原文地址:【概率论与数理统计(研究生课程)】知识点总结7(参数估计)
E X l = μ l , l = 1 , 2 , . . . A l = 1 n ∑ i = 1 n X i l m a k e μ l = A l \begin{aligned} EX^l &= \mu_l, \quad l=1,2,... \\ A_l &= \frac{1}{n}\sum\limits_{i=1}^{n}X_i^l \\ make \quad \mu_l &=A_l \end{aligned} EXlAlmakeμl=μl,l=1,2,...=n1i=1∑nXil=Al
例如,当只有一个参数时, l l l取1,则 E X = X ˉ EX=\bar{X} EX=Xˉ
解题步骤:
- 用 E X l EX^l EXl找到参数与 μ l \mu_l μl的关系;
- 带入 μ l = A l \mu_l=A_l μl=Al,用样本表示参数;
- 解方程(组)得到参数的矩估计值。
L ( θ ) = L ( x 1 , x 2 , ⋯ , x n ; θ ) = ∏ i = 1 n p ( x i ; θ ) L(\theta)=L(x_1, x_2, \cdots,x_n;\theta)=\prod_{i=1}^np(x_i;\theta) L(θ)=L(x1,x2,⋯,xn;θ)=i=1∏np(xi;θ)
L ( θ ) = L ( x 1 , x 2 , ⋯ , x n ; θ ) = ∏ i = 1 n f ( x i ; θ ) L(\theta)=L(x_1, x_2, \cdots,x_n;\theta)=\prod_{i=1}^nf(x_i;\theta) L(θ)=L(x1,x2,⋯,xn;θ)=i=1∏nf(xi;θ)
解题步骤:
- 先求密度函数(连续型),或者分布律(离散型)【针对题目中只给分布函数的题型】
- 构造似然函数【必须是样本 x 1 , x 2 , ⋯ , x n x_1,x_2,\cdots,x_n x1,x2,⋯,xn的函数,而不是 X 1 , X 2 , ⋯ , X n X_1,X_2,\cdots,X_n X1,X2,⋯,Xn的函数】
- 对似然函数取对数【视情况而定,如果似然函数复杂,则取对数】
- 【取对数后的似然函数或者原似然函数】对参数求导(只含有一个参数)或分别对参数求偏导(含有多个参数)
- 令导数值为零,求解参数值。【如果有解,则这个值就是极大似然估计值;如果没有解,则判断导数值正负情况,以推断似然函数的单调性,从而根据单调性取得参数的极大似然估计值使似然函数最大】
如果 θ ^ \hat\theta θ^满足:
E ( θ ^ ) = θ E(\hat{\theta})=\theta E(θ^)=θ
则称 θ ^ \hat\theta θ^是 θ \theta θ的无偏估计。
- X ˉ \bar{X} Xˉ是 μ \mu μ的无偏估计
- S 2 = 1 n − 1 ∑ i = 1 n ( X i − X ˉ ) 2 = 1 n − 1 ∑ i = 1 n X i 2 − n X ˉ 2 S^2=\frac{1}{n-1}\sum\limits_{i=1}^{n}(X_i-\bar{X})^2=\frac{1}{n-1}\sum\limits_{i=1}^{n}X_i^2-n\bar{X}^2 S2=n−11i=1∑n(Xi−Xˉ)2=n−11i=1∑nXi2−nXˉ2是 σ 2 \sigma^2 σ2的无偏估计
- 1 n ∑ i = 1 n ( X i − μ ) 2 \frac{1}{n}\sum\limits_{i=1}^{n}(X_i-\mu)^2 n1i=1∑n(Xi−μ)2是 σ 2 \sigma^2 σ2的无偏估计
- E ( X ˉ ) = E X E(\bar{X})=EX E(Xˉ)=EX
- E ( S 2 ) = D X E(S^2)=DX E(S2)=DX
θ ^ → P θ \hat{\theta}\xrightarrow{P} \theta θ^Pθ
θ 1 ^ \hat{\theta_1} θ1^和 θ 2 ^ \hat{\theta_2} θ2^都是 θ \theta θ的无偏估计量,若 D θ 1 ^ < D θ 2 ^ D\hat{\theta_1}
当估计量 θ ^ \hat{\theta} θ^的方差 D θ ^ D\hat{\theta} Dθ^达到罗—克拉美下界:
D θ ^ ≥ I R = 1 n I ( θ ) , I ( θ ) = − E [ ∂ 2 ln p ( x ; θ ) ∂ θ 2 ] D\hat{\theta}\ge I_R=\frac{1}{nI(\theta)},I(\theta)=-E[\frac{\partial^2 \ln p(x;\theta)}{\partial \theta^2}] Dθ^≥IR=nI(θ)1,I(θ)=−E[∂θ2∂2lnp(x;θ)]
有效率: e ( θ ^ ) = I R D θ ^ e(\hat{\theta})=\frac{I_R}{D\hat{\theta}} e(θ^)=Dθ^IR
分布名称 | 参数 | 矩估计 | 极大似然估计 |
---|---|---|---|
0-1分布 | p p p | p ^ M = X ˉ \hat{p}_M=\bar{X} p^M=Xˉ | p ^ L = X ˉ \hat{p}_L=\bar{X} p^L=Xˉ |
二项分布 | p p p | p ^ M = X ˉ n \hat{p}_M=\frac{\bar{X}}{n} p^M=nXˉ | p ^ L = X ˉ n \hat{p}_L=\frac{\bar{X}}{n} p^L=nXˉ |
泊松分布 | λ \lambda λ | λ ^ M = X ˉ \hat{\lambda}_M=\bar{X} λ^M=Xˉ | λ ^ L = X ˉ \hat{\lambda}_L=\bar{X} λ^L=Xˉ |
几何分布 | p p p | p ^ M = 1 X ˉ \hat{p}_M=\frac{1}{\bar{X}} p^M=Xˉ1 | p ^ L = 1 X ˉ \hat{p}_L=\frac{1}{\bar{X}} p^L=Xˉ1 |
均匀分布 | a , b a,b a,b | a ^ M = X ˉ − 3 n ∑ i = 1 n ( X i − X ˉ ) 2 , b ^ M = X ˉ + 3 n ∑ i = 1 n ( X i − X ˉ ) 2 \hat{a}_M=\bar{X}-\sqrt{\frac{3}{n}\sum\limits_{i=1}^n{(X_i-\bar{X})^2}},\hat{b}_M=\bar{X}+\sqrt{\frac{3}{n}\sum\limits_{i=1}^n{(X_i-\bar{X})^2}} a^M=Xˉ−n3i=1∑n(Xi−Xˉ)2,b^M=Xˉ+n3i=1∑n(Xi−Xˉ)2 | a ^ L = X ( 1 ) , b ^ L = X ( n ) \hat{a}_L=X_{(1)},\hat{b}_L=X_{(n)} a^L=X(1),b^L=X(n) |
指数分布 | λ \lambda λ | λ ^ M = 1 X ˉ \hat{\lambda}_M=\frac{1}{\bar{X}} λ^M=Xˉ1 | λ ^ L = 1 X ˉ \hat{\lambda}_L=\frac{1}{\bar{X}} λ^L=Xˉ1 |
正态分布 | μ , σ 2 \mu,\sigma^2 μ,σ2 | μ ^ M = X ˉ , σ 2 ^ M = 1 n ∑ i = 1 n ( X i − X ˉ ) 2 \hat{\mu}_M=\bar{X},\hat{\sigma^2}_M=\frac{1}{n}\sum\limits_{i=1}^{n}(X_i-\bar{X})^2 μ^M=Xˉ,σ2^M=n1i=1∑n(Xi−Xˉ)2 | μ ^ L = X ˉ , σ 2 ^ L = 1 n ∑ i = 1 n ( X i − X ˉ ) 2 \hat{\mu}_L=\bar{X},\hat{\sigma^2}_L=\frac{1}{n}\sum\limits_{i=1}^{n}(X_i-\bar{X})^2 μ^L=Xˉ,σ2^L=n1i=1∑n(Xi−Xˉ)2 |
置信概率、置信度、置信水平 1 − α 1-\alpha 1−α P { θ ‾ ≤ θ ≤ θ ˉ } = 1 − α P\{\underline{\theta} \le \theta \le \bar{\theta}\}=1-\alpha P{θ≤θ≤θˉ}=1−α, θ \theta θ置信度为 1 − α 1-\alpha 1−α的置信区间 ( θ ‾ , θ ˉ ) (\underline{\theta}, \bar{\theta}) (θ,θˉ)。
常用统计量:
Z = X ˉ − μ σ / n ∼ N ( 0 , 1 ) , P { ∣ Z ∣ > Z α 2 } = α Z=\frac{\bar{X}-\mu}{\sigma/\sqrt{n}} \sim N(0,1), \quad P\{|Z|>Z_{\frac{\alpha}{2}}\}=\alpha Z=σ/nXˉ−μ∼N(0,1),P{∣Z∣>Z2α}=α
T = X ˉ − μ S / n ∼ t ( n − 1 ) , P { ∣ T ∣ > t α 2 ( n − 1 ) } = α T=\frac{\bar{X}-\mu}{S/\sqrt{n}} \sim t(n-1), \quad P\{|T|>t_{\frac{\alpha}{2}}(n-1)\}=\alpha T=S/nXˉ−μ∼t(n−1),P{∣T∣>t2α(n−1)}=α
χ 2 = ( n − 1 ) S 2 σ 2 ∼ χ 2 ( n − 1 ) , P ( { χ 2 < χ 1 − α 2 2 ( n − 1 ) } ∪ { χ 2 > χ α 2 2 ( n − 1 ) } ) = α \chi^2=\frac{(n-1)S^2}{\sigma^2} \sim \chi^2(n-1), \quad P(\{\chi^2<\chi^2_{1-\frac{\alpha}{2}}(n-1)\}\cup \{\chi^2>\chi^2_{\frac{\alpha}{2}}(n-1)\})=\alpha χ2=σ2(n−1)S2∼χ2(n−1),P({χ2<χ1−2α2(n−1)}∪{χ2>χ2α2(n−1)})=α
F = S 1 2 / σ 1 2 S 2 2 / σ 2 2 ∼ F ( n 1 − 1 , n 2 − 1 ) , P ( { F < F 1 − α 2 ( n 1 − 1 , n 2 − 1 ) } ∪ { F > F α 2 ( n 1 − 1 , n 2 − 1 ) } ) = α F=\frac{S_1^2/\sigma_1^2}{S^2_2/\sigma_2^2} \sim F(n_1-1,n_2-1), \quad P(\{F
方差已知 σ 2 = σ 0 2 \sigma^2=\sigma_0^2 σ2=σ02,估计均值
构造统计量: Z = X ˉ − μ σ / n ∼ N ( 0 , 1 ) Z=\frac{\bar{X}-\mu}{\sigma/\sqrt{n}} \sim N(0,1) Z=σ/nXˉ−μ∼N(0,1)
P { − Z α 2 < X ˉ − μ σ 0 / n < Z α 2 } = 1 − α P\{-Z_{\frac{\alpha}{2}}<\frac{\bar{X}-\mu}{\sigma_0/\sqrt{n}}
推得: [ X ˉ ± Z α 2 σ 0 n ] [\bar{X}\pm Z_{\frac{\alpha}{2}}\frac{\sigma_0}{\sqrt{n}}] [Xˉ±Z2αnσ0]
方差未知,估计均值,用 S 2 S^2 S2代替 σ 2 \sigma^2 σ2
构造统计量: T = X ˉ − μ S / n ∼ t ( n − 1 ) T=\frac{\bar{X}-\mu}{S/\sqrt{n}} \sim t(n-1) T=S/nXˉ−μ∼t(n−1)
P { − t α 2 ( n − 1 ) < X ˉ − μ S / n < t α 2 ( n − 1 ) } = 1 − α P\{-t_{\frac{\alpha}{2}}(n-1)<\frac{\bar{X}-\mu}{S/\sqrt{n}}
推得: [ X ˉ ± t α 2 ( n − 1 ) S n ] [\bar{X}\pm t_{\frac{\alpha}{2}}(n-1)\frac{S}{\sqrt{n}}] [Xˉ±t2α(n−1)nS]
当 n > 50 n>50 n>50时, X ˉ − μ S / n ∼ N ( 0 , 1 ) \frac{\bar{X}-\mu}{S/\sqrt{n}} \sim N(0,1) S/nXˉ−μ∼N(0,1)(近似),则区间估计为 [ X ˉ ± Z α 2 S n ] [\bar{X}\pm Z_{\frac{\alpha}{2}}\frac{S}{\sqrt{n}}] [Xˉ±Z2αnS]
方差已知, σ 1 2 , σ 2 2 \sigma_1^2, \sigma_2^2 σ12,σ22
构造统计量: X ˉ − Y ˉ − ( μ 1 − μ 2 ) σ 1 2 n 1 + σ 2 2 n 2 ∼ N ( 0 , 1 ) \frac{\bar{X}-\bar{Y}-(\mu_1-\mu_2)}{\sqrt{\frac{\sigma_1^2}{n_1}}+\frac{\sigma_2^2}{n_2}} \sim N(0,1) n1σ12+n2σ22Xˉ−Yˉ−(μ1−μ2)∼N(0,1)
推得: [ X ˉ − Y ˉ ± Z α 2 σ 1 2 n 1 + σ 2 2 n 2 ] [\bar{X}-\bar{Y}\pm Z_{\frac{\alpha}{2}}\sqrt{\frac{\sigma_1^2}{n_1}+\frac{\sigma_2^2}{n_2}}] [Xˉ−Yˉ±Z2αn1σ12+n2σ22]
方差未知,用 S 1 2 , S 2 2 S_1^2,S_2^2 S12,S22代替 σ 1 2 , σ 2 2 \sigma_1^2,\sigma_2^2 σ12,σ22
构造统计量: X ˉ − Y ˉ − ( μ 1 − μ 2 ) S w 1 n 1 + 1 n 2 , S w = ( n 1 − 1 ) S 1 2 + ( n 2 − 1 S 2 2 ) n 1 + n 2 − 2 \frac{\bar{X}-\bar{Y}-(\mu_1-\mu_2)}{S_w\sqrt{\frac{1}{n_1}+\frac{1}{n_2}}},S_w=\sqrt{\frac{(n_1-1)S_1^2+(n_2-1S_2^2)}{n_1+n_2-2}} Swn11+n21Xˉ−Yˉ−(μ1−μ2),Sw=n1+n2−2(n1−1)S12+(n2−1S22)
推得: [ X ˉ − Y ˉ ± t α 2 ( n 1 + n 2 − 2 ) ] S w 1 n 1 + 1 n 2 [\bar{X}-\bar{Y}\pm t_{\frac{\alpha}{2}}(n_1+n_2-2)]S_w\sqrt{\frac{1}{n_1}+\frac{1}{n_2}} [Xˉ−Yˉ±t2α(n1+n2−2)]Swn11+n21
方差已知, X ˉ − Y ˉ − ( μ 1 − μ 2 ) σ 1 2 n 1 + σ 2 2 n 2 ∼ N ( 0 , 1 ) \frac{\bar{X}-\bar{Y}-(\mu_1-\mu_2)}{\sqrt{\frac{\sigma_1^2}{n_1}}+\frac{\sigma_2^2}{n_2}} \sim N(0,1) n1σ12+n2σ22Xˉ−Yˉ−(μ1−μ2)∼N(0,1)【大子样近似】
推得: [ X ˉ − Y ˉ ± Z α 2 σ 1 2 n 1 + σ 2 2 n 2 ] [\bar{X}-\bar{Y}\pm Z_{\frac{\alpha}{2}}\sqrt{\frac{\sigma_1^2}{n_1}+\frac{\sigma_2^2}{n_2}}] [Xˉ−Yˉ±Z2αn1σ12+n2σ22]
方差未知,用 S 1 2 , S 2 2 S_1^2,S_2^2 S12,S22代替 σ 1 2 , σ 2 2 \sigma_1^2,\sigma_2^2 σ12,σ22,当 n 1 , n 2 n_1,n_2 n1,n2都很大时, X ˉ − Y ˉ − ( μ 1 − μ 2 ) S 1 2 n 1 + S 2 2 n 2 ∼ N ( 0 , 1 ) \frac{\bar{X}-\bar{Y}-(\mu_1-\mu_2)}{\sqrt{\frac{S_1^2}{n_1}+\frac{S_2^2}{n_2}}} \sim N(0,1) n1S12+n2S22Xˉ−Yˉ−(μ1−μ2)∼N(0,1) 【大子样近似】
推得: [ X ˉ − Y ˉ ± Z α 2 S 1 2 n 1 + S 2 2 n 2 ] [\bar{X}-\bar{Y}\pm Z_{\frac{\alpha}{2}}\sqrt{\frac{S_1^2}{n_1}+\frac{S_2^2}{n_2}}] [Xˉ−Yˉ±Z2αn1S12+n2S22]
μ \mu μ已知
构造统计量: χ 2 = ∑ i = 1 n ( X i − μ ) 2 σ 2 ∼ χ 2 ( n ) \chi^2=\frac{\sum\limits_{i=1}^{n}(X_i-\mu)^2}{\sigma^2} \sim \chi^2(n) χ2=σ2i=1∑n(Xi−μ)2∼χ2(n)
P { χ 1 − α 2 2 ( n ) ≤ ∑ i = 1 n ( X i − μ ) 2 σ 2 ≤ χ α 2 ( n ) 2 } = 1 − α P\{\chi^2_{1-\frac{\alpha}{2}}(n) \le \frac{\sum\limits_{i=1}^{n}(X_i-\mu)^2}{\sigma^2} \le\chi^2_{\frac{\alpha}{2}(n)} \}=1-\alpha P{χ1−2α2(n)≤σ2i=1∑n(Xi−μ)2≤χ2α(n)2}=1−α
推得: [ ∑ i = 1 n ( X i − μ ) 2 χ α 2 2 ( n − 1 ) , ∑ i = 1 n ( X i − μ ) 2 χ 1 − α 2 2 ( n − 1 ) ] [\frac{\sum\limits_{i=1}^{n}(X_i-\mu)^2}{\chi^2_{\frac{\alpha}{2}}(n-1)},\frac{\sum\limits_{i=1}^{n}(X_i-\mu)^2}{\chi^2_{1-\frac{\alpha}{2}}(n-1)}] [χ2α2(n−1)i=1∑n(Xi−μ)2,χ1−2α2(n−1)i=1∑n(Xi−μ)2]
μ \mu μ未知
构造统计量: ( n − 1 ) S 2 σ 2 ∼ χ 2 ( n − 1 ) \frac{(n-1)S^2}{\sigma^2}\sim \chi^2(n-1) σ2(n−1)S2∼χ2(n−1)
P { χ 1 − α 2 2 ( n − 1 ) ≤ ( n − 1 ) S 2 σ 2 ≤ χ α 2 ( n − 1 ) 2 } = 1 − α P\{\chi^2_{1-\frac{\alpha}{2}}(n-1) \le \frac{(n-1)S^2}{\sigma^2} \le\chi^2_{\frac{\alpha}{2}(n-1)} \}=1-\alpha P{χ1−2α2(n−1)≤σ2(n−1)S2≤χ2α(n−1)2}=1−α
推得: [ ( n − 1 ) S 2 χ α 2 2 ( n − 1 ) , ( n − 1 ) S 2 χ 1 − α 2 2 ( n − 1 ) ] [\frac{(n-1)S^2}{\chi^2_{\frac{\alpha}{2}}(n-1)},\frac{(n-1)S^2}{\chi^2_{1-\frac{\alpha}{2}}(n-1)}] [χ2α2(n−1)(n−1)S2,χ1−2α2(n−1)(n−1)S2]
构造统计量: F = σ 1 2 / σ 2 2 S 1 2 / S 2 2 ∼ F ( n 2 − 1 , n 1 − 1 ) F=\frac{\sigma_1^2/\sigma_2^2}{S_1^2/S_2^2} \sim F(n_2-1,n_1-1) F=S12/S22σ12/σ22∼F(n2−1,n1−1)
P { F 1 − α 2 ( n 2 − 1 , n 1 − 1 ) ≤ σ 1 2 / σ 2 2 S 1 2 / S 2 2 ≤ F α 2 ( n 2 − 1 , n 1 − 1 ) } = 1 − α P\{F_{1-\frac{\alpha}{2}}(n_2-1,n_1-1)\le \frac{\sigma_1^2/\sigma_2^2}{S_1^2/S_2^2} \le F_{\frac{\alpha}{2}}(n_2-1,n_1-1) \}=1-\alpha P{F1−2α(n2−1,n1−1)≤S12/S22σ12/σ22≤F2α(n2−1,n1−1)}=1−α
推得: [ S 1 2 S 2 2 F 1 − α 2 ( n 2 − 1 , n 1 − 1 ) , S 1 2 S 2 2 F α 2 ( n 2 − 1 , n 1 − 1 ) ] = [ S 1 2 S 2 2 1 F α 2 ( n 1 − 1 , n 2 − 1 ) , S 1 2 S 2 2 F α 2 ( n 2 − 1 , n 1 − 1 ) ] [\frac{S1^2}{S_2^2}F_{1-\frac{\alpha}{2}}(n_2-1,n_1-1),\frac{S1^2}{S_2^2}F_{\frac{\alpha}{2}}(n_2-1,n_1-1)]=[\frac{S1^2}{S_2^2}\frac{1}{F_{\frac{\alpha}{2}}(n_1-1,n_2-1)},\frac{S1^2}{S_2^2}F_{\frac{\alpha}{2}}(n_2-1,n_1-1)] [S22S12F1−2α(n2−1,n1−1),S22S12F2α(n2−1,n1−1)]=[S22S12F2α(n1−1,n2−1)1,S22S12F2α(n2−1,n1−1)]
P { θ ≥ θ 1 ^ } = 1 − α , [ θ ^ 1 , + ∞ ) P { θ ≤ θ 2 ^ } = 1 − α , ( − ∞ , θ 2 ^ ] \begin{aligned} P\{\theta \ge \hat{\theta_1} \}=1-\alpha, \quad [\hat{\theta}_1, +\infty) \\ P\{\theta \le \hat{\theta_2} \}=1-\alpha, \quad (-\infty,\hat{\theta_2}] \\ \end{aligned} P{θ≥θ1^}=1−α,[θ^1,+∞)P{θ≤θ2^}=1−α,(−∞,θ2^]
单侧置信区间在统计量的构造上与双侧的一致,在求区间时 α 2 \frac{\alpha}{2} 2α换成 α \alpha α。