can have an impact on every aspect of hockey analytics and yet very little research has gone into it. As an example of what I am talking about, today “Garik16” had a post on hockey-graphs.com about neutral zone play. In it he talks about the importance of winning the neutral zone and how controlled entries into the offensive zone produce more shots than dump-ins. There may be something compelling here but I only come away wondering how coaching and/or player roles are playing a factor. Are the observations found driven by talent or driven by roles or more likely what combination of the two. Do offensively talented players who are assigned offensive roles carry the puck in more while defensive players whose primary role is to prevent goals (not score them) dump the puck in more and maybe not even press for puck recovery after the dump in? If we factored this into the analysis how much would our conclusions change? All of this matters because ultimately if we are trying to evaluate player talent we need to know how much coaching and role affects statistics (traditional, enhanced or tracked).
I don’t believe we fully understand how coaching and roles impact a players statistics and the relationship between shot generation/suppression and shooting/save percentage though I am certain a significant relationship exists. In 2016 I hope this is an area of hockey analytics that will gain more attention. It will definitely be an area of interest for me.