Garik16 from Hockey Graphs, Lighthouse Hockey, and Islander Analytics wrote a good piece today on the defensive shell (a topic that’s been on my list to address for a while), following up on David Johnson’s initial look into the subject a couple of years ago. I highly encourage you to read both stories, but the general conclusion from today was that the shell doesn’t actually help a team because the opponent’s scoring rate – what you’re trying to minimize – actually increases. I had a couple of thoughts on the issue, because while the material is interesting, I don’t necessarily agree with the conclusions.
First of all, it’s important to note that score effects are the result of a combination of four very different forces (more on this in a future post).
1. Players naturally playing harder/more aggressively when trailing
2. Coaches coaching trailing players to push ahead and take risks
3. Players naturally being risk averse and not going their hardest when leading
4. Coaches coaching leading players to make the safe play (ie a contain or prevent defense)
Generally speaking, items 2 and 4 are those that the coach can impact. A coach can tell his players to keep pushing up by a goal late in the third period (they generally claim that they do, and players often echo this notion) but items 1 and 3 will still naturally institute score effects, and therefore the idea that getting hemmed in one’s zone leading late is simply a poor strategy that should be discarded – like the 1-3-1 forecheck or the overload power play – is misguided.
Of course, there IS an element that is coaching-driven, but the problem you run into here is that the impact of such strategizing will tend to get swallowed in heaps of variance, since you’re adjusting one of the four factors, but the other three stay the same (the other team won’t, for example, suddenly agree to not take chances since you’re still pushing ahead).
So when I see something like this graph, where the author is using shooting percentage Down 1 as a proxy for results against a defensive shell, I worry that the intended effect isn’t being isolated.
We know that the shooting percentage on average Down 1 is greater than Tied, but if one removed items 1, 2 and 3, leaving only the strategized intentional defensive shell, would that still be the case? Maybe, but we can’t say that from just this data.
AC Thomas’ more precise graphs on shooting percentage as the game goes on don’t rid this analysis of confounding variables, either.
Sure, the shooting percentage when trailing is shown to be higher than that with the score tied right up until the end of regulation. But there’s a pretty logical reason for that as well. Teams tied near the end of regulation are playing for overtime. They are far more likely to dump shots from the blue line than to pinch to create opportunities. And both teams are far more worried about not allowing a goal against than scoring a goal for. There is massive loss aversion in play here; shooting percentage is bound to drop. You can’t compare the two situations because looming overtime (or more specifically the loser point) is a confounding variable.
So yes, scoring rates are higher against teams protecting leads than against teams late in tie games, but that doesn’t mean that the defensive shell, as a strategical maneuver, is responsible for that. Impending overtime, as well as natural factors that are difficult for coaches to account for, could just as easily be responsible.