2
$\begingroup$

My understanding is that the size $\alpha$ of hypothesis tests is defined as

$$ \alpha = \sup_{\theta \in \Theta_0} \mathbb{P}_\theta (X \in R) $$

where $\Theta_0$ is the subset of the parameter space associated with the null hypothesis $H_0: \theta \in \Theta_0$ and R is the rejection region such that

$$ \forall X \in R, \text{ reject } H_0 $$

So, in words, the size of hypothesis tests is the supremum of the probability to make type I error, i.e. rejecting $H_0$ when it is true. Correct?

$\endgroup$
2
$\begingroup$

Broadly yes, but $\Theta_0$ is not "the null hypothesis", but contains the set of parameters for which the null is true.

For example, when you test a composite hypothesis about the mean that $H_0:\mu\leq0$, then the null is true for any nonpositive $\mu$. The definition of the size now says that (a little heuristically in view of the proper definition of a $\sup$) when we compute the type-I error probabilities for all these nonpositive $\mu$, the largest of these probabilities, so the maximum type-I error probability, is $\alpha$.

$\endgroup$
  • 1
    $\begingroup$ I edited the question to exactify the definition of $\Theta_0$ to "$\Theta_0$ is the subset of the parameter space associated with the null hypothesis $H_0: \theta \in \Theta_0$". So what does the qualifier "broadly" mean here? $\endgroup$ – user78219 Nov 27 '15 at 7:30
  • $\begingroup$ Broadly referred to that you broadly got it right. $\endgroup$ – Christoph Hanck Nov 27 '15 at 7:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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