Goaltending is voodoo. Far and away, projecting goaltender impacts has been the most difficult piece to pull together in my models. My first attempt left me with some significant questions since the performance metrics of the model were much lower than I would like. Looking through the projections, there were some that seemed to align with the eye test but some seemed suspicious.
Robin Lehner of the Vegas Golden Knights was the poster boy for the projections that just didn’t seem right. While Lehner hasn’t been consistently in a starter’s role for the past three seasons (where the historical data used is from), he certainly has shown he is a capable NHL goaltender. His projection didn’t agree and is a big part of why the Vegas Golden Knights team projection was down with the re-building California teams.
The New Goaltending Model
I tried a few different types of regression models to find a better solution to the goaltending projections. Ridge regression has provided the best results so far. Using data from the past three seasons (as before), input variables of 5v5 GSAx/60, 5v5 SV%, and 5v5 HDSV% have provided a much improved model from the original for predicting 5v5 GSAx/60. The mean absolute error of the new projections is 0.24, which is a marked improvement from the previous model at 0.3. The R-squared value is up to 0.24, again much higher than the old model.
Top Projected 5v5 GSAx
- Chris Dreidger
- Adin Hill
- Connor Hellebuyck
- Igor Shesterkin
- James Reimer
With the new model, we see some familiar names in the top five projected 5v5 goals saved above expected rates. This gives some additional confidence that the model is working reasonably well.
We also see marked improvements for some of the goaltenders who has questionable projections in the previous model. Robin Lehner is in this category, along with Jonathan Quick and Dan Vladar.
Updated Goaltender and Team Projections
I’ll be rolling out the updated projections over the next week or so. The goaltender charts will be updated along with the full strength roster plots as teams hit the 10 game mark and I adjust the full strength rosters based on this season’s actual player usage. Team projections will be updated in time for next week’s regular season update post.
