How is the p-value of a hypothesis test related to type I and type II error?