My stats class is studying sufficient statistics. So far, every example we have done has assumed $h(\mathbf{X}) = 1$ and so all our examples have involved really trivial uses of the factorization theorem in this regard.

Fisher-Neyman factorisation theorem: If the probability density function is $ƒ_θ(x)$, then $T$ is sufficient for $θ$ if and only if nonnegative functions $g$ and $h$ can be found such that $$ f_\theta(x)=h(x) \, g_\theta(T(x)), \,\! $$

I am a little worried this might come up on an exam where $h(\mathbf{X}) \neq 1$.

My Question

Can someone provide a basic example of using the factorization theorem to show a statistic is sufficient when $h(\mathbf{X}) \ne 1$?


As explained in this answer to another question on sufficiency, if you consider an iid sample $(X_1,\ldots,X_n)$ from an arbitrary distribution with density $f$, the order statistic $S(\mathbf{X})=(X_{(1)},\ldots,X_{(n)})$ is sufficient. The distribution of the sample $(X_1,\ldots,X_n)$ conditionally on $S(\mathbf{X})$ is then uniform over all permutations of $S(\mathbf{X})$, with density $h(\mathbf{X})=1/n!$.

Note however that the function $h$ appearing in the factorisation theorem is somewhat arbitrary in that it is a density against a particular dominating measure. For instance the counting measure in the above example. If one incorporates $h$ into the measure, then the corresponding density becomes $1$, which means that we can always assume $h(x)\equiv 1$.

| cite | improve this answer | |

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.