Estimating NBA Free Throw % with Hierarchical Models

In this post I explore 3 different formulations for modeling repeated Bernoully / binary trail data: complete pooling where all items have the same chance of success, no pooling where each item has in independent chance of success and partial pooling where data across items are shared to estimate parameters. To do this I use an example where I try to estimate the batting average of baseball players, with inference in Numpyro. In a repeated Bernoully / binary trial, our data consists of $n$ units...

Submit