Bias in the evaluation of low-magnitude associations: an empirical perspective.