UA MATH566 统计理论7 似然比检验
Neyman-Pearson Lemma指出似然比检验是UMP检验,但第五讲只讨论了原假设和备择假设各只有一种取值的检验:
H 0 : θ = θ 0 H a : θ = θ 1 H_0:\theta = \theta_0 \\ H_a:\theta = \theta_1 H0:θ=θ0Ha:θ=θ1
这一讲把似然比检验推广到参数空间 Θ ⊂ R \Theta \subset \mathbb{R} Θ⊂R。
考虑假设检验:
H 0 : θ ∈ Θ 0 H a : θ ∈ Θ 1 H_0:\theta \in \Theta_0 \\ H_a:\theta \in \Theta_1 H0:θ∈Θ0Ha:θ∈Θ1
其中 Θ 0 ⊔ Θ 1 = Θ \Theta_0 \sqcup \Theta_1 = \Theta Θ0⊔Θ1=Θ。通常有三种设置假设的方法,选取某个参数 θ ∗ ∈ Θ \theta^* \in \Theta θ∗∈Θ:
- Increase: Θ 0 = { θ : θ ≤ θ ∗ } , Θ 0 = { θ : θ > θ ∗ } \Theta_0 = \{\theta:\theta \le \theta^*\},\Theta_0 = \{\theta:\theta > \theta^*\} Θ0={ θ:θ≤θ∗},Θ0={ θ:θ>θ∗}
- Decrease: Θ 0 = { θ : θ ≥ θ ∗ } , Θ 0 = { θ : θ > θ ∗ } \Theta_0 = \{\theta:\theta \ge \theta^*\},\Theta_0 = \{\theta:\theta > \theta^*\} Θ0={ θ:θ≥θ∗},Θ0={ θ:θ>θ∗}
- Change: Θ 0 = { θ = θ ∗ } , Θ 0 = { θ : θ ≠ θ ∗ } \Theta_0 = \{\theta = \theta^*\},\Theta_0 = \{\theta:\theta \ne \theta^*\} Θ0={ θ=θ∗},Θ0={ θ:θ=θ∗}
前两种叫单边检验(one-sided test),第三种叫双边检验(two-sided test)。仍然记拒绝域为 C C C,记样本为 X = { X 1 , ⋯ , X n } X=\{X_1,\cdots,X_n\} X={
X1,⋯,Xn},定义势函数为
P o w e r ( θ ) = P [ X ∈ C ] Power(\theta)=P[X \in C] Power(θ)=P[X∈C]
势函数是参数 θ \theta θ的函数,定义在 Θ \Theta Θ上,根据参数的不同它有两种含义:
- θ ∈ Θ 0 , P o w e r ( θ ) = t y p e I e r r o r \theta \in \Theta_0, Power(\theta)=type\ I\ error θ∈Θ0,Power(θ)=type I error
- θ ∈ Θ 1 , P o w e r ( θ ) = t y p e I I e r r o r \theta \in \Theta_1, Power(\theta)=type\ II\ error θ∈Θ