Following the 2004-2005 lockout, the NHL introduced a variety of rule changes intended to increase scoring and bring fans back to the rink. One of these changes which prevented a team that ices the puck from changing their defenders, was aimed at stopping the practice of pinned-in teams from sending the puck down the ice simply to get a line change and a fresh set of legs out. The idea being that giving the offensive team a chance to change and face a set of tired defenders would create a significant offensive advantage. Defensive zone faceoffs are the last place you want to start a shift, with Michael Schuckers estimating that starting in your defensive zone is worth approximately -0.0055 goals. The question that naturally comes up from this idea though is whether there is an incremental cost to a team icing the puck, and what that incremental cost is. Is the penalty for icing the puck simply the defensive zone faceoff, or does the fatigue factor play a significant role?
Using the NHL’s Play-by-Play files from 2010-2013, I identified all icing calls and who the offending team was by looking at the faceoff immediately following the icing and inferring the offending team from the faceoff zone and faceoff winner. After that, I followed Schuckers THoR methodology and calculated how often the team that iced the puck scored a goal in the next 20 seconds (see (a) below) and how often their opponents scored a goal in the next 20 seconds (see (b) below). I should note that this is a major simplification of Schuckers regression model which takes into account numerous other factors when determining the value of an event, but for the purposes of coming up with a rough estimate it should do.
|(a) Icing P20||
|(b) Icing Against P20||
|(c) Icing NP20 ((a)-(b))||
|(d) Zone Start||
|(e) Icing Fatigue Cost ((c)-(d))||
With these two numbers, we can then calculate the net probability of a goal from the point of view of the team that iced the puck, given in (c) above as -0.0090. What that’s saying is that each time a team ices the puck, they’re essentially costing themselves -0.009 goals in the long run. The cost of icing the puck though is made up of two factors though: the cost of taking a faceoff in the defensive zone, and the cost of having tired players on the ice. We know that a zone start is worth about -0.0055 goals from Schuckers work that I mentioned above, so the fatigue cost of an icing is simply the difference between the Icing NP20 (c) and the zone start cost (d), or -0.0035 goals.
This may not seem like a lot, but it’s roughly 3 times the per-play value of home ice advantage as given by the THoR model. What tempers the impact of icing is that icings are a relatively rare occurrence-within our sample there were an average of 6.7 icings per game, which isn’t game changing, but over the course of a full season a team that consistently ices the puck could see this effect add up. Jan Hejda was on the ice for 50 net icings in 2011-2012, which cost his team nearly half a goal between the zone-start and fatigue effects.
Now obviously our estimate doesn’t tell the whole story-there may be some gain to the team that iced the puck that we’re not capturing, as teams often ice the puck when they’re under a lot of pressure in their own zone. But it does illustrate the importance of accurate passing when starting on the attack, and how a missed connection when a team is moving up the ice can end up being more costly than just the lost offensive opportunity.
What about controlling for shift time prior to the icing? That is, is there in fact a greater fatigue effect the longer the players have been on the ice when the icing occurs?
That’s a very good point. I suspect that at least some of the cost is related to the fact that players who ice the puck are also more likely to be towards the lower end of the talent spectrum. I’ll see if I can pull something together to take a look. The trick may be figuring out how to quantify the fatigue effect. I think Gabe Desjardins looked into the effect of shift length on Corsi, but I’m not sure if anyone’s examined if it has any effect on shooting percentage (which I suspect that it may).
Thanks, I’ll have a look at Gabe’s stuff. Will be interested if you pull anything together.
A slightly related aside, are you aware of anyone having looked at “penalty adjusted corsi”. ie. including shots during a powerplay drawn/taken by a player when calculating corsiFor/Against?
I don’t know of anyone who’s looked at it although it’s always struck me as a bit of a gap that analytics tends to focus so much on even-strength play. Factoring it into Corsi For/Against might be tricky because the shot attempts on the PP/PK aren’t necessarily the same value as one’s taken at even-strength due to the differences in shooting percentage.