This is a follow-up on the phenomenon of "shot suppression". Weak stomach? Turn away. This might get a little gory. I started with a precision cut and ended up with splattered guts everywhere.
The underlying idea of shot suppression is that teams that dominate puck possession deprive the opposition of puck possession. Puck possession --> shots. Thinking about this over the course of a season, we can measure teams that dominate puck possession with Corsi For (CF), and teams that deprive the opposition of puck possession, Corsi Against (CA). Now, one might test for an inverse association between CF and CA for individual teams over the course of a season, but this was not well-supported (in either 5v5 or 5v5 Close for all teams since the 2007-2008 season). CF did correlate strongly with dCorsi (the difference between CF and CA), or relative puck possession.
So, let us define shot suppression in terms of a high dCorsi value (perhaps relative to other teams). Second, let us state that a high CF value is a necessary factor for shot suppression. (This was a move I did not make earlier. And I think it stands because we would not observe shot suppression without a team that did not have great puck possession figures.) I did this to think about cutting into the data and selecting particular cases. Yeah, here's where things get gory.
Now, my first cut was to take two teams with relatively close, but high, CF figures, and had different other attributes. The two teams (in 5v5 Close play) I chose were Carolina 2011-2012 and Vancouver 2010-2011.
Among all teams in the seasons I examined, these two had the 8th and 11st highest CF values (remember, CF = puck possession). Carolina and Vancouver had CF values of 2428 and 2407, respectively. In addition, relatively same PDOs (1007 and 1012). Yet their dCorsi values were -160 and 328. So, what explains this difference, given the similar other values? Given the similar PDO values, it can't be explained with SH% or SV%. Why would these matter? Well, if you have a high SH%, then your puck possession time would be diminished because you'd be scoring instead of possessing the puck in the offensive zone. SV% seems to matter because you need to stay in Close play. Prime candidate might to be: inability to retrieve and execute a successful zone exit. No data for this (see zone exit project), but it is likely worth a look.
The serendipity of having a PDO column got me thinking more about shot suppression if it were to happen on-ice, and the impact of SH% and SV% on shot suppression. As I was thinking this through, I checked out teams > 400 in dCorsi in the seasons (5v5 Close) examined. There were 10, and only 1 had a PDO > 1000 (St. Louis 2011-2012; sick SV%). My second cut was to look at SH% for each case, here they are ordered by SH% (descending).
|LA 2011-2012||6.35||29||8th West, CUP|
|San Jose 2008-2009||6.65||29||1st West, 1st RD|
|St Louis 2011-2012||6.88||26||2nd West, WCSF|
|Washington 2007-2008||7.15||24||3rd East, 1st RD|
|San Jose 2007-2008||7.47||21||2nd West, WCSF|
|Detroit 2008-2009||7.61||15||2nd West, SCF|
|Detroit 2007-2008||7.75||19||1st West, CUP|
|Pittsburgh 2011-2012||8.05||9||4th East, 1st RD|
|Calgary 2008-2009||8.32||6||5th West, 1st RD|
|Chicago 2009-2010||8.69||4||2nd West, CUP|
Table. Teams with >400 dCorsi, 2007-2008 to 2011-2012.
Couple points here. Note that the top 3 in SH% had sub-par goaltending; 4 out of the 10 are San Jose and Detroit; 3 Cup winners; 7 out of 10 are top 3 in their conference. So, these are teams that performed pretty well. Note how bad LA's SH% was.
And as I'm staring at this, it hits me. What do you actually need to do to get on this list? We need to have high Corsi For values for us to observe shot suppression. We also assume to need good zone exit efficiency. What else needs to happen? Well, either we need to not score so that Corsi For can accumulate or, if the SH% is high, to allow the opposition a high SH% so that Close play can accumulate. In other words, a team needs to not score often for puck possession domination to really show itself. And if the team is scoring at a high rate, the SV% might need to suffer in order for Close play to accumulate so that puck possession domination can be observed. In addition, teams might be scoring quickly to reduce Corsi Against, i.e. SV% might need to suffer. This is all reflected in how I have selected the sample above. I think we can only really "see" shot suppression in these circumstances. Note that teams that demonstrate this are generally pretty darn good, but good teams need not be demonstrate this characteristic.
Bottom line proposition, which I'm not sure we can run a test for: high Corsi For values + high zone exit efficiency + (somewhat low (SH% + SV%)) --> relatively greater time on attack ("shot suppression"). This means that shot suppression is probably not that important. It's ok to see, but you won't miss it when you are scoring goals (both at 5v5 and on the PP). Seeing how things work on the ice and figuring out what the metrics do (and don't do) were really important to working through this.
Stepping away, I thought about the Sharks. Does it matter if they're a shot suppressing team? Does the addition of PDO (broken into SH% and SV%) into the equation change things? Last year, when the Sharks not scoring at a high rate and not accumulating points in the standings, lots of people--including myself--said, "hey man, it's cool, puck possession." I realized that a possession metric (score-adjusted Fenwick) and PDO were the two primary statistics used to generate the Weekly Playoff Probabilities. Those two are probably going to do more than finding "shot suppression" to help me get through the valleys of an 82-game season.
To help with understanding what to do (and I assume Weekly Playoff Probabilities will return), here is a handy 2x2 response matrix. Feel free to cut out and put in your wallet, etc.
[Parting point: there were 8 teams in the seasons I examined who had a PDO > 1005 and CF% > 52; all made the playoffs, 3 were Vancouver teams, and included last year's Cup winner Chicago.]
I miss anything?