"... in an experiment, there are sort of two kinds of exercises. In one, you might collect more data and bring down your s. In another, you have a concern, what we call a , and you check it, whether it exists at all or not."