Using ifelse change the result of my calculus

I have a dataset corresponding to different trials in which are tested different treatments. Some trials include a Control group, some do not.

Here is an example where there is a Control group only in the trial 1:

data = data.frame(trial = c(1, 1, 2, 2),
                  treatment = c("control", "b", "b", "c"),
                  value = c(97.2, 99.3, 85.51, 85.01))

When a trial include a Control group, I want to make the ratio between the treatments of this trial and the Control of this trial. I first filtered all the trial with a Control group and it worked well:

data %>%
    filter(trial == 1) %>%
    group_by(trial) %>%
    summarise(treatment = treatment, 
              r = value / value[which(treatment == "control")])

But when I tried to do it without filtering the trials and using the ifelse() function, I got unexpected result, the ratio was always equals to 1 (or NA when no Control group).

data %>%
    group_by(trial) %>%
    summarise(treatment = treatment, 
              r = ifelse("control" %in% treatment, value / value[which(treatment == "control")], NA))

I think you need two corrections -

1. You need mutate instead of summarise if you want to keep the length of dataframe consistent.

2. "control" %in% treatment gives output of length 1. Even if you change it to treatment %in% "control" it will still give incorrect result since the condition is wrong itself.

Here is a shorter solution.

library(dplyr)

data %>%
  group_by(trial) %>%
  mutate(r = value/value[match("control", treatment)])

#  trial treatment value     r
#  <dbl> <chr>     <dbl> <dbl>
#1     1 control    97.2  1   
#2     1 b          99.3  1.02
#3     2 b          85.5 NA   
#4     2 c          85.0 NA   

We are using match because of two reasons here -

1. match ensures that you will always get 1 number even if you have two "control" for a trial. Even if it is highly unlikely it is good to be sure.
2. You can let go of ifelse because match would by default return NA if "control" does not exist.

As per new syntax you can use .by -

data %>%
  mutate(r = value/value[match("control", treatment)], .by = trial)

@RonakShah has explained why your code fails, and I also prefer his match() solution. Another option that minimally modifies your code is to replace ifelse() with if...else...

library(dplyr)

data %>%
  group_by(trial) %>%
  mutate(r = if("control" %in% treatment) value/value[treatment == "control"] else NA)

# # A tibble: 4 × 4
# # Groups:   trial [2]
#   trial treatment value     r
#   <dbl> <chr>     <dbl> <dbl>
# 1     1 control    97.2  1   
# 2     1 b          99.3  1.02
# 3     2 b          85.5 NA   
# 4     2 c          85.0 NA

Here is a base R solution with by.

with(data, by(data[-1], trial, \(X) {
  X$value <- X$value/X$value[match("control", X$treatment)]
  X
})) |> array2DF()
#>     trial treatment    value
#> 1       1   control 1.000000
#> 1.1     1         b 1.021605
#> 2       2         b       NA
#> 2.1     2         c       NA

^{Created on 2023-07-17 with reprex v2.0.2}