The Coaching-Corsi dilemma

The other day I wrote about Bozak-Corsi dilemma which basically goes as follows:

  • The coaching change in Toronto from Carlyle to Horachek resulted in Tyler Bozak and the rest of the Leafs top line posting dramatically improved Corsi (5v5 tied CF%). Does this mean Bozak et al. suddenly got good or does it mean that Corsi is largely driven by playing style which one can change and thus the value of Corsi in player evaluation is greatly minimized.

Today I will look at the rest of the Leafs 5v5 Tied CF% from Carlyle to Horachek as well as three other coaching changes that occurred during the 2011-12 season. Those are Bruce Boudreau to Dale Hunter in Washington, Randy Carlyle to Bruce Boudreau in Anaheim and Terry Murray to Darryl Sutter in Los Angeles. Expanding the analysis to more players/teams will determine whether the Bozak-Corsi dilemma can be expanded to a more general Corsi-Coaching dilemma. First, lets summarize how the team 5v5 tied CF% changed due to these coaching changes.

Coaching Change CF% Pre CF% Post Difference
Kings – Murray to Sutter 50.7 58.5 7.8
Ducks – Carlyle to Boudreau 42.6 48.7 6.1
Leafs – Carlyle to Horachek 45.2 51.1 5.9
Capitals – Boudreau to Hunter 56.7 48.1 -8.6

The biggest positive impact was with the Kings while the Capitals change from Boudreau to Hunter had the biggest negative impact and the biggest change overall. All four coaching changes saw significant impact in the teams overall CF%. Let’s look at the teams in order listed above starting with the Kings.

CoachingChangeCFPct_Kings201112

Shown here are each players CF% under Murray (in Blue) and under Sutter (in Orange) with the difference shown in the grey bars. Shown are players with at least 50 5v5tied minutes under both coaches. The first four players saw their CF% jump by at least 10% and the next four by at least 7.5% and 11 of the 14 players saw their CF% jump at least 5%. Only Drew Dougty saw his drop but he already had a team best 59.3 CF% under Murray.

Here is the chart for the Ducks.

CoachingChangeCFPct_Ducks201112

Every single player saw their CF% jump at least a little after the coaching change. Visnovsky’s and Getzlaf’s CF% jumped at least 10% while Perry, Sbisa and Lyudman jumped at least 7.5% and Selanne and Cogliano at least 5%.

Now for the Leafs.

CoachingChangeCFPct_Leafs201415

JVR, Bozak, Rielly and Kessel saw at least a 10% boost in their CF% while Polak and Gardiner were at least 8%. No other player saw a jump of more than 3% and Komarov actually has a significant (10.8%) drop off.

Now, for a reversal of fortunes here is the Capitals chart.

CoachingChangeCFPct_Caps201112

For the Capitals every single player saw at least a drop of 3.9% (in fact only Wideman saw a drop off of less than 5%) with the first 6 guys seeing a drop of at least 10% and three more at least 9%.

In total there are 53 players in the charts above, 17 of them saw an absolute change of at least 10% while another 13 saw an absolute change of at least 7.5%. The average change was just shy of 8%. By looking at these four coaching changes it is safe to say that it is not unusual for a coaching change, or a change in playing style, to impact a players 5v5 CF% by 10% or more (nearly one third of the players above saw that big of a change). If a normal range for 5v5tied CF% is between 40% and 60% I think it is safe to suggest that half or more of that spread might be due to playing style and not individual talent. Furthermore there are almost certainly different playing styles on a single team (some lines certainly play more defenisve roles while others play more offensive roles) so even looking at CorsiRel stats might not factor out all coaching decisions. It certainly appears that Kessel-Bozak-JVR have seen a far more significant boost in their CF% relative to the rest of the team indicating that they likely change their playing style the most.

Above I looked at four coaching changes which had an average absolute impact of an 8% change on 5v5tied save percentage with nearly one third the players having an absolute change of greater than 10%. The majority of NHL players end a season with a 5v5tied CF% of between 40% and 60%. Based on the above analysis it is probably reasonable to believe that at least half of that spread can be attributed to variations in coaching/playing style which means the actual talent spread is probably no more than 45% to 55%, possibly even less.

Furthermore, I have previously shown that FF% (and thus likely CF%) loses predictive ability over longer periods of time at the team level. A significant reason for this is likely the higher number of coaching and roster changes that occur over a 4 or 5 year span. Every coaching change and every time a player changes teams (or even the line they play on) can potentially lead to a playing style change which could impact their CF% significantly. Of course, none of this should really come as much of a surprise as we already know playing style can have a major impact on CF% because we know all about score effects. On average a teams 5v5 CF% when they are leading is about 10% higher than their 5v5 CF% when they are. This 10% difference in CF% due solely to playing style dictated by the score lines up fairly well with what we have seen above where a 10% change in CF% due to a coaching change is not abnormal. The Corsi-Coaching Dilemma is real.

What all this means is that we need to consider playing style when we evaluate players because playing style can have a major impact on a players statistics. In fact, it may be the most important factor in a players Corsi statistics. This is something that we rarely do in analytics but failure to do so could result in a very flawed player evaluation. This is something the hockey analytics community really needs to address in future research.