Q1: Let X 1 , … , X n X_1,\dots,X_n X1,…,Xn be a sample from density
f ( x ; σ ) = 1 2 σ e − ∣ x ∣ / σ , f(x;\sigma)=\frac{1}{2\sigma} e^{-|x|/\sigma}, f(x;σ)=2σ1e−∣x∣/σ,where σ > 0 \sigma>0 σ>0. Find the maximum likelihood estimator for σ \sigma σ.
Solution: 似然函数为:
L ( σ ) = ∏ i = 1 n f ( x i ) = ∏ i = 1 n ( 1 2 σ e − ∣ x i ∣ / σ ) = ( 2 σ ) − n e x p ( − 1 σ ∑ i = 1 n ∣ x i ∣ ) L(\sigma)=\displaystyle\prod_{i=1}^nf(x_i)=\displaystyle\prod_{i=1}^n(\frac{1}{2\sigma}e^{-|x_i|/\sigma})=(2\sigma)^{-n}exp(-\frac{1}{\sigma}\displaystyle\sum_{i=1}^n|x_i|) L(σ)=i=1∏nf(xi)=i=1∏n(2σ1e−∣xi∣/σ)=(2σ)−nexp(−σ1i=1∑n∣xi∣)于是有
ℓ ( σ ) = l n L ( σ ) = − n l n ( 2 σ ) − 1 σ ∑ i = 1 n ∣ x i ∣ \ell(\sigma)=lnL(\sigma)=-nln(2\sigma)-\frac{1}{\sigma}\displaystyle\sum_{i=1}^n|x_i| ℓ(σ)=lnL(σ)=−nln(2σ)−σ1i=1∑n∣xi∣令
∂ ∂ σ ℓ ( σ ) = − n σ + 1 σ 2 ∑ i = 1 n ∣ x i ∣ = 0 \frac{\partial}{\partial\sigma}\ell(\sigma)=-\frac{n}{\sigma}+\frac{1}{\sigma^2}\displaystyle\sum_{i=1}^n|x_i|=0 ∂σ∂ℓ(σ)=−σn+σ21i=1∑n∣xi∣=0可解得
σ ^ = ∑ i = 1 n ∣ x i ∣ n \hat{\sigma}=\frac{\displaystyle\sum_{i=1}^n|x_i|}{n} σ^=ni=1∑n∣xi∣因此,我们可以得到 σ \sigma σ的极大似然估计值 σ ^ M L E = 1 n ∑ i = 1 n ∣ X i ∣ \hat{\sigma}_{MLE}=\frac{1}{n}\sum_{i=1}^n|X_i| σ^MLE=n1∑i=1n∣Xi∣
Q2: Let X 1 , … , X n X_1,\dots,X_n X1,…,Xn be a sample from U [ θ , θ + 1 ] \mathbb{U}[\theta,\theta+1] U[θ,θ+1], where θ ∈ R \theta\in\mathbb{R} θ∈R. Prove that the maximum likelihood estimator (MLE) for θ \theta θ is not unique, and find out all the MLEs.
Solution: 随机变量 X 1 , X 2 , ⋯ , X n X_1,X_2,\cdots,X_n X1,X2,⋯,Xn服从均匀分布 U [ θ , θ + 1 ] \mathbb{U}[\theta,\theta+1] U[θ,θ+1],则其概率密度为:
f ( x ; θ ) = { 1 θ ≤ x ≤ θ + 1 0 其 他 f(x;\theta)= \begin{cases} 1 & \theta\leq x\leq\theta+1 \\ 0 & 其他 \end{cases} f(x;θ)={
10θ≤x≤θ+1其他似然函数为:
L ( θ ) = ∏ i = 1 n f ( x i ) = ∏ i = 1 n 1 { θ ≤ x i ≤ θ + 1 } = 1 { x ( n ) − 1 ≤ θ ≤ x ( 1 ) } L(\theta)=\displaystyle\prod_{i=1}^nf(x_i)=\displaystyle\prod_{i=1}^n1\{\theta\leq x_i\leq\theta+1\}=1\{x_{(n)}-1\leq\theta\leq x_{(1)}\} L(θ)=i=1∏nf(xi)=i=1∏n