Modeling Spatial Expected Points Allowed by NBA Defenses
STAT479 Group 28: Paddy Lavin, Owen Klabough, Barnabás Valkó, Steven Yang
Introduction: Beyond the Box Score
There are a great number of stats to measure offense in basketball, but not for defense
Traditional counting stats like steals and blocks are rare and do not reflect true defensive ability
Defensive metrics often ignore spatial context
Our Goal
To create a metric for analyzing team defensive efficiency that takes into account the location of shots taken.
Our Approach
We’ll use a Generalized Additive Model (GAM):
Smooths over the variance seen in binning to provide a continuous surface of expected points per shot taken.
Less susceptible to noise, captures the broad spatial trends of a defense.
We can use GAMs to create a new metric, EPx, representing the Expected Points a defense allows per shot taken at a point x.
The Code
We use the mgcv package in R to fit our GAMs based on the score_value (0, 2, or 3) of any shot taken as a function of its x and y coordinates on the half court. We use play-by-play data accessed via hoopR:
