Ultimate Records in Athletics using Extreme Value Theory

This paper addresses two questions: What is the ultimate record in a specific athletic event? and, how good is a current athletic world record? Data consist of all the personal bests of athletes in 28 events, 14 for both men and women. As an ultimate record is considered as a finite endpoint of an underlaying distribution, Extreme Value Theory and corresponding estimators are used to determine those endpoints. A simulation is performed to show the accuracy and consistency of the different estimators. After checking the finiteness of the endpoints, the ultimate records are calculated. The quality is measured by deriving its limiting distribution. Results indicate that some events have enough room for improvement (men’s marathon, 3 minutes), while other current world records lay close to the ultimate record (women’s shot put, 7 cm). Moreover, some current records have a high quality (women’s marathon), whereas other records are likely to be sharpened in the near future (men’s 110m hurdles). As extension, it is tested if the outdoor long jump data are contaminated by measurement errors, as the data do not satisfy initial assumptions. Results indicate that this is indeed the case, with the wind as possible factor that contributes to the measurement errors.