Let the distribution of $X_1,X_2,...X_n$ depend on two parameters $a, b$ such that there exists a single sufficient statistic, for either parameter when the other is fixed/known.

Show that there is necessarily a pair of jointly sufficient statistics when both parameters are unknown.

I tried by employing the factorization theorem in each when when one is known and the other is unknown, Say the parameters are $a,b$ then, by Factorisation theorem, $$\begin{array}{} f(x,a)=g_1(t_1,a)h(x) \\ f(x,b)=g_2(t_2,b)h'(x) \end{array}$$ Now, $$f(x,a)f(x,b)=f^2(x;a,b)=g_1(t_1,a)h(x)g_2(t_2,b)h'(x)$$

But, after that I can't think of anything to proceed with

  • 3
    $\begingroup$ I'm voting to close this question because there are ubiquitous spelling errors that show a lack of care by the OP. $\endgroup$
    – Ben
    Commented Aug 16, 2019 at 7:31
  • 4
    $\begingroup$ Aside from spelling, I actually found the format of the question not so bad (for instance, the question/intentions are clear and, while the question seems self-study, the OP has explained what has been tried by themselves). Therefore I took the effort to improve the question. (also I could not quickly imagine a solution/answer to this question myself, so it got my interest) $\endgroup$ Commented Aug 16, 2019 at 12:17
  • 1
    $\begingroup$ Thank you for the support, Sir! I'll be more careful from next time! $\endgroup$ Commented Aug 16, 2019 at 12:19
  • $\begingroup$ What happens when you take the square root of the final line? $\endgroup$ Commented Aug 16, 2019 at 12:23
  • 1
    $\begingroup$ Your notation may be misleading you: it does not involve two parameters, but only one, to which you variously give names "$a$" and "$b$". Correct notation would be of the form $f(x; a,b).$ $\endgroup$
    – whuber
    Commented Aug 16, 2019 at 12:36

1 Answer 1


As stressed in most of the comments, the notation is deficient and possibly explains why you cannot solve the problem.

The distribution of the sample $X_1,\cdots,X_n$ is parametrised by $(a,b)$, hence its density should be written as $$f(x_1,\ldots,x_n;a,b)$$ The assumptions made in the question about this density are that there exist two univariate functions $\hat a(x_1,\ldots,x_n;b)$ and $\hat b(x_1,\ldots,x_n;a)$ such that \begin{align} f(x_1,\ldots,x_n;a,b) &= f_a(\hat a(x_1,\ldots,x_n;b);a,b)g_a(x_1,\ldots,x_n;b)\\ &= f_b(\hat b(x_1,\ldots,x_n;a);a,b)g_b(x_1,\ldots,x_n;a) \end{align} by the factorisation theorem applied to both cases when $a$ and $b$ are the parameters (since all functions still depend on the other parameter even if it is considered as fixed).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.