Only thing I can think of is that if you pick a statistic with fewer confounding variables you have less of a chance to cherrypick?