Sufficient statistics, how compute it.

209 Views Asked by At

Guys can you help me with this? I’ve tried to apply the definition of sufficient statistics but the professor only gave us simple examples so I don’t know how to compute it.

$$X\sim f_θ$$

$$f_θ(x)=\frac{θ}{2}(θx)^2\exp\{−θx\}\text{ for }x>0$$

Moreover, let $X_1,…,X_n$ be a sample from $X$. a) Find a sufficient statistic for the model b) Find the MLE of $θ$ and the MLE of $E(X)=\frac{3}{θ}$

1

There are 1 best solutions below

2
On BEST ANSWER

\begin{align} \prod_{i=1}^n f_\theta(x_i) & = \left(\frac \theta 2\right)^n ((\theta x_1) \cdots (\theta x_n))^2 \exp\{−\theta (x_1+\cdots+x_n)\} \\[10pt] & = \Big(\, \overbrace{\theta^{3n} \exp(-\theta(x_1+\cdots + x_n))}^\text{first factor}\,\Big) \cdot \Big(\, \overbrace{ \frac{x_1\cdots x_n}{2^n}}^\text{second factor} \, \Big) \tag 1 \end{align}

Fisher's factorization theorem says that if this can be written as a product of two factors in which

  • one factor depends on $x_1,\ldots,x_n$ only through some specified function of $x_1,\ldots,x_n$ and
  • the other factor does not depend on $\theta,$

then that specified function, evaluated at the sample, is a sufficient statistic.

Therefore $X_1+\cdots+X_n$ is a sufficient statistic.

If $L(\theta) = $ the expression on line $(1)$ above, then $$ \ell(\theta) = \log L(\theta) = 3n\log\theta - \theta(x_1+\cdots+x_n) + \text{constant} $$ so $$ \ell\,'(\theta) = \frac{3n}\theta - (x_1+\cdots+x_n) \quad \begin{cases} >0 & \text{if } 0 \le \theta < 3/\overline{x}, \\[6pt] <0 & \text{if } \theta>3/\overline{x}. \end{cases} $$ Therefore $\widehat \theta = 3/\overline x.$

By equivariance of MLEs, the MLE of $3/\theta$ is $\overline x.$