Read on Twitter
Hey all I interrupt my normal CFB analysis for some NBA stuff. I’ve had this idea for a while but never had the data/skill set to make it a reality until now. I’m calling this metric expected points over average or xPOA. (Thread)
Inspired by xG in soccer/hockey, I wanted to create something similar for basketball which has the unique element of the three point arc. I hadn’t seen anyone create anything like this (at least publicly) so I really wanted to see what it’s like. The results are encouraging
The model is simple and straightforward. It’s a logistic regression model to predict how likely a shot is to be made based on the distance from the basket. Here’s what the model looks like on a half court.
Applying the model to any given field goal, we can calculate the expected points (xP) by taking P(make) * 2 or 3 based on the shot. After this, we subtract the actual points scored (0, 2, or 3) and subtract our xP for our xPOA.
For example, if you take a three from the top right corner of your half of the court, your xP is .3 points. If you make it, you have 2.7 points over average. If you miss you have -.3 points over average.
So how practical is this metric? There’s two ways of representing it: total xPOA and xPOA/shot. Here’s the top 10 on a per shot basis for players in the 90th percentile of FGs taken (800+)
And here’s total xPOA. There’s some issues I’ve noticed. Players who are good but not quite superstars on a good team will benefit from the luxury of not having to take the hardest shots. Conversely, better players on bad teams will have to take harder shots (Spencer Dinwiddie)
I’ll wrap the thread up here for now. This is by no means the be-all metric for how good a player is on the offensive end. This is also just a basic start for this metric so it can absolutely be improved. However I do believe initial results are promising
one thing I want to retroactively add to this thread is what the xPOA distribution for a single game looks like
