how to use case_when rather then if_else [Error in my code?]

Question

I'm trying to realize why I cannot use dplyr::case_when rather than dplyr::if_else.
Probably I'm missing something. Let me explain:

I got this operation which works fine:

df %>%
  mutate(
    keep = if_else(
      assembly_level != "Complete Genome" | genome_rep != "Full",
      FALSE,
      ifelse(
        version_status == "suppressed",
        FALSE,
        if_else(
          refseq_category %in% c("reference genome", "representative genome"),
          TRUE,
          if_else(
            rpseudo > 0.4,
            FALSE,
            TRUE
          )
        )
      )
    )
  )

but, when I try using case_when this way

df %>%
  mutate(
    keep = case_when(
      assembly_level != "Complete Genome" | genome_rep != "Full" ~ FALSE, 
      version_status == "suppressed" ~ FALSE,
      refseq_category %in% c("reference genome", "representative genome") ~ TRUE, 
      rpseudo > 0.4 ~ FALSE,
      TRUE ~ TRUE
    )
  )

I got different results.

I think the problem is just the use of the function.

If you need the data, it is a general public data and may be downloaded here: ftp://ftp.ncbi.nlm.nih.gov/genomes/ASSEMBLY_REPORTS/assembly_summary_refseq.txt

to get:

read_tsv("ftp://ftp.ncbi.nlm.nih.gov/genomes/ASSEMBLY_REPORTS/assembly_summary_refseq.txt", 
         comment = "#",
         col_names = c(
           "assembly", "bioproject", "biosample", 
           "wgs_master", "refseq_category", "taxid", 
           "species_taxid", "organism_name", "infraspecific_name", 
           "isolate", "version_status", "assembly_level", 
           "release_type", "genome_rep", "seq_rel_date", 
           "asm_name", "submitter", "gbrs_paired_asm", 
           "paired_asm_comp", "ftp_path", "excluded_from_refseq", "relation_to_type_material"
         )
) %>%
  select(assembly, refseq_category, 
         assembly_level, genome_rep, 
         version_status, release_type) %>%
  mutate(
    rpseudo = runif(nrow(.), 0, 1)
  ) -> df
# it will got some warnings

Thanks in advance,

Answer 1

There are NA's in the data. Store the output from if_else in df1 and the one with case_when in df2. The only difference between df1$keep and df2$keep is df1$keep has got few NAs in them and at those place case_when has got some real values. Check

table(df1$keep, useNA = "always")
# FALSE   TRUE   <NA> 
#156616  10386     79 

table(df2$keep, useNA = "always")
# FALSE   TRUE   <NA> 
#156647  10434      0 

and if you do

(156647 - 156616) + (10434 - 10386) #It gives exactly
#[1] 79

Also if you remove those NA values and then check values in df1 and df2 they are the same.

all(df1$keep[!is.na(df1$keep)] == df2$keep[!is.na(df1$keep)])
#[1] TRUE

---

The way NA is being handled in if_else and case_when is different. Consider this simplified example for better understanding.

library(dplyr)
df <- data.frame(a = c(1:3, NA, 4:7), b = c(NA, letters[1:7]))

Now let's create some random conditions to test. Using if_else

df %>%
  mutate(res = if_else(a > 3, "Yes", 
                   if_else(b == "c", "No", 
                           if_else(a > 5, "Maybe", "Done"))))

#   a    b  res
#1  1 <NA> <NA>
#2  2    a Done
#3  3    b Done
#4 NA    c <NA>
#5  4    d  Yes
#6  5    e  Yes
#7  6    f  Yes
#8  7    g  Yes

However, with case_when you get output as

df %>%
   mutate(res = case_when(a > 3 ~ "Yes", 
                          b == "c"~"No", 
                          a > 5 ~ "Maybe", 
                          TRUE ~ "Done"))

#   a    b  res
#1  1 <NA> Done
#2  2    a Done
#3  3    b Done
#4 NA    c   No
#5  4    d  Yes
#6  5    e  Yes
#7  6    f  Yes
#8  7    g  Yes

So if you notice in if_else if an NA is encountered it returns NA immediately. However, in case_when it treats NA as FALSE so if NA is encountered it goes to next condition until any condition is satisfied or else return value of TRUE.

data

set.seed(1234)
read_tsv("ftp://ftp.ncbi.nlm.nih.gov/genomes/ASSEMBLY_REPORTS/assembly_summary_refseq.txt",
comment = "#",
col_names = c(
       "assembly", "bioproject", "biosample", 
       "wgs_master", "refseq_category", "taxid", 
       "species_taxid", "organism_name", "infraspecific_name", 
       "isolate", "version_status", "assembly_level", 
       "release_type", "genome_rep", "seq_rel_date", 
       "asm_name", "submitter", "gbrs_paired_asm", 
       "paired_asm_comp", "ftp_path", "excluded_from_refseq", "relation_to_type_material"
     )
) %>%
select(assembly, refseq_category, 
      assembly_level, genome_rep, 
     version_status, release_type) %>%
 mutate(
  rpseudo = runif(nrow(.), 0, 1)
 ) -> df