Why the null matters: Statistical tests, random walks and evolution

摘要: A number of statistical tests have been developed to determine what type dynamics underlie observed changes in morphology evolutionary time series, based on the pattern change within series. The theory ‘scaled maximum’, ‘log-rate-interval’ (LRI) method, and Hurst exponent all operate same principle comparing maximum change, or rate dataset expected a random walk. Less than walk has interpreted as indicating stabilizing selection, while more implies directional selection. ‘runs test’ contrast, operates sequencing steps, rather excursion. Applications these computer generated, simulated series known dynamical form various levels additive noise indicate that there is fundamental asymmetry II errors excursion: they are highly sensitive models selection result linear trend but largely immune case simple model Additionally, LRI method lower sensitivity originally claimed, due large range rates produced by walks. Examination published results show seldom conclusion an was which needs closer examination light asymmetric response tests.