Corsi vs xG: Which New Jersey Devils look better in which metric?
Be sure to check out the inaugural Infernal Access merch drop! Hoodies, long sleeves, coffee mugs and stickers are available in plenty of colors and sizes. If you live in the United States, use promo code NICO13 for free shipping. If you don’t, you can get a discount with code HUGHES86.
Today’s guest post was written by C.J. Turtoro. You can find him on Twitter @CJTDevil.
Once upon a time in the hockey analytics world, it was Corsi or bust. Those who were woke enough would use Corsi and lament the mainstream stats like points and plus-minus. A few sizeable innovations later, we now more often use a metric called xG which stands for “expected goals”.
Like Corsi, it is a measure of shot attempts, but it weights them for danger -- rather than 1 unit per shot, each shot is with the probability of such an attempt producing a goal. This is a substantial improvement on the value of an individual shot. As such, most public analytics movement has been towards using that in player analysis over Corsi.
What sometimes gets lost in the shuffle, though, is that most public xG models continue to struggle to improve upon Corsi results when it comes to predicting future goal impact.
Don’t confuse this with saying “xG is useless.” It’s a better predictor of individual goal scoring and therefore is essential to estimating shooter talent, for instance. But when it comes to measuring on-ice impact, it’s fair to view Corsi as on equal (if not, higher) footing to xG.
So, how does this impact assessments of Devils players? Which ones struggle/thrive in each metric?
This article will serve as a summary rather than a deep dive. We’ll split it into 3 categories: players that are similar in both metrics, significantly better in xG, and significantly better in Corsi.
The metrics we will use are single-season, even-strength, 2019-20, RAPM ±/60s according to Evolving-Hockey. Since they are in different units, we will standardize the metric by using its z-score. Z-scores within 0.3 standard deviations will be considered roughly the same. Players who played at least 10 games and are on the 2020-21 roster were considered.
Players who look roughly the same in each
Not a whole lot to talk about here seeing as these are skaters that are almost the same in each metric. We can feel a little more confident about the descriptive assessment of the performance of players in this category given the consilience of the conclusions. Newcomer Andreas Johnsson comes in at roughly league-average along with Nico in a down year. Zacha and Wood show relatively equally poorly in both metrics as well as in the goal RAPM so the bottom-6 feels like where they belong.
Players who look better using xG
Players here will have benefited from the analytical innovations of the last few seasons, but, as mentioned in the introduction, should not be absolved of their Corsi. For many players it doesn’t make a difference. For instance, Bratt goes from being “extremely underrated” to “probably still underrated, but less so” when using Corsi. Others are more substantial, though.
Mcleod, in a small sample, goes from a half a SD above average in xG to a half a SD below average in Corsi. Given his career trajectory, it feels the latter is probably a more likely portrayal of his current talent level.
Palmieri goes from very good to only average -- worth noting given his age and contract status. And Hughes goes from below average to catastrophic -- a potential concern, though, at 18, he’s got plenty of time to figure it out.
Shiny new toy, Ryan Murray looks above average in xG, but has an average Corsi that’s something more in line with what’s been historically expected.
These players all generally will outperform their opponent when it comes to average danger per shot attempt.
It’s worth digging into the granular data, though to see their offensive/defensive splits. For instance, Hughes and Palmieri see a lot of improvement offensively from xG and that feels like it probably makes sense for them -- they do get into dangerous areas. Meanwhile, Butcher’s xG-improvement is mostly defensive -- the shots he allows are generally more benign.
Players who look better using Corsi
The Devils, as a whole, looked slightly better under xG so that is why this list is shorter than the previous one. But the xG underachievements among the players on this list are fairly massive. Gusev has the largest discrepancy on the team finishing with a 0.72 z-score (76th percentile) in Corsi, but a -0.62 (27th percentile) in xG. And it wasn’t confined to his offense or defense -- he was about a standard deviation better in Corsi than xG in both metrics. This will probably be the most interesting player to watch out for, analytically, in 2020-21.
Subban’s underachievement was slightly more confined to the defensive end of things. He was around league-average in shot impact against, but was 1.3 SDs below average (10th percentile) in xGA, indicating that the shots he allowed were generally very dangerous.
Last highlight I’ll point out is new acquistion, Dmitry Kulikov, who makes the cut by substantially underachieving offensively in xG -- WPG’s shot attempts with him on the ice were generally not particularly threatening.
Concluding Thoughts
When a team has a coaching change, it can be a bit of a shock to the system. Sometimes it’s so severe that the previous year's performance ends up being sheer noise. But other times, it just jogs something loose that was lingering beneath the surface anyway.
It could be a trend breaking towards xG-strong, towards Corsi-strong ones, or towards neither. But it’s important to realize that it’s not as easy as saying “this is a player ‘analytics’ likes” -- different metrics show different things.
Now, at least, you have something of a basis for the span of reasonable descriptions for the value of some 2019-20 performances. And that can contribute a valuable filter through which to view the coming 2020-21 season.
The more old data points you have, the less likely you are to be shocked by the new ones.
Think you would enjoy Infernal Access? Now is the time to join! You can get 35% off a subscription using this Thanksgiving Special! Oh, and year-long subscribers get entered in a contest for a chance to win a long sleeve of their choice from my merch drop.
