Stan model to fit ranks to match data like Bradley & Terry (1952)?



Is there a Stan model to demonstrate how to fit tournament data (e.g., pairwise match outcomes in a chess tournament) to obtain rankings on an absolute scale?


@Bob_Carpenter has some code in the example models repo:

stan-dev/example-models: knitr/bradley-terry

And there are blog posts I came across once but never actually read (so I can’t say if they cover exactly what you’re talking about). Maybe they’re useful:

Fitting Bradley-Terry models using Stan

The Bayesian Bradley-Terry model with draws




Warning—that’s not very well developed. I just banged it out for a friend of mine’s kid who was competing in a robotics tournament and asking about how to rank teams made up of multiple member (I just assumed abilities were additive on the team on the log odds scale).

I do go over the basic models with simulated data, too.