Excellent article. But I have found that running speed is better modeled by a lognormal distribution -- it would be interesting to see the same analysis with this model.