证明：I类错误在边界点取到最大值

问题描述：

$X_1,X_2,...,X_n\stackrel{i.i.d}{\sim }Pois(\lambda ),\lambda >0$ for some unknown $\lambda$

$H_0:\lambda \ge {\lambda }_0,H_1:\lambda <{\lambda }_0,{\lambda }_0>0$

theorem: the maximum of the type 1 error is achieved at the boundary of ${\Theta }_{0}and{\Theta }_1$ for a one-sided tests, where the parameter space is 1-dimensional.

the wrong proof:

$$\begin{gathered} {\alpha }_{\lambda }=P_{\lambda \in {\Theta }_0}(\psi =1) \\ where,\psi =\mathbb{I}(T_n<-q_{\alpha }),withlevel\alpha \end{gathered}$$

on this question, let $T_n=\sqrt{n}\frac{\overline{X_n}-\lambda }{\sqrt{\lambda }}$

then, ${\alpha }_{\lambda }=P_{\lambda \in {\Theta }_0}(T_n<-q_{\alpha })=P_{\lambda \ge {\lambda }_0}(\sqrt{n}\frac{\overline{X_n}-\lambda }{\sqrt{\lambda }}<q_{-\alpha })=P_{\lambda \ge {\lambda }_0}(\sqrt{n}\frac{\overline{X_n}-\lambda }{\sqrt{\lambda }}<-q_{\alpha })$

let $T_{n,\lambda }=\sqrt{n}\frac{\overline{X_n}-\lambda }{\sqrt{\lambda }}$ and $T_{n,{\lambda }_0}=\sqrt{n}\frac{\overline{X_n}-{\lambda }_0}{\sqrt{{\lambda }_0}}$

when $\lambda \ge {\lambda }_0$, we have:

$$\begin{gathered} \lambda \ge {\lambda }_0 \\ \Rightarrow -\lambda \le -{\lambda }_0 \\ \Rightarrow \overline{X_n}-\lambda \le \overline{X_n}-{\lambda }_0 \\ \Rightarrow \frac{\overline{X_n}-\lambda }{{\lambda }_0}\le \frac{\overline{X_n}-{\lambda }_0}{{\lambda }_0} \\ \Rightarrow \frac{\overline{X_n}-\lambda }{{\lambda }_0}\le \frac{\overline{X_n}-\lambda }{\lambda }\le \frac{\overline{X_n}-{\lambda }_0}{{\lambda }_0} \\ \Rightarrow T_{n,\lambda }\le T_{n,{\lambda }_0} \end{gathered}$$

let event $A:T_{n,\lambda }<-q_a$ and $B:T_{n,{\lambda }_0}<-q_a$,
$B\subseteq A\iff P(B)\le P(A)\iff P(T_{n,{\lambda }_0}<-q_a)\le P(T_{n,\lambda }<-q_a)$

which means: the minimum of the type 1 error is achieved at the boundary of ${\Theta }_0$

what’s the problem in this proof? --hint, the wrong $T_n$ and the confusion of variable and constant.

the correct proof:

$$\begin{gathered} {\alpha }_{\lambda }=P_{\lambda \in {\Theta }_0}(\psi =1) \\ where,\psi =\mathbb{I}(T_n<-q_{\alpha }),withlevel\alpha \end{gathered}$$

on this question, let $T_n=\sqrt{n}\frac{\overline{X_n}-{\lambda }_0}{\sqrt{{\lambda }_0}}$

then, ${\alpha }_{\lambda }=P_{\lambda \in {\Theta }_0}(T_n<-q_{\alpha })=P_{\lambda \ge {\lambda }_0}(\sqrt{n}\frac{\overline{X_n}-{\lambda }_0}{\sqrt{{\lambda }_0}}<q_{-\alpha })=P_{\lambda \ge {\lambda }_0}(\sqrt{n}\frac{\overline{X_n}-\lambda }{\sqrt{\lambda }}<-q_{\alpha })$

let $T_{n,{\lambda }_0}=\sqrt{n}\frac{\overline{X_n}-{\lambda }_0}{\sqrt{{\lambda }_0}}$

when $\lambda ={\lambda }_0$ we have:

$\overline{X_n}\stackrel{\mathbb{P}}{⟶}{\lambda }_0$, then

$T_{n,{\lambda }_0}\stackrel{d}{⟶}N(0,1)$, because of CLT

${\alpha }_{\lambda }=P_{\lambda ={\lambda }_0}(\sqrt{n}\frac{\overline{X_n}-{\lambda }_0}{\sqrt{{\lambda }_0}}<-q_{\alpha })=\Phi (-q_{\alpha })=1-\Phi (q_{\alpha })$

when $\lambda >{\lambda }_0$, we have:

$\overline{X_n}\stackrel{\mathbb{P}}{⟶}\lambda$, then by continuous mapping theorem

$T_{n,{\lambda }_0}\stackrel{\mathbb{P}}{⟶}\sqrt{n}\frac{\lambda -{\lambda }_0}{\sqrt{{\lambda }_0}}\underset{n\to \infty }{\stackrel{\mathbb{P}}{⟶}}+\infty$, therefore,

$$\begin{gathered} {\alpha }_{\lambda }=P_{\lambda >{\lambda }_0}(\sqrt{n}\frac{\overline{X_n}-{\lambda }_0}{\sqrt{{\lambda }_0}}<-q_{\alpha }) \\ \stackrel{\mathbb{P}}{⟶}{P}_{\lambda >{\lambda }_0}(\sqrt{n}\frac{\lambda -{\lambda }_0}{\sqrt{{\lambda }_0}}<-q_{\alpha }) \\ \stackrel{\mathbb{P}}{⟶}{P}_{\lambda >{\lambda }_0}(+\infty <-q_{\alpha })\to 0 \end{gathered}$$

thus, the maximum of ${\alpha }_{\lambda }$ is achieved at $\lambda ={\lambda }_0$, which is the boundary of ${\Theta }_0$