A bayesian spatial hierarchical model for putting in golf

In this project, a novel statistic to evaluate the putting performance of professional golfers is developed. The methodology provided in this paper borrows ideas from Bayesian spatial statistics to determine the expected number of putts at various green locations. After constructing a Bayesian hierarchical model, the necessary derivation and computations of the relevant full conditional distributions are discussed. Data from the 2012 Honda Classic tournament obtained from the ShotLink website is then used to investigate the approach. The Metropolis within Gibbs algorithm is run to generate samples from the posterior distribution with the ultimate goal of determining the posterior mean of expected number of putts at various green locations. A golfer's performance is assessed against the expected number of putts generated from the MCMC run. Finally, the statistic developed in this paper is compared to total putts and the strokes gained-putting statistic. The results indicate that the difficulty of a putt is influenced by both the distance and region from which the shot is taken.