在汇总统计表中添加更高级别的分组

Question

前段时间问了如何做分组汇总table：

我想做与此类似的事情，但还需要几步，但我不确定如何进行。

这是我目前的情况：

data %>%
  dplyr::filter_all(all_vars(!is.na(.))) %>%
  group_by(Type.Time, Type.Perc, Grp) %>%
  dplyr::summarise(mean.ms = sprintf("%.2f", mean(Time, na.rm = TRUE)),
                   se.ms = sprintf("%.2f", (sd(Time, na.rm = T))/sqrt(data %>% filter(Grp == 1) %>% nrow())),
                   mean.perc = sprintf("%.2f", mean(Percentage, na.rm = TRUE)),
                   se.perc = sprintf("%.2f", (sd(Percentage, na.rm = T))/sqrt(data %>% filter(Grp == 1) %>% nrow())),
                   ) %>%
  gather(key, value, mean.ms:se.perc) %>%
  unite(Group, Grp, key) %>%
  spread(Group, value)

这给了我想要的信息，但格式错误且值是原来的两倍：

| Type.Time | Type.Perc | 1_mean.ms | 1_mean.perc | 1_se.ms | 1_se.perc | 2_mean.ms | 2_mean.perc | 2_se.ms | 2_se.perc|
|-----------|-----------|-----------|-------------|---------|-----------|-----------|-------------|---------|----------|
| TType2    | PType2    | 703       | 15          | 15      | 1.4       | 573       | 8           | 22      | 1.3      |       
| TType2    | PType1    | 703       | 10          | 15      | 1.8       | 573       | 13          | 22      | 3.1      |
| TType1    | PType2    | 710       | 15          | 18      | 1.4       | 622       | 8           | 29      | 1.3      |
| TType1    | PType1    | 710       | 10          | 18      | 1.8       | 622       | 13          | 29      | 3.1      |

我希望我的新 table 中的顶部分组是 'mean'/'se' 之前的 1 或 2（即 Grp [Group]）。然后是 Type1 和 Type 2 的子组，前面的 T 和 P 被拆分为行（分别为 ms 和 %）...所以我的目标是生成一个 table 这种格式：

     |         Group1         |          Group2           |
     |------------------------|---------------------------|
     |    Type1   |   Type2   |    Type1   |    Type2     |
     |------------|-----------|------------|--------------|
     |   M |  SE  |  M  | SE  |   M  | SE  |   M  |  SE   |
|----|-----|------|-----|-----|------|-----|------|-------|
|ms  | [values calculated from 'Time' variable]           |
|%   | [values calculated from 'Percentage' variable]     |

我希望这是有道理的！

示例数据：

structure(list(ID = c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 
2L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L), Grp = c(1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), Type.Time = c("TType1", 
"TType1", "TType2", "TType2", "TType1", "TType1", "TType2", "TType2", 
"TType1", "TType1", "TType2", "TType2", "TType1", "TType1", 
"TType2", "TType2"), Time = c(711, 711, 669, 669, 765, 765, 876, 876, 740, 
740, 658, 658, 456, 456, 423, 423), Type.Perc = c("PType1", 
"PType2", "PType1", "PType2", "PType1", "PType2", 
"PType1", "PType2", "PType1", "PType2", "PType1", 
"PType2", "PType1", "PType2", "PType1", "PType2"
), Percentage = c(8, 3, 9, 7, 19, 22, 30, 21, 10, 5, 10, 5, 8, 7, 
13, 5)), row.names = c(NA, -16L), class = c("tbl_df", 
"tbl", "data.frame"))

Answer 1

配置此类 header 分组的一个选项是使用 kableExtra 包。

对于数据准备，我做了两个主要更改 - 仅考虑 Type.Time == Type.Perc（以避免问题中显示的过多组合），并计算每个类型和组的 SE 值（在示例代码中混合不同的分组，我认为这不是故意的）。

library(tidyverse)
df <- data %>%
  dplyr::filter_all(all_vars(!is.na(.))) %>%
  dplyr::mutate(
    Type = stringr::str_extract(Type.Time, "Type[0-9]"),
    Type.Perc = stringr::str_extract(Type.Perc, "Type[0-9]"),
  ) %>%
  dplyr::filter(Type == Type.Perc) %>%
  dplyr::select(-Type.Perc, -Type.Time, -ID) %>%
  pivot_longer(c(Percentage, Time), names_to = "parameter") %>%
  group_by(Type, Grp, parameter) %>%
  dplyr::summarise(
    mean = sprintf("%.2f", mean(value, na.rm = TRUE)),
    se = sprintf("%.2f", (sd(value, na.rm = T))/sqrt(n())),
    .groups = "drop"
  ) %>%
  tidyr::pivot_longer(c(mean, se)) %>%
  arrange(Grp, Type) %>%
  tidyr::pivot_wider(id_cols = "parameter", names_from = c("Grp", "Type", "name"))

# A tibble: 2 x 9
  parameter  `1_Type1_mean` `1_Type1_se` `1_Type2_mean` `1_Type2_se` `2_Type1_mean`
  <chr>      <chr>          <chr>        <chr>          <chr>        <chr>         
1 Percentage 13.50          5.50         14.00          7.00         9.00          
2 Time       738.00         27.00        772.50         103.50       598.00        
# ... with 3 more variables: `2_Type1_se` <chr>, `2_Type2_mean` <chr>,
#   `2_Type2_se` <chr>

这些值已经是正确的格式，我们可以简单地用 add_header_above 定义几个 header 分组。并且 kableExtra provides plenty of additional options for modyfing the output format.

library(kableExtra)

kable(df, col.names = c("", "M", "SE", "M", "SE", "M", "SE", "M", "SE"),
      align = c("l", "r", "r", "r", "r", "r", "r", "r", "r", "r"),
      format = "html") %>%
  kable_styling() %>%
  add_header_above(c(" ", "Type1" = 2, "Type2" = 2, "Type1" = 2, "Type2" = 2)) %>%
  add_header_above(c(" ", "Group1" = 4, "Group2" = 4))

在汇总统计表中添加更高级别的分组

Adding higher level groupings in summary statistic tables

r

kableextra