Is possession hockey providing diminishing returns?

Ian Cooper sent me a link to an article he recently wrote how hockey analytics has driven the importance of puck possession hockey which in turn can change the dynamics of hockey analytics. In his article he showed that in recent years there has been a smaller spread in team CF% talent which he believes is due to a higher percentage of teams focusing on puck possession hockey.

That’s more consistent with a league in which teams are all deploying tactics in order to optimize their SAT% and pummeling a small number of holdouts who either don’t accept analytics orthodoxy or simply aren’t good enough to compete. In that case, you would expect fewer teams at the high end and a bigger cluster around the middle.

Essentially as teams focus on puck possession hockey it becomes less of an advantage to do so. This is something that I wrote about several years ago.

Furthermore, if General Managers as a whole started paying primarily for corsi we will start to find that corsi talent becomes more evenly distributed across teams and thus shooting percentage would become much more highly correlated with winning (even after adjusting for luck). Furthermore, paying players based on corsi would potentially lead to players altering their style of play to optimize their corsi statistics to the detriment of the ultimate goal, out scoring the opponent.

I want to see if we are seeing a decrease in importance of puck possession hockey and an increase in importance in shooting percentage. To do this I took three groups of 3 seasons.

  • Group 1: 2007-08, 2008-09, 2009-10
  • Group 2: 2009-10, 2010-11, 2011-12
  • Group 3: 2013-14, 2014-15, 2015-16

I have not included the 2012-13 shortened season in any group as it was a short season and likely has some unique qualities associated with it.

Now for each group I was able to create two separate Year1 vs Year2 comparisons. For group 1 this would be 2007-08 vs 2008-09 and 2008-09 vs 2009-10.  For each of these year over year comparisons I calculated each teams change in CF%, GF% and Sh%. This gave me 60 sets of year over year comparisons for each group. (Note: all stats in this post are 5v5close stats to minimize influence of score effects)

Because shooting percentage can be influenced by randomness in an attempt to reduce the effects of randomness I sorted the 60 year over year comparisons by change in CF% and then created subgroups of 5. So subgroup 1 would be the 5 teams with the largest change in CF%, subgroup 2 would be the next 5 teams with the largest change in CF%, etc. This gave me 12 subgroups per group with average changes in CF%, GF% and Sh%.

I can now plot these to get an idea of the relationship between changes in CF% with changes in GF% and Sh%. Here are the plots for group 1.


You can see from the charts that changes in CF% resulted in significant changes in GF% and there is a high positive correlation between the two. Changing CF% only had a relatively small negative impact on Sh%. Improving CF% appears to matter a lot back then.

Here are the charts for group 2.


The slope in the CF% vs GF% chart has dropped considerably though the relationship is still fairly strong. For CF% vs Sh% the negative relationship has grown some.

Now for group 3.


Look at that. The relationship between CF% and GF% has shrunk even more while the relationship between CF% and Sh% has grown.

Let’s look at the slope and R^2 for each group in a summary table.

CF% vs GF% Slope CF% vs GF% R^2 CF% vs Sh% Slope CF% vs Sh% R^2
Group1 0.8171 0.7291 -0.0546 0.1131
Group2 0.5569 0.5835 -0.0924 0.2235
Group3 0.4207 0.3542 -0.1211 0.5924

This table indicates that in 200708 to 200910 an 1 point increase in CF% would be expected to translate into a 0.82 point increase in GF% and a -0.05 point decrease in Sh%.

In recent years a 1 point increase in CF% can only be expected to increase GF% by 0.42 points and the corresponding drop in Sh% is 0.12.

It is still good to improve your Corsi but we are hitting the point of diminishing returns. In fact if we just look at last season to this season the negative relationship between CF% and Sh% is even stronger (slope=-0.1717, R^2=0.8627) while the relationship between change in CF% and change in GF% is non-existant (slope=0.0197, R^2=0.0008). That that trend holds there may be no benefit to valuing CF% over shooting and save percentage.

These observations could be a sign that hockey analytics is working. Several years ago analytics showed that puck possession hockey was undervalued by teams and teams are now giving puck possession its proper weight. Of course it could also mean that teams are starting to give puck possession too much focus and any benefits they are seeing from a better puck possession game is getting cancelled out by worse shooting percentages (and possibly save percentages). Teams really need to seek to improve puck possession with as minimal an impact on the percentages as possible. We have seen several teams improve Corsi in recent years only to see their shooting percentage drop significantly. I’ll be interested in revisiting this in a year or two to see how it continues to progress.

As I wrote before, Corsi and puck possession should not be the goal. Scoring and preventing goals should.


This article has 1 Comment

  1. Hi David,

    Interesting analysis. The idea of diminishing turns over time in regards to possession hockey had not occurred to me before. The data certainly seems to indicate that at this point the goal of GM’s should not to be to sign as many players as possible with high CF%, as the time for that has passed for most teams (except for maybe Colorado and a couple others…)

    I wonder if a new stat (or existing stat I don’t know about) will come about that combines CF% and SH% in a meaningful way, such that we can identify which players could improve a team’s CF% while at the same time not adversely affecting the team’s SH%.

    I also think it would be an interesting investigation to see whether teams who improved their CF% from one season to the next also saw a higher percentage of shots coming from low and medium danger locations, and a lower percentage of shots from high-danger locations (based on the slope of the CF% vs SH% graph, I would imagine this is the case).

    A final thought is will the analytics-savvy management groups in the NHL identify this trend of diminishing returns, and will high SH% players begin to be valued more compared to high CF% players? (And will the superstars of the future be the players with the highest combined CF% + SH% ?)

    Thanks again for the analysis, David.

Comments are closed.