Online Program Home
My Program

Abstract Details

Activity Number: 169
Type: Contributed
Date/Time: Monday, August 1, 2016 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistics in Sports
Abstract #321283 View Presentation
Title: Improved Expected Points and Win Probability Models for NFL Coach and Player Evaluation
Author(s): Gregory Miller* and Gabrielle Flynt and Sam Ventura and Andrew Crossett
Companies: Bucknell University and Bucknell University and Carnegie Mellon University and West Chester University
Keywords: EPA ; WPA ; Expected points added ; win probability added ; National Football League ; NFL
Abstract:

Expected points added (EPA) and win probability added (WPA) are common approaches to evaluate the effect that individual plays/players have on the outcome of National Football League games. However, popular existing models used for EPA and WPA can be improved upon, both in calculation and in use. Many of these models are limited in scope by their assumptions, in that they negate the use of key situational data ranging from the 2nd & 4th quarters to instances of large score disparities. In observing other sport literature, (e.g. McCurdy, 2014 - hockey) where controlling for score situation and using data from all parts of the game has proven advantageous, we have used all data from all situations to build our model. Using our revamped EPA and WPA metrics, we provide an empirical evaluation of NFL coach decision making, independent of actual outcomes of plays they call. Similarly, we provide an easily interpretable measure of NFL player value, grounded in common currency of points/wins, for both offensive and defensive players. With these two statistical advancements, we hope to offer insight as to how NFL organizations can use this information to evaluate coaches and players alike.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2016 program

 
 
Copyright © American Statistical Association