Minimal Sufficient Statistics theorem 6.6.5 C&B proof

211 Views Asked by At

The theorem goes as follows:

Suppose the family of densities $\{f_0(x), .., f_k(0)\}$ all have common support. Then,

(a)

$$T(X) = \Big(\frac{f_1(X)}{f_0(X)}, \frac{f_2(X)}{f_0(X)}, .., \frac{f_k(X)}{f_0(X)}\Big)$$

is minimal sufficient for the family above.

(b) If $\mathcal{F}$ is a family of densities with common support, and

$\quad$ i) $f_i(x) \in \mathcal{F}, i= 0, 1, .., k$

$\quad$ ii) T(x) is sufficient for $\mathcal{F}$

then $T(x)$ is minimal sufficient for $\mathcal{F}$.

I already proved (a), but the textbook (Casella-Berger exercise 6.28) suggests to show that any sufficient statistic of $\mathcal{F}$ is a function of $T(x)$ defined in (a) somehow.

I was thinking of picking a random set of densities from $\mathcal{F}$, assuming that $T(x) = T(y)$, and then showing that some arbitrary sufficient statistic $T'(x) = T'(y)$ for $\mathcal{F}$ is implied. But somehow got stuck.