The minimum sufficient statistics for $f(x) = \frac12 exp(-|x-\theta|)$ for $-\infty < \theta < +\infty$ is $ T(X) = \{X_{(1)},X_{(n)} \}$. I want to show that the above is complete.

$f(x) = \frac12 exp(-|x-\theta|)$ is location family therefore the range is an ancillary statistic which is free of $\theta$

Let $h((T(x))$ be $X_{(n)}-X_{(n)}$

$0 = E_\theta [ g(h(T(X))] = \int_{-\infty}^{+\infty} g(h(t)) f(h(t))dt$ $.

$f(h(t))$ is now $\frac12 exp(-|(t_{(n)}-\theta) - (t_{(1)}-\theta)|)$ which reduces to

$\frac12 exp(-|(t_{(n)}- t_{(1)}|)$.

When$ \int_{-\infty}^{+\infty} g(h(t)) \frac12 exp(-|(t_{(n)}- t_{(1)}|)dt = 0$, then $g(h(t)) = 0$ $\forall h(t)$

Now, I want to claim that since $h(T)$ is complete and there is a 1-1 function between $h(T)$ and $T$ therefore $T$ is also complete. Is this approach correct?

  • $\begingroup$ The minimal sufficient statistic is not complete in this case. Consider the vector of the pairwise differences, with fixed expectation.. $\endgroup$ – Xi'an Sep 5 '20 at 13:16
  • $\begingroup$ @Xian, just to follow your suggestion. You would want me to apply a function on T s.t I have something like $( X_(1)-X_(2),,,X(n-1) - X(n)) as my new t correct? $\endgroup$ – user1916067 Sep 5 '20 at 16:16
  • 1
    $\begingroup$ Minimal sufficient statistic is the full set of order statistics $(X_{(1)},X_{(2)},\ldots,X_{(n)})$. $\endgroup$ – StubbornAtom Sep 5 '20 at 19:34
  • $\begingroup$ @StubbornAtom: the set $\{X_1,\ldots,X_n\}$ is equally minimal sufficient. $\endgroup$ – Xi'an Sep 6 '20 at 8:03

Since the double exponential distribution is not part of an exponential family, there cannot exist a sufficient distribution with fixed (in $n$) dimension. The minimal sufficient statistic in this case is $(X_{(1)},\ldots,X_{(n)})$ or, equivalently, $\{X_1,\ldots,X_n\}$. Since $\mathbb{E}[X_i-X_j]=0$, this statistic cannot be complete.


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.