This problem is from a qualifying exam of mathematical statistics of Department of Mathematics, POSTECH, and I'm trying to solve this problem on reference of Casella & Berger's Statistical Inference.

Let $X$ be a random variable with pdf $f(x;\theta)$ and $T=T(X)$ be a sufficient statistic for $\theta$. A statistic $\psi$ is called a test function if $0\le \psi \le 1$. Show that for every test function $\psi(X)$, there exists a test function $\tilde{\psi}(T)$ such that $E_\theta\tilde{\psi}(T)=E_\theta\psi(X)$ for all $\theta$.

My attempt: $E_\theta\psi(X)=\int\psi(x)f(x;\theta)dx=\int\psi(x) \frac{f(x;\theta)}{g(t;\theta)}g(t;\theta)dx$

Here $g(t;\theta)$ is the pdf for $T=T(X)$. Now I stuck in manipulating the above integral formula. I'd like to exploit the sufficiency of $T$, but I can't find the way how to. Does anyone have ideas?

Any hints or advices are welcome! Thanks!

  • 2
    $\begingroup$ Hi there and welcome. Please include an exact/full reference. Also, please inlcude the self-study tag. Best of luck. – Reviewer $\endgroup$
    – Jim
    Jun 20, 2018 at 15:12
  • 2
    $\begingroup$ Consider constructing the function $\tilde \psi$ directly from $\psi$ by exploiting the definition (not characterization!) of sufficient statistics. $\endgroup$
    – whuber
    Jun 20, 2018 at 15:54
  • 1
    $\begingroup$ @whuber Thanks! I finally figured out the answer. I'll post right here soon. $\endgroup$
    – bellcircle
    Jun 21, 2018 at 5:59

1 Answer 1


By sufficiency of $T$, we get $f(x|t)f(t|\theta)=f(x|t,\theta)f(t|\theta)=f(x,t|\theta)$

Therefore, $E_\theta\psi(X)=\int\int\psi(x)f(x,t|\theta)dxdt=\int\int\psi(x)f(x|t)f(t|\theta)dxdt$

Letting $\tilde\psi(t)=\int\psi(x)f(x|t)dt$, we get the desired result.


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.