summarize_at和细分分组变量

Question

如何使用summarize_at实现这一目标？

数据：

library(dplyr)
set.seed(100)
test_df <- data.frame(var_name=c(rep(LETTERS[1:3],each=3),"C"),
                      group_name=c(1,1,0,0,1,0,1,1,1,1),
                      obs_1=rnorm(10),
                      obs_2=rnorm(10))

我想要达到的目标：

test_df %>%
  group_by(var_name) %>%
  summarise(delta_obs1 = median(obs_1[group_name==1])-median(obs_1[group_name==0]),
            delta_obs2 = median(obs_2[group_name==1])-median(obs_2[group_name==0]),
            n_group1   = length(which(group_name==0)),
            n_group0   = length(which(group_name==1)))

# A tibble: 3 x 5
  var_name delta_obs1 delta_obs2 n_group1 n_group0
    <fctr>      <dbl>      <dbl>    <int>    <int>
1        A -0.1064135  0.2947143        1        2
2        B -0.4857362 -0.2318824        2        1
3        C         NA         NA        0        4

但是，如果有许多列(如我真实的例子)，这是相当混乱和乏味的。

我无法工作的summarize_at 版本：

fun_obs_median <-
  function(x) {
    median(x[.$group_name == 1]) - median(x[.$group_name == 0])
  }

test_df %>%
  group_by(var_name) %>%
  summarize_at(.vars = colnames(.)[3:4],
               .funs=fun_obs_median)

Error in summarise_impl(.data, dots) : Evaluation error: object '.' not found.

Answer 1

这将有助于：

library(tidyverse)

set.seed(100)
test_df <- data.frame(var_name=c(rep(LETTERS[1:3],each=3),"C"),
                      group_name=c(1,1,0,0,1,0,1,1,1,1),
                      obs_1=rnorm(10),
                      obs_2=rnorm(10))

# function to calculate delta
delta_f = function(x) x[2]-x[1]

test_df %>%
  group_by(var_name, group_name) %>%                          # for each combination of var and group
  summarise_at(vars(matches("obs")), median) %>%              # get the median for all columns that match "obs"
  arrange(var_name, group_name) %>%                           # for each var get group == 0 in first row and group == 1 in second row
  summarise_at(vars(matches("obs")), funs(delta = delta_f))   # apply delta function

# # A tibble: 3 x 3
#   var_name obs_1_delta obs_2_delta
#   <fct>          <dbl>       <dbl>
# 1 A             -0.106       0.295
# 2 B             -0.486      -0.232
# 3 C             NA          NA 

看起来，arrange()部件是不必要的，因为分组会按照您想要的方式自动排序行。但是，保持这种状态是很好的，以防将来的行为因包更新而发生变化。

对于计数，您可以使用这个

test_df %>%
  mutate(group_name = paste0("n_group", group_name)) %>%
  count(var_name, group_name) %>%
  spread(group_name, n, fill = 0)

# # A tibble: 3 x 3
#   var_name n_group0 n_group1
#   <fct>       <dbl>    <dbl>
# 1 A               1        2
# 2 B               2        1
# 3 C               0        4

然后通过var_name连接这两个表。