I get that and let me preface this by saying that I am a big fan of numbers/objectivity: one of the reasons I am a huge fan of your articles is because of the numbers-driven analysis that you provide.

That said, statistical numbers need to be taken in context. To use an example of a shot from the slot – averaged over X shots from that area over 1300-odd games a season, yes, they can be high danger shots. But that doesn’t mean a smaller set of 5 shots from the slot in a particular game are necessarily high danger shots. Each shot has significant variance and the smaller your sample size, the greater the variance.

To use a different analogy – if stats say that the average suburban household has 0.8 dogs, that doesn’t mean a street with 10 houses is going to have 8 dogs.

To use a few specific examples from this game:
– the Flyers got off several shots from inside the slot, while having some separation from the defensemen (which is factored into the model). So in theory, shots from a relatively high danger area. Except in atleast 3 of those instances, the D-man got a stick on the shot and deflected it. So in this instance, not exactly as dangerous as they could have been
– There is one shot the Devils took from the point – a seeing-eye wrister that slithered through a whole lot of players, got deflected and required Ersson to make a save that was part positioning/part luck/part reflexes. That, typically, would be a pretty low-percentage chance but in this case, was not.
– There was atleast one instance where the Devils had a great cross-ice pace that had Ersson scrambling across the ice but their forward whiffed. Does that count as a high-danger opportunity?

Without getting into the meat and potatoes of how these numbers are calculated, I think it is fairly clear that each individual “shot from the slot” has a reasonably high variance of danger than implied by the cumulative “danger rating” of that shot. So while these analytics numbers are directionally useful over large sample sizes, it doesn’t make sense to hold on to them as gospel truth with much smaller sample sizes. Subjective assessment can sometimes have a lot of validity as well (especially in small sample sizes, where you dont have to rely on using statistical proxies for “danger” but can actually just count them out).

TL;DR: If you want to have a discussion on the specifics of the high danger chances, or you feel that the analytics tell a more accurate story than subjective assessment in this case, that’s fine.

But less of the dismissive “You can’t just state something incorrect with absolute certainty and make it true” – leaving aside the slight condescension in that (which doesn’t bother me that much), that statement implies a level of .. ahem…absolute certainty.. in analytics which is not be merited.

As an aside that isn’t really relevant to anything we were discussing – one of the models apparently did try to take into account a shooter’s skill level in determining the level of danger in a shot. Interestingly, one of the more popular models that came afterwards stopped doing so. I would have thought that factoring in skill level would refine the model more, but apparently not.

(I forget which models – I had tried to do some research to understand the methodologies but either found something incredible facile or something requiring me to go through tens of citations, which aint happening for a hobby)