case_when doesn't work with multiple conditions over multiple variables

December 20, 2022

I just discovered that, case_when might not work if a variable is recoded based on multiple variables.

Reproducible data:

data <- data.frame(f103 = c(2, NA, NA, 1, 2, 2),
                       f76 = c(2, NA, NA, NA, 3, 3),
                       f4 = c(1,3,3,1,1,2))

The following code produces the same results for var1 and var 2 (which is not what I want):

reprdata <- reprdata %>%
  mutate(var1 = f4) %>% 
  mutate(var1 = case_when(f103 == 2 ~ 3, TRUE ~ as.numeric(var1))) %>%
  mutate(var2 = f4) %>% 
  mutate(var2 = case_when(f103 == 2 ~ 3, f76 == 1 ~ 1, f76 == 2 ~ 2, f76 == 3 ~ 3, TRUE ~ as.numeric(var2)))

The following produces the correct result (i.e., the solution to my problem):

reprdata <- reprdata %>%
  mutate(var1 = f4) %>% 
  mutate(var1 = case_when(f103 == 2 ~ 3, TRUE ~ as.numeric(var1))) %>%
  mutate(var2 = f4) %>% 
  mutate(var2 = case_when(f103 == 2 ~ 3, TRUE ~ as.numeric(var2))) %>%
  mutate(var2 = case_when(f76 == 1 ~ 1, f76 == 2 ~ 2, f76 == 3 ~ 3, TRUE ~ as.numeric(var2)))

(I am aware that in this snippet of my data, the f103 condition for var1 is superfluous, still, I wouldn’t expect it to cause this issue.)

I’d be interested to know if someone can explain to my why this problem occurs and how to prevent it in future.

>Solution :

It has to do with how case_when evaluates: It’s evaluating from the bottom and up, which is contrary to what most people think intuitively (my experience). I.e.

f76 wins (what you expect!)

library(dplyr)

data |>
    mutate(var1 = case_when(f103 == 2 ~ 3,
                            TRUE ~ f4)) |>
    mutate(var2 = case_when(f76 %in% 1:3 ~ f76,
                            f103 == 2 ~ 3, # NB!
                            TRUE ~ f4))

  f103 f76 f4 var1 var2
1    2   2  1    3    2
2   NA  NA  3    3    3
3   NA  NA  3    3    3
4    1  NA  1    1    1
5    2   3  1    3    3
6    2   3  2    3    3

f103 wins (what you don’t expect)

library(dplyr)

data |>
    mutate(var1 = case_when(f103 == 2 ~ 3,
                            TRUE ~ f4)) |>
    mutate(var2 = case_when(f103 == 2 ~ 3, # NB!
                            f76 %in% 1:3 ~ f76 
                            TRUE ~ f4))

  f103 f76 f4 var1 var2
1    2   2  1    3    3
2   NA  NA  3    3    3
3   NA  NA  3    3    3
4    1  NA  1    1    1
5    2   3  1    3    3
6    2   3  2    3    3