I was reading Statistical Inference (2nd edition) by George Casella & Roger L. Berger when I came across a formal definition of p-value, followed by a theorem stating that

Let W(X) be a test statistic such that large values of W give evidence that $H_1$ is true. For each sample point x, define $$p(x)=\sup_{\theta \in \Theta_0} P_{\theta}(W(X) \geq W(x))$$ Then, p(X) is a valid p-value.

My question is what if the test statistic is the one whose small values give evidence that $H_1$ is true? Is the $p$-value defined as $\inf_{\theta \in \Theta_0} P_{\theta}(W(X) \leq W(x))$ in that case? Thank you a lot.

  • 2
    $\begingroup$ That's an awful definition. It seems to define "a valid p-value", so presumably that contrasts with an invalid p-value. $\endgroup$ Commented Aug 1, 2018 at 4:57
  • 1
    $\begingroup$ If by "small" you mean "very negative," then replace $W$ by $-W$ and apply the definition. For other concepts of "small" it's likely a similar procedure will work. $\endgroup$
    – whuber
    Commented Aug 1, 2018 at 12:53

1 Answer 1


No, you would still use the supremum, so in that case the p-value would be:

$$p(x)=\sup_{\theta \in \Theta_0} P_{\theta}(W(X) \leqslant W(x))$$

  • 3
    $\begingroup$ The intuition here is that we’re considering the worst case scenario: what’s the biggest our probability could be given any parameters? Then we know that the real probability of getting that small of a statistic is bounded above by that. $\endgroup$
    – Removed
    Commented Aug 1, 2018 at 6:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.