For which cases is it appropriate to calculate a probability about a difference in two sample means based upon an approximately Normal sampling distribution?