The definition in the book I am reading says the following:

A test with power function $\beta(\theta)$ is unbiased if $\beta(\theta') \geq \beta(\theta'')$, $\forall \theta' \in \Theta_{0}^{\complement}, \theta'' \in \Theta_{0}$.

Now, $\beta(\theta)=\mathbb{P}_{\theta}(\mathbf{X}\in R)$ where $R$ is the rejection/critical region.We also know that:

$\mathbb{P}_{\theta}(\mathbf{X}\in R|\theta \in \Theta_{0})=\mathbb{P}(\text{Type I Error})$, and
$\mathbb{P}_{\theta}(\mathbf{X}\in R|\theta \in \Theta_{0}^{\complement})=1-\mathbb{P}(\text{Type II Error})$

That means we can also write:

$\beta({\theta'|\theta' \in \Theta_{0}})=\mathbb{P}(\text{Type I Error})$, and
$\beta({\theta''|\theta'' \in \Theta_{0}^{\complement}})=1-\mathbb{P}(\text{Type II Error})$

And from these, would it be right to say that a test is unbiased if

$$ \mathbb{P}(\text{Type I Error}) + \mathbb{P}(\text{Type II Error}) \leq 1 $$

If yes, how do I interpret that?


2 Answers 2


To ensure that $\beta(\theta_1) \geq \beta(\theta_0)$ for all $\theta_1 \in \Theta_{0}^{\complement}, \theta_0 \in \Theta_{0}$, it suffices to check that the inequality holds when the LHS is minimised and the RHS is maximised.

Notice that $\max_{\theta \in \Theta_{0}} \beta(\theta)=\max_{\theta \in \Theta_{0}}\mathbb{P}_\theta(\text{Type I error})=\alpha$, the size of the test.

You can, thus, rewrite the condition as $\beta(\theta_1) \geq \alpha$ for all $\theta_1 \in \Theta_{0}^{\complement}$. In words, when the alternative hypothesis is true, the power of the test is always greater than or equal to the size.

  • $\begingroup$ +1. This can't be clearer (in OP's notations). $\endgroup$ Mar 27 at 11:37
  • $\begingroup$ @User1865345 Thanks! $\endgroup$ Mar 27 at 11:43

Let me re-write the definition in a more familiar expression:

A size $\alpha$ test $\varphi$ is unbiased if

\begin{align}\mathbf E_\theta\varphi(X)&\leq \alpha,~~\theta\in\Theta_0,\\\mathbf E_\theta\varphi(X)&\geq \alpha,~~\theta\in\Theta_1.\end{align}

Here $\beta_\varphi(\theta) :=\mathbf E_\theta\varphi(X) $ is the power function of the test $\varphi.$

The interpretation is clear: the probability of $\varphi$ rejecting $\rm H_0$ when false is never smaller than that of rejecting of $\rm H_0$ when it is true.


$\rm[I]$ Mathematical Statistics: A Decision Theoretic Approach, Thomas S. Ferguson, Academic Press, $1967, $ p. $224.$


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.