case_when formula not working as expected

[gorkang]

When using case_when with a formula such as Age == 25 & Brochure == "New", its behavior seems not to be equivalent to filter(Age == 25 & Brochure == "New").

This is not expected (at least I didn't expected it) and can cause problems when there are NAs and we do an arithmetic operation as the replacement value.

Please see a reprex where a minimal case is shown.

library(tidyverse)

Age = c(25, 25, NA)
Brochure = c("New", "Old", "New")
PPV_Cond1 = c(NA, 1, NA)
PPV_Cond2 = c(1, NA, 1)
DF = data.frame(Age, Brochure, PPV_Cond1, PPV_Cond2)
DF
#>   Age Brochure PPV_Cond1 PPV_Cond2
#> 1  25      New        NA         1
#> 2  25      Old         1        NA
#> 3  NA      New        NA         1

# One case with Brochure == "New" have NA in Age. If we filter using Age == 25 & Brochure == "New",
# that row disappears as expected
DF %>% filter(Age == 25 & Brochure == "New")
#>   Age Brochure PPV_Cond1 PPV_Cond2
#> 1  25      New        NA         1

# Despite using the same "filter" here, the NA causes problems
DF %>% mutate(Variable =
                case_when(Age == 25 & Brochure == "Old" ~ PPV_Cond1 - 50,
                          Age == 25 & Brochure == "New" ~ PPV_Cond2 - 50))
#> Error in mutate_impl(.data, dots): Evaluation error: NAs are not allowed in subscripted assignments.

# If we add the !is.na(Age) to the case_when condition, everything works as expected
DF %>% mutate(Variable =
                case_when(Age == 25 & Brochure == "Old" ~ PPV_Cond1 - 50,
                     !is.na(Age) & Age == 25 & Brochure == "New" ~ PPV_Cond2 - 50))
#>   Age Brochure PPV_Cond1 PPV_Cond2 Variable
#> 1  25      New        NA         1      -49
#> 2  25      Old         1        NA      -49
#> 3  NA      New        NA         1       NA

[krlmlr]

Thanks, confirmed. Simpler reprex:

dplyr::case_when(c(1:3, NA) == 1 ~ 1:4)
#> Error in x[i] <- val[i]: NAs are not allowed in subscripted assignments