Ramblings: A new goal model; John Gibson, Linus Ullmark, and Cory Schneider – August 15

Michael Clifford



Be sure to head to the Dobber Shop to get your copy of the 2019-20 Dobber Hockey Fantasy Guide! It has everything that fantasy owners need from projections and rankings, to depth charts and prospects, over- and under-valued players, and a whole lot more. And it’s updated all throughout training camp! Get to our store and grab your copy now.


I wanted to share a very exciting piece of research done recently by Alex Novet over at Hockey Graphs.

Hockey analytics started out by counting shot attempts. It feels weird to say that it took the general hockey world about 100 years to figure out that generating more shots than your opponent is a good thing, but innovation in sports has lagged behind almost across the board. Anyway, that was a big breakthrough, and about 10 years after that we got our first expected goals model. These models are effectively another counting model, but they count the probability of each shot actually going in based on historical scoring rates on similar shots from similar locations rather than just the shot itself. It was another improvement in analytics and a big step forward.

Now, that piece from Novet starts to cover one weakness of expected goals models and that’s pre-shot movement. For example, expected goals models count a wrist shot from the right faceoff dot as all effectively the same, even though some are just wasted shots while some are cross-ice one-timers that can have an open net. Both shots don’t have the same probability of going in but are treated the same in expected goals models. That’s where Novet’s research comes in.

He takes the data from Corey Sznajder’s data set – he does much of the public game-tracking on things like pass locations, zone entries/exits, and a lot more – and creates a model that tries to incorporate basic pre-shot movement based on pass totals and pass locations. Now, as Mr. Novet points out, the data set is far from complete, but it’s a start in the right direction in preparation for whenever the NHL actually rolls out the player tracking technology and we can be more precise with these things. It’s an exciting time for hockey stat heads.


Let’s talk about goalies for a second.

Realistically, I don’t like talking about goalies very much. The reasons for this are as follows:

  • It is a very random position. It is a very important position in fantasy (one of the most important in all fantasy sports, really), but it’s still random.
  • It takes a long time (several seasons) to know how good a goalie is, and even after several seasons to evaluate, there’s still a margin