[Feature Request]: a_summary.logical formatting not dealing as a_summary.factor when denom = 0

[imazubi]

Feature description

Hi @shajoezhu ,

When using the summarize_vars function, I realized that when the denominator is 0, summarize_vars() returns NA for logical variables while it returns 0 for factor variables for .stats = count_fraction. I feel both should act in the same way. I am proposing the solution.

See example:

df <- data.frame(var1 = c(TRUE, TRUE, FALSE, FALSE),
                 var2 = c("d", "d", "e", "e"),
                 var3 = c("f", "f", "f", "f"))

df <- df %>% dplyr::mutate_if(is.character,as.factor) %>%
  mutate(var3 = factor(var3, levels = c("f", "g")))

my_table <- basic_table() %>%
  split_cols_by("var3") %>%
  summarize_vars(
    vars = c("var1", "var2")
    ) %>%
  build_table(df = df)

Look at the red and green circles.

The following is the reason why this is happening:

• In s_summary.logical
y$count_fraction <- c(count, ifelse(dn > 0, count / dn, NA))
• In s_summary.factor

y$count_fraction <- lapply(
   y$count,
   function(x) {
     c(x, ifelse(dn > 0, x / dn, 0))
   }
 )

In the logical part we are assigning the NA value if denom is 0, In the factor case we are assigning the 0 value for the same case. Then, if we go to format_count_fraction() which is the function used for formatting, if x[1] or x[2] is NA (which is the case for the logical scenario above), then the output will be NA.

So the solution would be:

• To act in the same in both cases. I would change s_summary.logical to be aligned with s_summary.factor:
  y$count_fraction <- c(count, ifelse(dn > 0, count / dn, NA))

• Don't you think that when showing counts and percentages (xx (xx%)), would be even better the output to be i.e 0/0%, instead of just 0 when denom is 0?

See the new proposal on format_count_fraction() code:

...
result <- if (x[1] == 0 && x[2] != 0) {
  "0"
} else if (x[2] == 0) {
  "0/0%"
} else {
    paste0(x[1], " (", round(x[2] * 100, 1), "%)")
  }...

Code of Conduct

• I agree to follow this project's Code of Conduct.

Contribution Guidelines

• I agree to follow this project's Contribution Guidelines.

Security Policy

• I agree to follow this project's Security Policy.

[Melkiades]

I think this is a fast fix. Should we use just 0 or NA? @edelarua @ayogasekaram @shajoezhu

[shajoezhu]

NA

[Melkiades]

NA

This changes a lot of things. Keeping it in todo. If we want to unify this behavior, we need to be sure that NA is an universal standard for / 0. Can we assume so? @anajens @barnett11 ?

[barnett11]

The actual standards guidance (which was written pre-R onboarding I should caveat) is below,

3.1. Presentation of counts and percentages
3.1.1. When a treatment group has 0 subjects, report “0” (no %) for sample size and all values of categorical variables
3.1.2. When the level of a categorical variable has zero subjects but the denominator is non-zero, report only “0” (no %) for that category.

3.2. Presentation of Summary Statistics
3.2.1. When a treatment group has 0 subjects, report “0” for sample size and report “NE” (not estimable) for all summary statistics
3.2.2. When a treatment group has insufficient number of subjects to compute a summary statistic (e.g. SD based on a single subject), report that statistic as “NE”

[shajoezhu]

@Melkiades sorry for the very brief reply, was meant to get back to this thread, and I was hoping for dig up the rationale we did the way we have now. but got side tracked, and many thanks to @barnett11 , this is exactly the context i was looking for. When I was thinking of NA, I was thinking to specify use the NA_str to modify the "reporting" here.

pernally, i hope we can seperate the actual value we compute here and the values that we reporting here.
it is confusing, and we wont be able to distinguish the case of 0/0 vs 0/10,

back in the days when we first wrote these summary statistics function, rtables did not have the option for na_str display. I think that is why we mixed these computation together.

I think another good examples can be found at, #543 (comment)
and https://github.com/insightsengineering/tlg-catalog/pull/104/files

I was thinking, maybe we can do something more explicit to the reporting option, the NA is caused by "dividing a treatment group of 0", the computation should be mathematical, letting 0/0 be computed as 0 is less ideal I think. but if we are more explicit with what "NA" is displayed as, to users, this option would be more flexible? what do you think? @Melkiades @barnett11 @anajens @khatril

[Melkiades]

If I understood correctly @barnett11 comment, it seems that any summary statistic derived from a 0 population should be NE, while counts should just be 0 (rightfully so). I would put NA for tern to stay general, while having NE in tlg-catalog through na_str display. Does this make sense @barnett11 @anajens?

[shajoezhu]

checked with @barnett11 ,

specify na_str = "NE" for this could work