Statistics of Extremes in Athletics

Abstract

TV shows on any athletic event make clear that those who want gold medals cannot dispense statistics. And the statistics more appealing to champions and coachers are the extreme order statistics, and in particular maximum (or minimum) values and records. The models in statistics of extremes are usually semi-parametric or even non-parametric in nature, with the imposition of a few regularity conditions in the appropriate tail of the unknown model underlying the available data. The primordial parameter is the extreme value index, the shape parameter in the (unified) extreme value distribution. The estimation of the extreme value index is one of the basis for the estimation of other parameters of rare events, like the right endpoint of the model underlying the data, a high quantile, the return period and the probability of exceedance of a high level. In this paper, we are interested in an application of statistics of extremes to the best personal marks in a few athletic events. Due to the way data are collected, we begin with a parametric data analysis, but we pay special attention to the semi-parametric estimation of the extreme value index and the right endpoint whenever finite, the possible world record, given the actual conditions. In order to achieve a better decision we consider a few alternative semi-parametric estimators available in the literature, and heuristic rules for the choice of thresholds.