Propagate special missing values through dplyr operations etc.

[SchmidtPaul]

Current Behavior

When using naniar's special missing values (via bind_shadow() and recode_shadow()), the missing value types are not propagated through dplyr operations like mutate(). This means we lose information about why a result is NA when performing calculations.

I am aware that this is probably out of scope of what naniar can do, but do you happen to have a suggested approach here?

Example

Here's a minimal reproducible example where the result is completely uninformed about why it is NA.

library(naniar)
library(tidyverse)

tbl <- tibble(val1 = c('12', '*', 'x'), val2 = c('1', '2', 'x'))
cols <- c('val1', 'val2')

tbl_shadow <- tbl %>%  
  bind_shadow() %>% 
  recode_shadow(val1 = .where(val1 == '*' ~ "*", val1 == 'x' ~ 'x')) %>% 
  recode_shadow(val2 = .where(val2 == '*' ~ "*", val2 == 'x' ~ 'x')) %>% 
  replace_with_na(replace = setNames(lapply(cols, function(x) c('*', 'x')), cols)) %>%
  mutate(across(all_of(cols), as.numeric))

tbl_shadow %>%
  mutate(result = val1 / val2) 
#> # A tibble: 3 × 5
#>    val1  val2 val1_NA val2_NA result
#>   <dbl> <dbl> <fct>   <fct>    <dbl>
#> 1    12     1 !NA     !NA         12
#> 2    NA     2 NA_*    !NA         NA
#> 3    NA    NA NA_x    NA_x        NA

^{Created on 2024-12-16 with reprex v2.1.1}