5
$\begingroup$

Let $X_1, X_2 , \dots, X_n$ be i.i.d observations from $N(\theta, \theta)$ (where $\theta$ is the variance). I intend to find the minimal sufficient statistics (M.S.S) for the joint distribution.

My steps ultimately lead me to the following:

$\dfrac{f_\theta(x)}{f_\theta(y)} = \text{exp}{ \left\{ \dfrac{-1}{2\theta} \left[ \left(\sum x_i^2 - \sum y_i^2 \right) + 2\theta \left( \sum y_i - \sum x_i \right) \right] \right) }$.

My M.S.S are therefore $\left( \sum x_i^2 , \sum x_i \right)$

Am I on track?

$\endgroup$
5
  • 2
    $\begingroup$ You are on the write track, this is one approach to solve it. But I believe that your solution is incorrect. You have to be careful with appealing to the theorem, which states that T(X) is a minimally sufficient statistic if and only if for T(X) = T(Y), that $\frac{f_{\theta}(x)}{f_{\theta}(y)} = \phi(x,y)$, where $\phi(x,y)$ is free of $\theta \forall \theta \in \Theta$. The hint I'll give is that one of your MSS solutions is not necessary for this to be satisfied. $\endgroup$ Aug 27, 2020 at 3:14
  • $\begingroup$ Since this is a regular exponential family, a minimal sufficient statistic is obtained directly through 'exponential family factorization'. $\endgroup$ Aug 27, 2020 at 7:38
  • $\begingroup$ @Tyrel Stokes: I have realized it. the $\Sigma x_i $ is not necessary. $\endgroup$ Aug 27, 2020 at 12:12
  • $\begingroup$ @Scortchi-ReinstateMonica That term does not depend on $\theta$. Our aim is to find a solution so that the ratio is free of $\theta$. $\endgroup$ Aug 29, 2020 at 3:10
  • $\begingroup$ People use the notation $N(\theta, \theta)$ in two senses, $\theta$ the variance or $\theta$ the standar deviation. You should specify which (but from the rest of the post we can infer first option, so I did add it). $\endgroup$ Apr 20, 2021 at 18:20

2 Answers 2

2
$\begingroup$

It turns out that after further expansion, the above becomes

$\dfrac{f_\theta(x)}{f_\theta(y)} = \exp \left\{ - \left[ \dfrac{1}{2\theta} \left( \sum x_i^2 - \sum y_i^2 \right) + \left( \sum y_i - \sum x_i \right) \right] \right\}$.

Since it is only the first term that depends on $\theta$, the M.S.S is $\sum x_i^2$.

$\endgroup$
1
  • 1
    $\begingroup$ Perhaps clearer with something like $\exp \left(- \frac{ \Sigma x_i^2 - \Sigma y_i^2}{2\theta}\right)\;\exp\left(- {(\Sigma y_i - \Sigma x_i)}\right)$ $\endgroup$
    – Henry
    Aug 27, 2020 at 12:48
1
$\begingroup$

By writing the joint density in exponential family form we find

$$ f(x,\theta) = \frac1{\sqrt{2\pi}} \exp\left\{ -\frac{n}{2}\log\theta - \frac1{2\theta} \sum_i x_i^2 + \sum_i x_i -n\theta^2 \right\} $$ and then the result follows by general exponential family theory, as noted in comments.

$\endgroup$

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.