Is this statistic minimal sufficient? Exponential family method

155 Views Asked by At

My task is to first show that the binomial pdf belongs to the exponential family, and then to identify the minimal sufficient statistic when $0<p<1$ is unknown (nothing is said about $n$ so I'm assuming that it is known).

First, I rewrite the pdf $f(x|\theta)={n \choose x} p^x(1-p)^{n-x}$ as $f(x|\theta)={n \choose x}e^{xln(\theta)+(n-x)ln(1-\theta)}$. From here, I can use the definition of a k-parameter exponential family and assert that if $a(\theta)=1$, $g(x)={n \choose x}$, $R_1(x)=x$, $b_1(\theta)=ln(\theta)$, $R_2(x)=(n-x)$, and $b_2(\theta)=ln(1-\theta)$, then $f(x|\theta)$ belongs to the 2 parameter exponential family.

Accordingly, the statistic $T=(\sum_{i=1}^nx, \sum_{i=1}^n(n-x))$ would be minimally sufficient. However, $T_2$ is just a 1-1 function of $T_1$, so this can't be minimally sufficient. Does anyone see where I may have messed up?