What is the relationship between the significance level $\alpha$ and the probability of a Type II error $\beta$ if the sample size remains constant?