|
Activity Number:
|
19
- Bayesian Methods for Sports Data
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 3, 2020 : 10:00 AM to 11:50 AM
|
|
Sponsor:
|
Section on Statistics in Sports
|
|
Abstract #309305
|
|
|
Title:
|
Bayesian Prediction of Performance in Professional Athletes
|
|
Author(s):
|
Gilbert Fellingham* and Richard Warr and Garritt L Page
|
|
Companies:
|
Dept of Statistics, Brigham Young University and BYU and BYU
|
|
Keywords:
|
hierarchical Bayes;
nonparametric Bayes;
prediction
|
|
Abstract:
|
We test the predictive ability of various Bayesian models using two sets of professional sports data. The first data set contains the scores for a group of 22 members of the Professional Golf Association (PGA) in 2014. The other data set contains home runs per at bat for major league baseball players with at least 100 at bats in the 2015 season. We fit six different models with the intention of determining which predicts better in these two disparate data sets. We varied model complexity across two different dimensions. In one dimension we fit model intercepts using parametric Bayesian, nonparametric Bayesian, and hierarchical Bayesian methods. In the other dimension, we either included covariates for each sport or we did not include the covariates. We then use these models to predict performance in the following season for the same golfers/players as well as other golfers/players. Results indicate that nonparametric Bayesian methods seem marginally better, and using covariates to predict outcomes performs more poorly than expected. We extended these ideas and found that prediction is enhanced by using a covariate dependent random partition model.
|
Authors who are presenting talks have a * after their name.
