One interesting compare-and-contrast exercise going into this next game is that despite a huge gap in standings position, both the Avalanche and Flames have similarly atrocious Corsi ratings at even strength – that is, they tend to give up far more shot attempts than they take. At first glance, they appear to be equally poor possession teams and if you were to cite the main difference between them, it would have to be the level of goaltending each is currently getting. This seems counter-intuitive; how can a team with the top record in the league be playing on the same level as a side with David Jones on its first line?
Corsi (+/- for shot attempts) is considered an excellent proxy for possession, which should make sense intuitively: if you’re taking most of the shots in a game, you probably have the puck more. The theory is that while you win games by scoring goals, too few goals are scored each season is to be statistically relevant and they to be highly influenced by random chance – how many times has a team outplayed its opponent but lost to a bad bounce or hot goalie? Shots are a useful stand-in because they directly lead to goals and teams accumulate several times as many, which eliminates quite a bit of that random chance and produces a more reliable measurement of which teams are outplaying their opponents throughout games.
To make a long story short, a team that dominates shot totals (read: possession) is both generating more offensive chances and playing better defense by keeping the puck out of their own zone. For these reasons, it is generally accepted in the advanced stats community that Corsi is a better predictor of a team’s true performance than goals and even wins. Over a large enough quantity of games, a team’s win and goal numbers are expected to regress to the quality indicated by their possession stats.
So does this mean the Avalanche are a struggling team masquerading as a contender thanks to some exemplary play from their netminders and a schedule packed with games against the weaker Eastern Conference? Not quite – consider that when the score is close (within one goal’s difference) they suddenly break even and jump to the middle of the pack in Corsi percentage. In tied situations, their Corsi is 11th, hanging near elite teams such as the Red Wings, Canucks, and Penguins. In contrast, the Flames continue to languish at 27th-28th in all even strength contexts with the bottom-feeders of the league.
This raises the question of why close and tied situations even matter. If we only consider shots when the game is close or tied, we decrease the sample size and introduce more random noise back into our data. If that was the reason wins and goals are a poor metric in the first place, why do sites like Extra Skater even give them as parameters?
Score effects are a phenomenon where the trailing team will tend to control play due to its imperative of catching up, whereas playing defensively is more favourable for the leading team. It’s driven both by offensive desperation from the losing side and a goal prevention mentality for the team up one, and has a significant effect on shooting numbers. Consider the amount of time the Avalanche have spent leading games this season, it shouldn’t be surprising to learn that they’ve been largely outshot over them. The reason that Corsi close and tied are tracked is to mitigate this and give a less biased representation of a team’s possession.
But even for close situations, the Avalanche have spent a great deal more time leading than trailing. Over the course of a full season, this should even out and unless they’re playing really well or like complete trash, teams will see a balance of leads and deficits over many close games. But as of today we have 14 games of data to work with and the Avalanche have won 12 of them. Our numbers are subject to massive score effect bias and even in close situations are misrepresented.
Corsi tied is the ideal measurement for possession – teams won’t be pressing for a goal or content to sit back with the score knotted, so it’s our best estimate for whether a team is outpossessing its opponents with all else equal. Its inherent issue is that at this point we’ve shrunk our sample size enough that the exact same statistical pitfalls as for goals and wins come in. Corsi close is a better substitute if given enough games to reduce score effects, but that isn’t a luxury that we have at hand. As far as we know, Corsi tied is the most accurate (albeit the least precise) value for how well the Avalanche are possessing the puck.
None of this means that the Avalanche are now definitively the 10th-best possession team in the league, but there is reasonable evidence to suggest that they are above average in that regard and will remain so even once the hot goaltending ends. (Though they certainly aren’t performing as well as they could be if the difference between situations is that pronounced – note that teams like the Sharks, Hawks, and Wild are up there in all three contexts despite similarly obscene goal differentials.)
Again, there is plenty of noise in these numbers so early in the season, and none of these concepts are concrete rules – for example, the Avalanche played from behind for most of last game and their shooting numbers actually deteriorated despite what score effects would predict. And just as wins and goals are expected to regress to the mean shown by a team’s possession, those selfsame possession numbers will over time regress to the team’s actual level of play. But for this year’s Avalanche, said level is a lot higher than what even strength Corsi indicates on the surface. The Flames, on the other hand, are just a crap possession team across the board.