R 中的 Function 计算线性回归的异方差稳健置信区间

Question

下午好，我对下面的 function 有疑问。 任务是在 R 中开发一个 function，用于计算线性回归的 beta 结果的异方差稳健置信区间。

正如我试图这样做的那样，我的 function 没有返回任何 output。 控制台在尝试从中获得一些结果后根本不做任何事情。 我真的很争论为什么特别是如果我通过代码的最后两行手动计算它，它工作得很好。 即使您没有必要的 data.frames，也许您可以查看我的代码并告诉我它有什么问题或提出解决我的问题的替代方法:)

为了清楚起见：系数的原始众多值（每个使用所有 200 个数据点）是 c(463.2121, 139.5762)，stdHC 是 c(74.705054, 5.548689)，由 lm model 给出，对于 HC-robust 标准错误我使用package 三明治。

my_CI <- function (mod, level = 0.95)
{
  `%>%` <- magrittr::`%>%`
  standard_deviation <- stderrorHC
  Margin_Error <- abs(qnorm((1-0.95)/2))*standard_deviation 
  df_out <- data.frame(stderrorHC, mod,Margin_Error=Margin_Error,
                       'CI lower limit'=(mod - Margin_Error),
                       'CI Upper limit'=(mod + Margin_Error)) %>%
    return(df_out)
}

my_CI(mod, level = 0.95) #retrieving does not return any results for me

Definitions:
women <- read.table("women.txt")
men <- read.table("men.txt")
converged <- merge(women, men, all = TRUE)
level <- c(0.95, 0.975)
modell <- lm(formula = loan ~ education, data = converged)
mod <- modell$coefficients
vcov <- vcovHC(modell, type = "HC1")
stderrorHC <- sqrt(diag(vcov))

mod - abs(qnorm((1-level[1])/2))*stderrorHC 
mod + abs(qnorm((1-level[1])/2))*stderrorHC

补充：这是原始数据集中的一些数据。 我只包括了十个数据点，因此在这种情况下，我们需要在 t 分布上构建置信区间。

dataMenEductaion <- c(12, 17, 16, 11, 20, 20 , 11, 19, 15, 16)
dataMenLoan <- c(2404.72, 3075.313, 2769.543, 2009.295, 3105.121, 4269.216
                   2213.730, 4025.136, 2605.191, 2760.186)
dataWomenEducation <- c(12, 14, 16, 19 , 12, 19, 20, 17, 16, 10)
dataWomenLoan <- c(1920.667, 2278.255, 2296.804, 2977.048, 1915.740, 3557.991, 
                   3336.683, 2923.040, 2628.351, 1918.218)

Answer 1

相信下面为您提供了想要的output。

# install.packages('sandwich')
library(sandwich) # contains vcovHC()

# data
df <- data.frame(education = c(12, 17, 16, 11, 20, 20, 11, 19, 15, 16,
                              12, 14, 16, 19 , 12, 19, 20, 17, 16, 10),
                loan = c(2404.72, 3075.313, 2769.543, 2009.295, 3105.121, 4269.216,
                         2213.730, 4025.136, 2605.191, 2760.186,
                         1920.667, 2278.255, 2296.804, 2977.048, 1915.740, 3557.991, 
                         3336.683, 2923.040, 2628.351, 1918.218))
df$sex <- factor(gl(2, nrow(df)/2, labels = c('males', 'females')))

# linear model
fit <- lm(loan ~ education + sex, data = df)
coefs <- fit$coefficients
vcov <- vcovHC(fit, type = "HC1")
stderrorHC <- sqrt(diag(vcov))

# function to compute robust SEs
my_CIs <- function (coefs, level = 0.95) {
  standard_deviation <- stderrorHC
  Margin_Error <- abs( qnorm( (1-level)/ 2) ) * standard_deviation 
  df_out <- data.frame(stderrorHC, coefs, Margin_Error = Margin_Error,
                       'CI lower limit' = (coefs - Margin_Error),
                       'CI Upper limit' = (coefs + Margin_Error))
  return(df_out)
}

Output

> my_CIs(coefs = coefs)
stderrorHC     coefs Margin_Error CI.lower.limit CI.Upper.limit
(Intercept)  295.86900  160.3716    579.89259      -419.5210      740.26416
education     23.64313  176.0111     46.33968       129.6714      222.35073
sexfemales   132.07169 -313.2632    258.85576      -572.1189      -54.40743