I decided to test a novel methodology for predicting 100 meter and 110 meter hurdler medalists in the the 2017 World Championships. Specfically, I wanted to builder a model that predicted both placing and the probability of medaling. I decided to predict these odds based on each athlete`s results during the 2017 track and field season. In order to test the validity of this approach, I built a model for each World Championship year from 2005 to 2017. I scraped the IAAF website for all 2005-2017 outdoor results (https://www.iaaf.org/records/toplists/sprints/). I followed a rather basic approach to predicting medalists: I fit linear models to the data using performance during a given season (sans World Championship results) as my predictor variable for that season`s World Championship medalists. I fit a different model for each season. I fit a binomial generalized linear model (GLM) to predict the probability of a given athlete medaling in a given year. I did try fitting one binomial GLM for all the data, however, when I applied the coefficients to predict medalists for one season (2017), the model overestimated probability of medaling for each athlete because it assumed there were 21 medal positions (3 medals x 7 years).
© Copyright 2018 The World through data. All rights reserved.