Let $X_1,\ldots, X_n \sim f(x\mid \theta)=\frac{x}{\theta}e^{-x^2/(2\theta)}, x > 0$ independently. $\theta > 0$ is unknown.

How can I show that $\sum X_i$ is not a sufficient statistic for $\theta$?

I found from the likelihood ratio that

$$\frac{f(\mathbf{x}\mid \theta)}{f(\mathbf{y}\mid \theta)}=\prod^n_{i=1}\left(\frac{x_i}{y_i}\right)\frac{e^{-\frac{1}{2\theta}\sum x_i^2}}{e^{-\frac{1}{2\theta}\sum y_i^2}}$$

Since we need this likelihood ratio to not be a function of $\theta$, $\sum x_i^2=\sum y_i^2$. Thus, $T(\mathbf{X})=\sum x_i^2$ is our minimal sufficient statistic. (Is this logic correct?)

How can I show that $\sum x_i$ cannot be a sufficient statistic since $T(\mathbf{X})=\sum x_i^2$ cannot be written as a function of $\sum x_i$?

  • 2
    $\begingroup$ Hint: You will need to assume $n \ne 1$ for this to be true. $\endgroup$ – whuber Feb 6 at 4:57
  • $\begingroup$ @whuber I think I see what you're implying: to use $s^2$. I'll try playing around with this a bit more. Thank you! $\endgroup$ – Ron Snow Feb 6 at 17:23

It looks like you are trying to figure out how to prove that one thing is not a function of another thing. This is generally quite simple using a proof by counter-example. For any data vector $\mathbf{x}$, let $S(\mathbf{x}) \equiv \sum x_i$ denote its sum. If $T$ can be written as a composite function of $S$ then this would mean that $S(\mathbf{x}) = S(\mathbf{x}')$ implies $T(\mathbf{x}) = T(\mathbf{x}')$. Thus, to show that $T$ cannot be written as a function of $S$, all you need to do is find any two vectors $\mathbf{x}$ and $\mathbf{x}'$ that falsify this implication ---i.e. you need to establish a counter-example with:

$$S(\mathbf{x}) = S(\mathbf{x}') \quad \quad \text{ and } \quad \quad T(\mathbf{x}) \neq T(\mathbf{x}').$$

As whuber points out in the comments, you will actually need to assume that $n>1$ to get this result (since $S$ is sufficient if $n=1$). If you can establish a counter-example of this kind (which should be quite trivial), then you have shown that the minimal sufficient statistic $T$ is not a function of $S$, and so $S$ is not a sufficient statistic.

| cite | improve this answer | |
  • 2
    $\begingroup$ Such as letting $x_1=-2, x_2=2$ and $x'_1=-3, x'_2=3$? It appears that $S(\mathbf{x})=S(\mathbf{x'})=0$, while $T(\mathbf{x})=4, T(\mathbf{x'})=9$. Is this the right idea? $\endgroup$ – Ron Snow Feb 6 at 17:28
  • 1
    $\begingroup$ Yes, that is a valid counter-example. $\endgroup$ – Ben - Reinstate Monica Feb 6 at 20:41
  • $\begingroup$ Awesome- thank you for the clarification! $\endgroup$ – Ron Snow Feb 7 at 1:36

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.