Date of Original Version
Abstract or Description
The International Association of Ultrarunners 24 Hour World Championships holds a 24 hour race during which each entrant tries to run as many laps as they can. There are several different strategies to running a race, and these may be characteristic to the runner. Some runners may race at a consistent pace, dropout, take breaks and/or fluctuate their pace. Records containing how many laps a particular entrant runs over each hour in the race allow us the opportunity to model the different types of strategies and how successful they are. Building on previous work by White and Murphy (2013), we use mixture models, latent class analysis, and model based clustering for mixed data. We alter the model based clustering for mixed data framework, clustMD, that extends the estimation capability of the method in scenarios it was not able to prior. We use this method to determine running strategies in 24 hour races. In future work we plan to develop a longitudinal model-based clustering approach using Poisson processes. This approach will have a large scale impact on other applications outside the field of sports. The same type of model-based clustering can be used in various medical research questions, including clinical depression scores of patients over a period of time.