自助法和水手刀法

R包boot中含有数据集aircondit,里面含有12个观测值用来记录空调设备两次故障之间的小时数。假设故障之间的时间服从指数分布 \(Exp(\lambda)\)。

  1. 给出故障率 \(\lambda\) 的极大似然估计;

\(f(x)=\lambda e^{-\lambda x},\)独立同分布的样本联合密度分布为\(f(x)=\lambda^n e^{-\lambda \sum_{i=1}^{n} x_i},\lambda_{MLE}=\frac {1}{\bar{x}},n=12\)(本题)。

  1. 使用自助法对该估计的偏差和标准差进行估计;
library(boot)
m <- 1/mean(aircondit$hours)   #检查了aircondit只有hours这一列
#初始化
B <- 200 # Number of bootstrapping
n <- nrow(aircondit) # Sample size
MLE <- numeric(B)

for(b in 1:B){
  i <- sample(1:n, size=n, replace = TRUE)
  hour <- aircondit$hours[i]
  MLE[b] <- 1/mean(hour)
}
se_theta_MLE <- sd(MLE)
bias <- mean(MLE-m)
cat("该估计标准误差是",se_theta_MLE,",偏差的估计是",bias,".\n")
## 该估计标准误差是 0.004764472 ,偏差的估计是 0.001918553 .
  1. 分别用标准正态、基本、百分位数、自助法t区间和BCa法计算故障间隔平均时间 \(1/\lambda\) 的95%自助法置信区间。

本题求的是\(1/\lambda\) 的95%自助法置信区间,所以其似然估计为\(\bar{x}\).

标准正态法
\[(\hat{\theta} - u_{\alpha/2} se(\hat{\theta}), \ \hat{\theta} + u_{\alpha/2} se(\hat{\theta})),\]

# Set up the bootstrap
B <- 200 # Number of bootstrapping
n <- nrow(aircondit) # Sample size
MLE <- numeric(B)
MLE.hat <- mean(aircondit$hours)
alpha <- 0.05   

for(b in 1:B){
  i <- sample(1:n, size=n, replace = TRUE)
  hour <- aircondit$hours[i]
  MLE[b] <- mean(hour)
}
#标准正态法
se_theta_MLE <- sd(MLE)
cat("置信区间为[",MLE.hat-qnorm(p = alpha/2,lower.tail = FALSE)*se_theta_MLE,",",MLE.hat+qnorm(p = alpha/2,lower.tail = FALSE)*se_theta_MLE,"].",sep="")
## 置信区间为[38.46575,177.7009].

基本法:
利用经验分布得到置信区间公式如下
\[(2\hat{\theta} - \hat\theta_{\alpha/2}, \ 2\hat{\theta} - \hat\theta_{1-\alpha/2} ).\]

#基本法
MLE.sort <- sort(MLE)
cat("置信区间为[",2*MLE.hat-MLE.sort[(1-alpha/2)*B],",",2*MLE.hat-MLE.sort[alpha/2*B],"].",sep="")
## 置信区间为[27,171.0833].

百分位数法:
利用自助法重复试验的经验分布作为参考。 \[(\hat{\theta}_{1-\alpha/2},\ \hat{\theta}_{\alpha/2}),\]

#百分位数法
MLE.sort <- sort(MLE)
cat("置信区间为[",MLE.sort[alpha/2*B],",",MLE.sort[(1-alpha/2)*B],"].",sep="")
## 置信区间为[45.08333,189.1667].

自助法t区间
\[(\hat{\theta} - t^*_{1-\alpha/2} \hat{se}(\hat{\theta}), \ \hat{\theta} + t^*_{\alpha/2} \hat{se}(\hat{\theta}))\]

#为在相同随机样本下比较几种方法,认为保持向量MLE一致比较好
t.stat <- numeric(B)
for(b in 1:B){
  MLE.mat <- replicate(100, expr = {
      i <- sample(1:n, size=n, replace = TRUE)
      hour <- aircondit$hours[i]
      mean(hour)
  })
  se <- sd(MLE.mat)
  t.stat[b] <- (MLE[b] - MLE.hat) / se
}

Qt <- quantile(t.stat, c(1-alpha/2, alpha/2), type = 1)
#print(MLE.hat - Qt * sd(MLE))
cat("置信区间为[",MLE.hat-Qt[1] * sd(MLE),",",MLE.hat-Qt[2] * sd(MLE),"].",sep="")
## 置信区间为[25.95517,169.3914].
#若按照老师的代码
#MLE <- t.stat <- numeric(B)
#for(b in 1:B){
  #i <- sample(1:n, size=n, replace = TRUE)
  #hour <- aircondit$hours[i]
  #MLE[b] <- mean(hour)
  #MLE.mat <- replicate(100, expr = {
      #i <- sample(1:n, size=n, replace = TRUE)
      #hour <- aircondit$hours[i]
      #mean(hour)
  #})
  #se <- sd(MLE.mat)
  #t.stat[b] <- (MLE[b] - MLE.hat) / se
#}

#Qt <- quantile(t.stat, c(1-alpha/2, alpha/2), type = 1)
#print(MLE.hat - Qt * sd(MLE))
#cat("置信区间为[",MLE.hat-Qt[1] * sd(MLE),",",MLE.hat-Qt[2] * sd(MLE),"].",sep="")

BCa法:
\(100(1 - \alpha)\%\) 加速偏差修正百分位数(BCa)自助法置信区间为 \((\hat{\theta}_{\alpha_1}, \ \hat{\theta}_{\alpha_2})\),其中

\[\alpha_1 = \Phi\left(\hat{z}_0 + \frac{\hat{z}_0 + z_{1-\alpha/2}}{1-\hat{a}(\hat{z}_0 + z_{1-\alpha/2})}\right),\] \[\alpha_2 = \Phi\left(\hat{z}_0 + \frac{\hat{z}_0 + z_{\alpha/2}}{1-\hat{a}(\hat{z}_0 + z_{\alpha/2})}\right),\]

z0 <- qnorm(mean(MLE<MLE.hat))
theta.jack <- numeric(n)
for(i in 1:n){
  theta.jack[i] <- mean(aircondit$hours[-i])
}
bias <- mean(theta.jack) - theta.jack
a <- sum(bias^3)/(6 * sum(bias^2) ^(3/2))

alpha1 <- pnorm(z0 + (z0+qnorm(p = 1-alpha/2,lower.tail = FALSE))/(1-a*(z0+qnorm(p = 1-alpha/2,lower.tail = FALSE))))
alpha2 <- pnorm(z0 + (z0+qnorm(p = alpha/2,lower.tail = FALSE))/(1-a*(z0+qnorm(p = alpha/2,lower.tail = FALSE))))

print(quantile(MLE.sort, c(alpha1, alpha2)))
## 8.477838%  99.7565% 
##  60.70423 207.77469
q <- quantile(MLE.sort, c(alpha1, alpha2))
cat("置信区间为[",q[1],",",q[2],"].",sep="")
## 置信区间为[60.70423,207.7747].

置换检验

R函数cor 中提供了三种不同的相关性度量,其中 method="spearman" 提供的是Spearman秩相关性检验统计量。

  1. 产生一组样本量 \(n=20\) 的数据如下(种子设置为 set.seed(123)): \[X_i \sim N(0,1), \quad Y_i =\exp(X_i), \quad i=1,\dots,n.\]
set.seed(123)
n <- 20
x <- rnorm(n, 0, 1)
y <- exp(x)
  1. 用置换检验方法计算上述样本的显著性水平,其中置换次数设置为199。
#采取spearman检验  仿照独立性检验计算p值和对独立性进行检验
cor.obs <- cor(x,y,method="spearman")
cat("The obtained correlation is ",cor.obs,'\n')
## The obtained correlation is  1
nreps<-199
cor.perm <- numeric(nreps)
for (i in 1:nreps) {
  Y <- y
  X <- sample(x, n, replace = FALSE)
  cor.perm[i] <- cor(X,Y,method="spearman")
}
p_0 <- mean(abs(cor.obs) <= abs(c(cor.obs, cor.perm)))   #pdf P7 p-值计算公式
cat("采用spearman检验得到显著性水平为",p_0,".\n",sep="")
## 采用spearman检验得到显著性水平为0.005.
#采取spearman检验  仿照独立性检验采取置换检验
cor_1.obs <- cor(x,y,method="spearman")
cat("The obtained correlation is ",cor_1.obs,'\n')
## The obtained correlation is  1
nreps<-199
cor_1.perm <- numeric(nreps)
for (i in 1:nreps) {
  tmp <- sample(c(x,y))
  cor_1.perm[i] <- cor(tmp[1:20],tmp[21:40],method="spearman")
}
p_1 <- mean(abs(cor_1.obs) <= abs(c(cor_1.obs, cor_1.perm)))   #pdf P7 p-值计算公式
cat("采用spearman检验得到显著性水平为",p_1,".\n",sep="")
## 采用spearman检验得到显著性水平为0.005.
  1. 比较上述显著性水平和使用 cor.test函数得到的p值,可用表格汇总或者画图展示。
#将第二题过程设置不同的种子重复模拟100次,与相同100次模拟下的cor.test进行对比  表格呈现结果
d <- data.frame("cor"=seq(1,100,1),"cor.test"=seq(1,100,1))
for(count in 1:100){
  set.seed(count)
  n <- 20
  x <- rnorm(n, 0, 1)
  y <- exp(x)
  cor.obs <- cor(x,y,method="spearman")
  #cat("The obtained correlation(by cor) is ",cor.obs,'\n')
  nreps<-199
  cor.perm <- numeric(nreps)
  for (i in 1:nreps) {
    Y <- y
    X <- sample(x, n, replace = FALSE)
    cor.perm[i] <- cor(X,Y,method="spearman")
  }
  p_0 <- mean(abs(cor.obs) <= abs(c(cor.obs, cor.perm)))
  
#采用cor.test
  p_1 <- cor.test(x,y,method="spearman")$p.value   #直接用p值
  cor_1.obs <- cor.test(x,y,method="spearman")$estimate
  #cat("The obtained correlation(by cor.test) is ",cor_1.obs,'\n')
  
  d[count,1] <- p_0;d[count,2] <- p_1
}
print(d)
##       cor     cor.test
## 1   0.005 5.976356e-06
## 2   0.005 5.976356e-06
## 3   0.005 5.976356e-06
## 4   0.005 5.976356e-06
## 5   0.005 5.976356e-06
## 6   0.005 5.976356e-06
## 7   0.005 5.976356e-06
## 8   0.005 5.976356e-06
## 9   0.005 5.976356e-06
## 10  0.005 5.976356e-06
## 11  0.005 5.976356e-06
## 12  0.005 5.976356e-06
## 13  0.005 5.976356e-06
## 14  0.005 5.976356e-06
## 15  0.005 5.976356e-06
## 16  0.005 5.976356e-06
## 17  0.005 5.976356e-06
## 18  0.005 5.976356e-06
## 19  0.005 5.976356e-06
## 20  0.005 5.976356e-06
## 21  0.005 5.976356e-06
## 22  0.005 5.976356e-06
## 23  0.005 5.976356e-06
## 24  0.005 5.976356e-06
## 25  0.005 5.976356e-06
## 26  0.005 5.976356e-06
## 27  0.005 5.976356e-06
## 28  0.005 5.976356e-06
## 29  0.005 5.976356e-06
## 30  0.005 5.976356e-06
## 31  0.005 5.976356e-06
## 32  0.005 5.976356e-06
## 33  0.005 5.976356e-06
## 34  0.005 5.976356e-06
## 35  0.005 5.976356e-06
## 36  0.005 5.976356e-06
## 37  0.005 5.976356e-06
## 38  0.005 5.976356e-06
## 39  0.005 5.976356e-06
## 40  0.005 5.976356e-06
## 41  0.005 5.976356e-06
## 42  0.005 5.976356e-06
## 43  0.005 5.976356e-06
## 44  0.005 5.976356e-06
## 45  0.005 5.976356e-06
## 46  0.005 5.976356e-06
## 47  0.005 5.976356e-06
## 48  0.005 5.976356e-06
## 49  0.005 5.976356e-06
## 50  0.005 5.976356e-06
## 51  0.005 5.976356e-06
## 52  0.005 5.976356e-06
## 53  0.005 5.976356e-06
## 54  0.005 5.976356e-06
## 55  0.005 5.976356e-06
## 56  0.005 5.976356e-06
## 57  0.005 5.976356e-06
## 58  0.005 5.976356e-06
## 59  0.005 5.976356e-06
## 60  0.005 5.976356e-06
## 61  0.005 5.976356e-06
## 62  0.005 5.976356e-06
## 63  0.005 5.976356e-06
## 64  0.005 5.976356e-06
## 65  0.005 5.976356e-06
## 66  0.005 5.976356e-06
## 67  0.005 5.976356e-06
## 68  0.005 5.976356e-06
## 69  0.005 5.976356e-06
## 70  0.005 5.976356e-06
## 71  0.005 5.976356e-06
## 72  0.005 5.976356e-06
## 73  0.005 5.976356e-06
## 74  0.005 5.976356e-06
## 75  0.005 5.976356e-06
## 76  0.005 5.976356e-06
## 77  0.005 5.976356e-06
## 78  0.005 5.976356e-06
## 79  0.005 5.976356e-06
## 80  0.005 5.976356e-06
## 81  0.005 5.976356e-06
## 82  0.005 5.976356e-06
## 83  0.005 5.976356e-06
## 84  0.005 5.976356e-06
## 85  0.005 5.976356e-06
## 86  0.005 5.976356e-06
## 87  0.005 5.976356e-06
## 88  0.005 5.976356e-06
## 89  0.005 5.976356e-06
## 90  0.005 5.976356e-06
## 91  0.005 5.976356e-06
## 92  0.005 5.976356e-06
## 93  0.005 5.976356e-06
## 94  0.005 5.976356e-06
## 95  0.005 5.976356e-06
## 96  0.005 5.976356e-06
## 97  0.005 5.976356e-06
## 98  0.005 5.976356e-06
## 99  0.005 5.976356e-06
## 100 0.005 5.976356e-06
#将第二题过程设置不同的种子重复模拟100次,与相同100次模拟下的cor.test进行对比  表格呈现结果
for(count in 1:100){
  set.seed(count)
  n <- 20
  x <- rnorm(n, 0, 1)
  y <- exp(x)
  cor.obs <- cor(x,y,method="spearman")
  #cat("The obtained correlation(by cor) is ",cor.obs,'\n')
  nreps<-199
  cor.perm <- numeric(nreps)
  p_0 <- numeric(100)
  p_1 <- numeric(100)
  for (i in 1:nreps) {
    Y <- y
    X <- sample(x, n, replace = FALSE)
    cor.perm[i] <- cor(X,Y,method="spearman")
  }
  p_0[count] <- mean(abs(cor.obs) <= abs(c(cor.obs, cor.perm)))
  
#采用cor.test
  p_1[count] <- cor.test(x,y,method="spearman")$p.value   #直接用p值
}
hist(p_0,main="cor方法检验",xlim=c(-1,1))

hist(p_1,main="cor.test方法检验",xlim=c(-1,1))