Nature 432, 147 (11 November 2004) | doi:10.1038/432147a; Published online 10 November 2004

Linear models can't keep up with sport gender gap

Weia Reinboud1

  1. Simon Bolivarstraat 87, NL 3573 ZK Utrecht, the Netherlands


Will women runners ever overtake men at the Olympics? Don't hold your breath.


Women sprinters may one day overtake men, according to A. J. Tatem and colleagues ("Momentous sprint at the 2156 Olympics?" Nature 431, 525; 200410.1038/431525a). As the holder of the world and European high-jump records for women over 50, I must say that this statistical prediction has been greeted with much laughter in athletic circles. The authors show linear regression lines and state that this model does not differ significantly from nonlinear approaches. I'm prepared to believe that. But much criticism is possible on both the data set and the logic behind the model.

The data set is very small: only one sprint result per Olympiad, with a large variation in results. Taking the mean of the best 10 per year provides 40 times more data, leads to much less deviation and clearly shows nonlinearity (see

A logical critique goes like this: an athlete can improve greatly by training three times instead of twice a week and can improve further by adding a fourth training session, and so on — but each additional session will give less improvement than the one before. It follows that the sport as a whole will show a similar nonlinear improvement. When statistics, nevertheless, point to linear development, there must be something wrong. Most likely the 'linear' graph in fact consists of more nonlinear parts. For example, one part for the period when athletes were adding ever more training sessions, one part for when they reached a ceiling in adding sessions (around 1980), one part for when drug users were filtered out, and so on.

In which Olympiad will the form of the real nonlinear development become clear? I dare not guess.