The evolving value of corsi

My understanding of success in hockey is composed of four factors:

1. Shot metrics (corsi, fenwick, etc.)

2. PDO (SH% + SV%)

3. Special teams

4. Luck

The influence of each of these factors is up for debate. This old piece from PPP found that the most advanced shot metrics constituted about 30-35% of winning in hockey, as measured by point percentage and winning percentage. The same piece found that another 30% could be explained by PP%+PK%. I have seen smarter people than I refer to luck as 30% of hockey, though I am not capable of explaining why or how that number was found. I have not seen any analysis on the relationship between PDO and winning, though I'm sure it exists. In any case, PDO is often one of the ways that luck manifests itself, so a relationship between PDO and winning might not be that interesting anyway.

This piece will focus on the shot metric and PDO influence on winning. Before I go on, I will briefly explain the argument for building a team around shot metrics: shot metrics are repeatable. As the old PPP post finds, shot metrics are "reliable," which means that a team which did well in shot metrics one year has a good chance of being good at shot metrics the next year. In other words, it's more skill-based than luck-based. PDO, PP% and PK% are all extremely unreliable and are very difficult to consistently do well in. Therefore, it makes sense to build a team around a metric which can be repeated reliably.

However, my impression prior to doing this work is that shot metrics are not as strongly correlated to winning this year as they have been in the past. Excellent corsi teams like Edmonton and Carolina are simply not winning like they should be. Good teams like Toronto, New Jersey and Vegas have mediocre shot metrics, or worse.

In order to test the value of shot metrics, I took ten years of data and correlated several 5v5 team metrics (all score and venue adjusted) to 5v5 GF%. All data was taken from Natural Stat Trick. I included metrics like scoring chance goals for%, which I actually do not understand whatsoever, and high danger corsi for% just for the sake of thoroughness. Each relationship is measured by r-squared, or the degree to which the variance in the first variable can be explained by the second variable. In English that means that an r-squared of .20 for the relationship between CF% and GF% means that 20% of GF% can be explained by CF%, and vice versa. The results:

Quick disclaimer—I have no idea how reliable any of NTS's data is. Or how consistently these metrics have been tracked. Any variance in tracking methods immediately invalidates this entire argument.

There's a lot to unpack here. First, it's clear that there was a corsi golden age from 12/13 to 14/15. For those three seasons, corsi explained roughly half of GF%. In the four seasons prior, it was about 30% of GF%. In the three seasons since it's been about 25%. Right now, we're on pace for the lowest performances of CF%, FF% and SF% ever. Meanwhile, PDO has grown to explain 83% of GF%, its best performance ever.

Over the past decade, PDO has claimed a larger and larger share of GF%. There are two possible explanations for this. Either luck has become a larger and larger share of hockey, or teams have become capable of succeeding in percentage hockey in a way which they were not capable of ten years ago. My data does not measure the repeatability of PDO, which is where we'll find our answer. If PDO is as unrepeatable as it was in 2013, then this is likely luck taking more control of hockey, perhaps as a result of greater parity between teams. If PDO is more repeatable, then perhaps teams have put their efforts into taking better shots instead of more shots.

As fans of the Leafs, we have a unique perspective on this. We've seen our team take loads of excellent shots, but not exactly dominate shot metrics as a whole. The difference between Toronto and a team like Carolina is that we've got excellent finish, or SH%. It's reasonable to think that we can continue to finish at a higher rate than other teams because of the elite offensive talent we feature. I have no idea if this is true or not, but I believe it is plausible. However, we're an outlier.

What about teams like Vegas or New Jersey who categorically lack finishing talent? Or Philadelphia, the 8th best 5v5 GF% team, but the 18th best CF% team. It's a similar story for Washington and the Rangers—both very good GF% teams and much weaker corsi teams. All five of these teams have PDOs in the top 13. Which, no big deal. Every year, teams get PDO and outperform shot metrics. The only difference is that PDO is claiming a larger and larger share of GF%. In other words, there are more and more teams who are outperforming shot metrics every year.

Which all leads me back to my original question, are shot metrics as valuable as they used to be? I don't know. Testing the repeatability of PDO in the post-corsi golden age might answer this question, but I do not know how to do that. I think parity can explain the reduced influence of shot metrics, but that begs the question—how does one get an edge anymore if corsi doesn't get it done alone anymore?

This post is as much as an argument as it is an open question to the community. Has PDO become a real and repeatable ability, or is it just luck taking a larger share of hockey? Also please respond to the poll so I can correlate the effect of my writing against the quality of day being had by my audience.