Bayesian hierarchical modelling in motorsports

Abstract

In this dissertation, a thorough analysis of MotoGP race results between 2016 and 2021 is presented by using Bayesian hierarchical models (BHMs) to obtain estimates for skill variables for riders, teams, and manufacturers. This research in MotoGP, which has yet to be explored with Bayesian models, was inspired by the existing research on other motorsports such as Formula 1. The dataset used for this dissertation includes all race results, filtered to only include riders who have participated in a substantial amount of races to ensure reliability in the results. Two models were implemented using the rjags package in RStudio; one defined on the riders and the teams, whilst the other on the riders and manufacturers. To estimate the skill parameters for the entities in the hierarchy, being rider and team or rider and manufacturer, a logit function on the proportion of riders beaten was used. The results show the abilities of Bayesian models within the context of MotoGP with both models achieving MCMC convergence and outputting reliable estimates for the different skill parameters. Both models also showed good results when it came to predicting rider standings, particularly when considering the top and bottom riders. The team and manufacturer standings were also considered for each model respectively and both showed accurate outputs predicting most of them perfectly or just one place off. The similarity between the results and the Bayesian RMSE of both models is an indication that most of the variability in the results of MotoGP comes from the riders themselves.