Math isn’t always the worst! (Image courtesy of Boing Boing)

‘Stats-ing’ With Steve, Part II: Expected Goals, a Deeper Understanding of Shots

Diving deep into Expected Goals to find out how it helps us learn about goal scoring

Folks, hello. The first installment of this series involved Corsi, a proxy for puck possession measured by counting all shot attempts. The only problem with Corsi is that it treats every shot attempt as equal. It’s still an extremely useful tool and simple to understand, so it’s used often (it’s usually the first statistic I check when looking to find out about a team or player, for example). It just doesn’t always tell the whole story.

That’s where Expected Goals (xG) come in. An idea first introduced in hockey by Brian MacDonald, @DTMAboutHeart developed an improved version based on Michael Caley’s version for soccer. DTM has made a few adjustments to his own model (here and the most recent here with the help of Asmean), and Emmanuel Perry built a model similar to his (updated here) on his site:

xG is intended to combine shot volume and shot quality into one number to determine the likelihood of more goals in the future. For this article, I will focus on Perry’s xG model because it is the most readily available and easiest to find and cite (DTM hasn’t updated his site with xG numbers since the 2014–15 season, and while he does dump much of his data periodically on Twitter, it is harder to find).

It uses the elements to determine this, according to Perry’s model, are shot type (wrist shot, slap shot, deflection, etc.), shot distance (adjusted distance from the net), shot angle (angle in absolute degrees at which shot was taken), whether or not the shot was a rebound, whether or not the shot was a rush shot (or off the rush), and the strength state (5v5 vs. power play).

By combining these elements, we can get a better idea of how well an individual player or a team has been performing in terms of how many goals they should be scoring. If you want to view how a team is doing in xG, the quickest way is to take it as a percentage, like with Corsi. You can go to and whether you look at teams or individual skaters you would look at the xGF% column. This stat, like Corsi, can and should be adjusted for score and venue.

The development of this stat is important because it says more than “take more shots and you’ll score more goals.” That’s a true statement, but we know not all shots are created equally — in fact, that’s the central pillar of xG. For example, shots taken down low (in the “high-danger” areas) have a higher likelihood of going in the net than do shots from the outside (in the “medium-danger” or “low-danger” areas).

Unfortunately, at this time Manny’s model is not a better predictor of future goals than CF%, but it is more descriptive of past play in terms of its correlation to goals. The model developed by DTM and Asmean does have a higher correlation with future goal scoring, but again, recent data generated by that model can only be found on Twitter. Manny has said that he plans to update his xG model and will surely publish his findings on his site.

Well folks, I hope you found this informative. My next foray into this series will likely concern hockey WAR (wins above replacement, derived from GAR, or goals above replacement), a stat inspired by baseball’s WAR. I look forward to writing the next installment.