6
$\begingroup$

Find a two dimensional minimal sufficient statistic for $\theta$ from $n$ independent random variables $X_k\sim > U(-k\theta+k,k\theta+k)$, $k\in\{1,\cdots,n\}$

Here is what I've attempted.

The pdf of X is $$\delta_\theta(x)=\prod_{i=1}^n\frac{1}{2k\theta}\mathbb{1}_{(1-\theta)k\leq x_k\leq(1+\theta)k}.$$ This means that $$\theta\ge\max\{1-\frac{x_k}{k},\frac{x_k}{k}-1\},\quad\forall k\in\{1,\cdots,n\}$$ or equivalently $|\frac{x_k}{k}-1|\leq\theta,\quad\forall k\in\{1,\cdots,n\}$. What I've got is $\underset{k\in\{1,\cdots,n\}}{\max}\left\{|\frac{x_k}{k}-1|\right\}$ as a minimal sufficient statistic. However, I am not sure if this is correct and the question was about there should be a 2-dimensional minimal statistic.

--Update It is always possible that the problem itself is wrong, i.e. there is no 2-dim minimal sufficient statistic. If this is the case, how to disprove it?

$\endgroup$
3
  • $\begingroup$ The value $k$ is an index, not a fixed quantity, so it should not appear in your answer. $\endgroup$
    – Ben
    Jan 5 at 0:45
  • $\begingroup$ Updated. Thanks. @Ben $\endgroup$
    – Tan
    Jan 5 at 0:49
  • $\begingroup$ Similar: stats.stackexchange.com/q/510173/119261 $\endgroup$ Feb 19 at 19:06
4
$\begingroup$

Your own attempted answer incorrectly treats $k$ as a fixed value, rather than an index. This gives you an incorrect likelihood function, which means that your subsequent work is also incorrect. We first observe the event equivalence:

$$-\theta k + k \leqslant x_k \leqslant \theta k + k \quad \quad \quad \iff \quad \quad \quad \Big| \frac{x_k}{k} - 1 \Big| \leqslant \theta,$$

which means that the correct likelihood function is:

$$\begin{align} L_\theta(\mathbf{x}_n) &= \prod_{k=1}^n \frac{1}{2k\theta} \cdot \mathbb{I}(\theta k - k \leqslant x_k \leqslant \theta k + k) \\[6pt] &= \prod_{k=1}^n \frac{1}{2k\theta} \cdot \mathbb{I} \Bigg( \theta \geqslant \Big| \frac{x_k}{k} - 1 \Big| \Bigg) \\[6pt] &= \frac{1}{(2\theta)^n n!} \cdot \mathbb{I} \Bigg( \theta \geqslant \max_k \Big| \frac{x_k}{k} - 1 \Big| \Bigg). \\[6pt] \end{align}$$

Consequently, a minimal sufficient statistic for the parameter $\theta$ is:

$$\max_k \Big| \frac{x_k}{k} - 1 \Big| .$$

$\endgroup$
4
  • 2
    $\begingroup$ But what is 'two dimensional' about this sufficient statistic as asked in the question? $\endgroup$ Jan 5 at 10:36
  • $\begingroup$ It is not two-dimensional. Is it possible you have misstated the bound in the problem? (I ask because if you change the lower bound to $-k + k \theta$ then you get a two dimensional minimum sufficient statistic instead.) $\endgroup$
    – Ben
    Jan 5 at 21:16
  • $\begingroup$ No, I double-checked.@Ben It is indeed $-k\theta+k$. I would be interesting then to show there is no two-dimensional minimal sufficient statistic. I don't think there is any theorem about the smallest dimension of minimal sufficient statistics. $\endgroup$
    – Tan
    Jan 7 at 5:47
  • $\begingroup$ On the contrary, if you show that you have a scalar statistic that is sufficient, then that is proof that there is no (non-degenerate) two-dimensional statistic that is also minimal sufficient. It is a fairly simple application of the definition of minimal sufficiency. $\endgroup$
    – Ben
    Jan 7 at 6:03

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.