多项摘要输出到R中的数据框

Question

我在下面创建了一个多项逻辑回归：

我已将系数、标准误差、z 统计量和 p 值存储在不同的变量中。 我正在尝试创建一个数据框来存储所有这些信息。

library(foreign)
library(nnet)
ml <- read.dta("https://stats.idre.ucla.edu/stat/data/hsbdemo.dta")
ml$prog2 <- relevel(ml$prog, ref = "academic")
test <- multinom(prog2 ~ 1, data = ml)

coeff <- summary(test)$coefficients
std.errs <- summary(test)$standard.errors

# calculate z-statistics of coefficients
z_stats <- summary(test)$coefficients/
  summary(test)$standard.errors

# convert to p-values
p_values <- (1 - pnorm(abs(z_stats)))*2

> std.errs
         (Intercept)
general    0.1781742
vocation   0.1718249
> coeff
         (Intercept)
general   -0.8472980
vocation  -0.7419374
> p_values
          (Intercept)
general  1.980067e-06
vocation 1.574605e-05
> z_stats
         (Intercept)
general    -4.755448
vocation   -4.317984

这是我的预期输出。如何从我创建的变量创建这个数据框？我在使用 (Intercept) 列创建 general_intercept 时遇到问题。我希望通过 dplyr 来完成。

feature             coefficient std.error  pval z.stat
general_intercept   -0.8472980  0.1781742  1.980067e-06  -4.755448       
vocation_intercept  -0.7419374  0.1718249  1.574605e-05 -4.317984
         

我需要它对变量的数量也是动态的。 例如，如果我运行这个多项逻辑回归

multinom(prog ~ ses + write, data=mdata)

我会得到这个表结构：

feature             coefficient  std.error p.val z_stat
academic_intercept         
academic_sesmiddle              
academic_seshigh
vocation_intercept 
vocation_sesmiddle
vocation_seshigh

Answer 1

df <- cbind(coeff, std.errs, p_values, z_stats)
colnames(df) <- c("coefficient", "SE", "p_val", "z_stat")

         coefficient        SE        p_val    z_stat
general   -0.8472980 0.1781742 1.980067e-06 -4.755448
vocation  -0.7419374 0.1718249 1.574605e-05 -4.317984

如果要重命名行索引：

rownames(df) <- c("general_intercept", "vocation_intercept")

                   coefficient        SE        p_val    z_stat
general_intercept   -0.8472980 0.1781742 1.980067e-06 -4.755448
vocation_intercept  -0.7419374 0.1718249 1.574605e-05 -4.317984

要动态更改行名，您可以：

rownames(df) <- paste(rownames(df), tolower(gsub("[^a-zA-Z]", "", colnames(coeff))), sep = "_")

                   coefficient        SE        p_val    z_stat
general_intercept   -0.8472980 0.1781742 1.980067e-06 -4.755448
vocation_intercept  -0.7419374 0.1718249 1.574605e-05 -4.317984

编辑：有了新的要求，我们可以编写一个可以做到这一点的函数。这有点单一，可能会分成几个功能。请注意，为了使其工作，对象必须作为命名列表传入。编写此函数是为了使用您已经定义的各种统计信息。 coeff <- summary(test)$coefficients。除了你使用的包外，还需要reshape2和dplyr。

create_summary = function(stats) {
  if (!is.list(stats)) {
    stop("Argument must be a named list")
  }
  df = NULL
  for (i in 1:length(stats)) {
    stat = as.data.frame(stats[[i]])
    stat$feature = rownames(stat)
    stat = stat %>%
      melt(id.vars = c("feature")) %>%
      mutate(feature = paste(feature,
                             tolower(gsub(pattern = "[^a-zA-Z]",
                                          replacement = "",
                                          variable)),
                             sep = "_")) %>%
      select(feature, value)

    colnames(stat) = c("feature", names(stats)[i])
    if (is.null(df)) {
      df = stat
    } else {
      df = merge(df, stat, by = "feature")
    }
  }
  return(df)
}

用初始模型test <- multinom(prog2 ~ 1, data=ml)：

create_summary((list(coefficients = coeff,
                     se = std.errs,
                     p_val = p_values,
                     z_stat = z_stats)))

             feature coefficients        se        p_val    z_stat
1  general_intercept   -0.8472980 0.1781742 1.980067e-06 -4.755448
2 vocation_intercept   -0.7419374 0.1718249 1.574605e-05 -4.317984

使用您的第二个模型test <- multinom(prog ~ ses + write, data=ml)：

             feature coefficients         se      p_val     z_stat
1 academic_intercept  -2.85197258 1.16643741 0.01448407 -2.4450284
2   academic_seshigh   1.16282574 0.51422150 0.02373868  2.2613324
3 academic_sesmiddle   0.53329140 0.44373191 0.22942845  1.2018324
4     academic_write   0.05792480 0.02141092 0.00682251  2.7053858
5 vocation_intercept   2.36609732 1.17425127 0.04390634  2.0149838
6   vocation_seshigh   0.18021764 0.64845085 0.78107356  0.2779203
7 vocation_sesmiddle   0.82463840 0.49012366 0.09246981  1.6825109
8     vocation_write  -0.05567514 0.02333135 0.01701977 -2.3862803

您可以继续添加功能，摘要的长度会增加。