使用dplyr的across函数后，列名为何显示为pos$cured和prop$cured？

你的代码示例 #

n <- 500
durations <- c(3, 5, 6, 8, 10)
prop_fever_resolve <- c(0.5, 0.6, 0.7, 0.8, 0.9)
prop_urine_sterile <- c(0.8, 0.85, 0.88, 0.9, 0.93)
prop_additional_abx <- c(0.2, 0.17, 0.14, 0.1, 0.05)

duration_props <- data.frame(duration = durations,
                             prop_fever_resolve = prop_fever_resolve,
                             prop_urine_sterile = prop_urine_sterile,
                             prop_additional_abx = prop_additional_abx)

set.seed(6942069)

data <- duration_props %>%
  slice_sample(n = n, replace = TRUE) %>%
  rowwise() %>%
  mutate(fever_resolve = sample(c(FALSE, TRUE), 1, prob = c(1 - prop_fever_resolve, prop_fever_resolve)),
         urine_sterile = sample(c(FALSE, TRUE), 1, prob = c(1 - prop_urine_sterile, prop_urine_sterile)),
         additional_abx = sample(c(FALSE, TRUE), 1, prob = c(1 - prop_additional_abx, prop_additional_abx)),
         cured = fever_resolve & urine_sterile & !additional_abx)

data %>%
  group_by(across("duration")) %>%
  summarize(n = n(),
            pos = across("cured", sum),
            prop = pos / n)

问题原因 #

across()是为批量处理多列设计的，它的返回值是一个嵌套的tibble（数据框），而非单个数值向量。当你在summarize()里写pos = across("cured", sum)时，相当于把cured列的求和结果打包成了一个迷你数据框，后续计算prop = pos / n时，本质是用这个嵌套数据框除以标量n，结果依然是嵌套结构，最终就会出现pos$cured、prop$cured这种表示嵌套列的名称。

另外补充：group_by(across("duration"))其实没必要用across()，直接写group_by(duration)更简洁，因为你只分组单个列。

解决办法 #

方法1：直接不用across（推荐，因为只处理单个列）

既然只对cured一列求和，直接调用sum(cured)即可，完全不需要across()：

data %>%
  group_by(duration) %>%
  summarize(n = n(),
            pos = sum(cured),
            prop = pos / n)

方法2：如果要保留across（比如后续扩展多列）

可以用.names参数指定输出列名，或者提取嵌套中的向量：

• 用.names批量命名：

data %>%
  group_by(duration) %>%
  summarize(n = n(),
            across("cured", sum, .names = "pos"),
            across("pos", ~ .x / n, .names = "prop"))

• 用pull()提取嵌套向量：

data %>%
  group_by(duration) %>%
  summarize(n = n(),
            pos = pull(across("cured", sum)),
            prop = pos / n)

内容的提问来源于stack exchange，提问作者jackahall