Why is $E_A(Pr(\Theta \leq 0 | A= \alpha, X=x) | X=x)) = Pr_{\pi}(\Theta \leq 0 | X = x)$?

65 Views Asked by At

I'm reading Casella & Berger (1987) [1]. On page 107, the following can be found:

We use the notation $Pr_{\pi}(H_0|x)$ to indicate that $\pi$ is the prior used in calculating a posterior probability. Consider the random triple $(A, \theta,x)$ with joint distribution defined by the following. The distribution of $X|\Theta=\theta$ has density $f(x-\theta)$, the distribution of $\Theta|A=\alpha$ is $\pi_{\alpha}$, and the distribution of $A$ is $P$. Then for any $\pi \in \Gamma_M$, $$\begin{align} Pr_{\pi}(H_0|x) & = Pr_{\pi}(\Theta \leq 0 | X = x) \\ &= E_A(Pr(\Theta \leq 0 | A= \alpha, X=x) | X=x))\end{align}$$

Here, $\Gamma_M$ is the mixture of all elements of the set $\Gamma = \{\pi_{\alpha}: \alpha \in \mathcal{A}\}$, which is a class of prior distributions on the real line indexed by the set $\mathcal{A}$. $P$ is some probability measure on $\mathcal{A}$ and $f(x-\theta)$ is symmetric about zero and has monotone likelihood ratio.

I've got a hard time to understand the last equality in the equation stated above, namely:

$$E_A(Pr(\Theta \leq 0 | A= \alpha, X=x) | X=x)) = Pr_{\pi}(\Theta \leq 0 | X = x)$$

Any hints and help would be greatly appreciated.

[1] Casella, G., & Berger, R. L. (1987). Reconciling Bayesian and frequentist evidence in the one-sided testing problem. Journal of the American Statistical Association, 82(397), 106-111.

1

There are 1 best solutions below

0
On BEST ANSWER

to show

$$ Pr_{\pi}(\Theta \leq 0 | X = x) = E(Pr(\Theta \leq 0 | A= \alpha, X=x) | X=x))$$

it is better to prove

$$ Pr_{\pi}(\Theta \leq 0 | X ) = E(Pr(\Theta \leq 0 | A, X) | X))$$

start with

$E(P(\Theta \leq 0 | A, X) | X))=E(E(I_{\Theta \leq 0} | A, X) | X))$

$=E(E(I_{\Theta \leq 0} | \sigma(A, X)) | \sigma(X)))$

since $\sigma(X) \subset \sigma(A, X)$ by tower property

$=E(I_{\Theta \leq 0} |\sigma(X))$

$=E(I_{\Theta \leq 0} |X)$

$=P(\Theta \leq 0 |X)$ so

$$ Pr_{\pi}(\Theta \leq 0 | X ) = E(Pr(\Theta \leq 0 | A, X) | X))$$

and in hence

$$ Pr_{\pi}(\Theta \leq 0 | X = x) = E(Pr(\Theta \leq 0 | A= \alpha, X=x) | X=x))$$