In the previous article, I found some differences in the shot tracking data depending on the arena. I therefore defined the Arena Effect as the difference between shot quality at home and on the road. Those Arena Effects can be downloaded here.
Now, I want to add these effects to my sGAA model, which you can read more about here. It’s basically a model combined of GAR, xGAR and GSAx from EvolvingHockey.com, that I have adjusted so sGAA correlates directly to goal differential on the team level.
Adjusting the goaltending:
Let’s start by Arena-adjusting the goaltending. To do so, I will need the home ice fenwick against for every goalie. I have found this through the shot data from MoneyPuck.com. Then I will simply adjust by multiplying FA(Home) and the Arena Effect.
Here is the top 10 goalies (career) before the adjustment:
And here’s the new top 10:
We could also look at per 100 fenwick data. Here’s that data (FA>5000):
Henrik Lundqvist is hit pretty hard with these adjustments, but I already discussed that in the previous article. Based on longevity and workload he has probably still been the best goaltender over the last 10+ years. I’m more surprised to see Thomas Vokoun near the top though. He posted really good numbers in a tough environment in Florida.
Adjusting the skaters:
Things get a bit more complicated when we look at the skaters. We saw in the previous piece, that Arena Effects don’t really have an impact on the offensive components of sGAA, so I will instead turn my focus towards the defensive components (sEVD and sSHD). I also want the overall team sGAA to stay the same, since the current model correlates really well goal differential:
Therefore, I want the adjustment to a team’s goalies to equal the adjustments to the skaters. So, if I add 10 goals to the goalies, I need to subtract 10 goals from the skaters as well.
I’m using on ice fenwick against at home (from NaturalStatTrick.com) to determine the adjustment. After some refinement I got to the following adjustments:
sEVD adj = 0.1845 * FA(home, even strength)
sSHD adj = 0.2606 * FA(home, shorthanded)
These adjustments are added to sGAA model, and the updated data can be downloaded as an excel workbook here.
Obviously, goalies are more impacted by Arena Effects, but here’s the most impacted skaters.
Biggest positive impact:
Biggest negative impact:
Unsurprisingly, we see players from NYR and NYI (high xG) gain most from these adjustments, and we see players from T.B and MIN (low xG) decrease the most. This is in line with the idea that skater adjustments should offset goalie adjustments.
We can compare Performance Charts before and after the adjustments, to illustrate how the Arena Effects impact teams. Updated Performance Charts can be found here. The x-axis is the year the season ended and the y-axis is the estimated goal differential based on the sGAA model.
Here’s NYR before the adjustments – offense is sEVO, sPPO and sDraw – defense is sEVD, sSHD and sTake:
And here’s the Performance chart with the arena adjustments:
The new data shows that those early NYR teams were a combination of good defense and good goaltending. According to the old data it was all Henrik Lundqvist.
We see the opposite effect with Tampa Bay:
I didn’t want to change the overall team sGAA, since it correlates really well with the goal differential of the team. So, the changes I’ve made are only at the individual player level. I’m very confident that the new arena adjusted sGK is a better goaltender metric than the old one – which was derived from GSAx on Evolving-Hockey.com.
It makes sense that the defensive metrics are affected by the Arena Effects, but you would also expect the offensive components of xGAR (part of my model) to be affected.
xGAR consists of 3 sub-models: Shot rates, Shot quality and Shooting ability. Clearly, Arena Effect should impact the Shot quality, but at the team level that effect is offset by an inverse effect on the Shooting ability. However, this doesn’t mean the effects are offset on the player level. You would expect shooters to be affected in one direction and non-shooters/playmakers in the other direction.
Steven Stamkos is probably a good example of this. He plays in Tampa Bay with the lowest Arena Effect (low xG) and he shoots a lot. He therefore scores more than expected giving him a high Shooting component. Stamkos is great shooter, so he would likely have a good Shooting component no matter what, but I think it’s increased by Arena Effects.
You could perhaps adjust for this by having iFF (individual fenwicks for) pull in one direction and on ice fenwick for pull in the other. However, I won’t adjust for offensive effects at this time. For now, it’s just something to be aware of, when analyzing the data.
It’s important to know that sGAA primarily is a descriptive metric. It’s a useful tool for player and team evaluation. The next step is to define projected sGAA (p-sGAA), which will be the predictive counterpart to sGAA. This is very similar to Dom Luszczyszyn’s approach with his metrics: GSVA (descriptive) and GS (predictive).
Predicting hockey results is very difficult – Especially when it comes to goaltending. I hope the arena adjustments above increases the predictability slightly.
Stats from Evolving-Hockey.com, NaturalStatTrick.com and MoneyPuck.com