Note: Modeling Future Record Performances in Athletics

In their article in the December 2008 issue of JASA, Einmahl and Magnus (E–M; 2008) used an extreme-value distribution to predict ultimate world records in track & field events. They also used this method to classify the highest-quality records; that is, records unlikely to be surpassed soon. Unfortunately, the E–M method predicts both ultimate record performances, and the highest-quality records, that are largely at odds with the facts. To demonstrate the problem with the E–M model, I give two examples from each of men’s and women’s track and field events. The authors list a men’s world marathon best of 2:04.26, and predict an estimated ultimate record (EUR) of 2:04.06 with a standard error (SE) of 57 seconds (i.e., a 95 % confidence inter-val (CI) of 2:02.33–2:06.20). They specifically emphasize that of all events, the men’s marathon is least likely to show im-provement. Yet before their article was published in December 2008, Haile Gebrselassie broke his own world record with a time of 2:03:59—an improvement of 27 seconds, and 7 seconds under the EUR predicted by the E–M model. The second example is the EUR for the men’s 200 meters. The E–M model predicts an ultimate best time of 18.63 seconds with a SE of 0.88 seconds. The 95 % CI gives a range of 16.87– 20.39 seconds. The current world record is 19.30, set by Usia