即时从 mutate(across()) 中子串化 {.col}

Question

假设我有以下数据：

df <- structure(list(treat = structure(1:4, levels = c("Control", "Alexander Hamilton", 
"Politicians pay attention", "Mark your calendar"), class = "factor"), 
    female_n = c(314709L, 10456L, 10481L, 10455L), female_mean = c(0.506, 
    0.506, 0.504, 0.5), female_sd = c(0.5, 0.5, 0.5, 0.5), birth_year_n = c(314709L, 
    10456L, 10481L, 10455L), birth_year_mean = c(1973.74, 1973.654, 
    1973.486, 1973.766), birth_year_sd = c(16.867, 16.997, 16.869, 
    16.89), provided_phone_no_n = c(314709L, 10456L, 10481L, 
    10455L), provided_phone_no_mean = c(0.656, 0.666, 0.663, 
    0.647), provided_phone_no_sd = c(0.475, 0.472, 0.473, 0.478
    ), dem_n = c(314709L, 10456L, 10481L, 10455L), dem_mean = c(0.48, 
    0.474, 0.482, 0.478), dem_sd = c(0.5, 0.499, 0.5, 0.5), rep_n = c(314709L, 
    10456L, 10481L, 10455L), rep_mean = c(0.136, 0.141, 0.142, 
    0.138), rep_sd = c(0.343, 0.348, 0.349, 0.345), uaf_n = c(314709L, 
    10456L, 10481L, 10455L), uaf_mean = c(0.363, 0.365, 0.357, 
    0.363), uaf_sd = c(0.481, 0.481, 0.479, 0.481)), class = c("tbl_df", 
"tbl", "data.frame"), row.names = c(NA, -4L))

我想添加一个新的

*_se

列，它将数据中每个变量组的现有

*_n

和

*_sd

列作为输入。 IE。

female_*

、

birth_year_*

、

provided_phone_no_*

、

dem_*

、

rep_*

和

uaf_*

各一个。

尝试执行此操作，我认为

mutate(across())

可能是正确的辅助函数，但我在对

{.col}

进行子字符串化并让 R 将其识别为列名称时遇到一些问题。这是我迄今为止的尝试：

df %>%
    mutate(
    across(ends_with("_sd"),
           list(
            se = ~.x / sqrt(!!ensym("{str_replace(.col, '_sd', '_n')}"))
           )
    )

上面返回错误：

Error in `ensym()`:
! `arg` must be a symbol
Backtrace:
  1. ... %>% ...
 10. rlang::abort(message = message)

有人能看出我哪里出错了吗？

Answer 1

这确实是将数据转换为长格式然后再转换回宽格式的理想情况：

library(tidyr)
library(dplyr)

df |>
  pivot_longer(cols = -treat, names_pattern = "(.*)_(.*)", names_to = c("grp", ".value")) |>
  mutate(se = sd / sqrt(n)) |>
  pivot_wider(names_from = grp, values_from = n:se, names_glue = "{grp}_{.value}", names_vary = "slowest")