Analytics, Coaching Tactics, and Observable Results

Today the Carolina Hurricanes officially announced that Eric Tulsky has been hired in the position of “Hockey Analyst” with his role being defined as follows:

As Hockey Analyst, Tulsky will provide and analyze data to assist the hockey operations department and coaching staff.

The official announcement also mentioned that Tulsky had worked for the team last season on a part-time basis which we already knew. For me the interesting thing here is that we learn that Tulsky will be working with the coaching staff and thus I think it is safe to assume that he probably did some of that last season too. We do know that a significant portion of Tulsky’s analytics work has been related to zone entries and exits which really gets to the heart of coaching tactics.

We also know that Tyler Dellow worked closely with the coaching staff in Edmonton, especially Dallas Eakins (who was a major reason for Dellow’s hire in the first place). We also know that Kyle Dubas in Toronto is a big promoter of “puck possession” hockey and this likely influenced coaching tactics and playing style with the Leafs last season, especially after Randy Carlyle was fired.

I asked twitter earlier tonight whether any of the other analytics hires from last season had direct or indirect influence over coaching tactics and playing style (as opposed to just consulting in front office decisions) and as of yet no one has come up with a conclusive example. This leaves Carolina, Edmonton and Toronto as the best examples of where analytics may have had a significant impact in on-ice playing style and tactics which is where we might first see analytics having an impact (takes much longer to change roster than playing style).

Back in January I looked at the relationship between possession (Corsi%) and Shooting percentage (or Corsi Shooting% to be more precise) in 5v5close situations and found that there is an inverse relationship and I even estimated  the impact the Oilers improved Corsi Percentage (CF%) would have on their Corsi Shooting Percentage (CSh%). This is what I found:

So, their 5v5 CSh% has improved from 43.4% to 48.7%. If we plug that 5.3% improvement into the regression equation above we would expect that their CSh% would drop 0.65% where it actually dropped 0.89%. Edmonton dropped from 11th in CSh% last season to 27th this season.

We can now presume that the Leafs and the Hurricanes playing style may also have been influenced by “puck possession” analytics and we can see how it impacted their statistics as well as looking at full season stats for the Oilers. Here is what I find.

Edmonton CF% Sh% CSh%
2013-14 43.40% 8.25% 4.38%
2014-15 47.40% 7.16% 3.74%
Difference 4.00% -1.09% -0.63%
Expected Difference -0.49%
Carolina CF% Sh% CSh%
2013-14 49.20% 7.39% 3.78%
2014-15 51.70% 6.38% 3.36%
Difference 2.50% -1.01% -0.42%
Expected Difference -0.31%
Toronto CF% Sh% CSh%
2013-14 42.10% 8.24% 4.46%
2014-15 45.20% 7.52% 3.99%
Difference 3.10% -0.72% -0.47%
Expected Difference -0.38%

What is interesting is that each of these teams saw their puck possession stats increase (CF%) and their shooting percentage stats decrease (Sh% and CSh%). We also see that the expected change in CF% based on their change in CF% was fairly reasonably predicted by the negative relationship I found in January though in all three instances it underestimated the drop in CSh%.

Now, this is just three teams and it may just be random luck but it is quite an interesting observation and I have shown that there does seem to be an inverse relationship between CF% and Sh% when teams change coaches. Really, this all makes sense because if you take more risks offensively you might give up the puck more (hurting your possession stats) but you might also generate higher quality scoring chances if those risks pay off (boosting your shooting percentage). And of course the opposite is true if you play a more conservative offensive game.

Anyway, I found today’s announcement that Tulsky will be (and thus probably did last season) working with Carolina coaches an interesting one as it provided an additional test case.