Problem Statement

Today one of my colleagues posted a question into our R slack community asking why when she was using ifelse that her date field was being converted into a numeric.

date1<-as.Date(c("2019-01-01","2019-01-02","2019-01-03"))
date2<-as.Date(c("2019-01-02","2018-01-02","2019-01-03"))
ifelse(date1>date2,date1,date2)

## [1] 17898 17898 17899

My first instinct was that date objects (and POSIXct objects) are numeric objects that are counting the number of days or seconds since the origin (January 1, 1970 in R). So, comparing the two would result in the values being converted into numeric since comparisons (>,<,==) are not implemented in POSIXct and Date objects.

Before I posted, thankfully someone else correctly identified using if_else would solve my colleagues issue, and the thread moved along.

However, the question was stuck in my craw: WHY was ifelse returning numeric values when the yes/no values were Date objects?

ifelse

Upon inspection of the ifelse function, it was this set of lines really helped me understand the issue:

ifelse <- function(test,yes,no){
  ...
  ans <- test
  ok <- !is.na(test)
  if (any(test[ok])) 
     ans[test & ok] <- rep(yes, length.out = length(ans))[test & ok]
  if (any(!test[ok])) 
     ans[!test & ok] <- rep(no, length.out = length(ans))[!test & ok]
  ...
}

The crux is that ans is a copy of the test argument, which is a logical vector. When you are trying to write into ans, R will try to find the “lowest” common object type between the assigned new value and the current existing values.

In the case of my colleague, it was numeric. Since Date objects are actually numeric values, and logical values can also be seen as 1 / 0 values for TRUE/FALSE, R converts the Date objects and the logical values into numeric. This can be demonstrated using the code below:

logical_vector<-c(TRUE,TRUE,FALSE)
newvalue<-as.Date("2019-01-01")
print(newvalue)

## [1] "2019-01-01"

logical_vector[2]<-newvalue

print(logical_vector)

## [1]     1 17897     0

Fortunately there is a solution - You can convert the numeric values on the other end of the ifelse back into date objects using the as.Date(x,origin=‘1970-01-01’). However, this adds complexity to the workflow that might not be desired.

if_else

if_else is a “Vectorised if” according to the documentation and is more strict than the base ifelse function by strictly checking that the true/false outputs are the same type. In addition, it will give you a warning if the input is invalid. The implementation of if_else is the same as your base ifelse, so the learning curve should not be large, just relearning to type the new function.

Some of the benefits of if_else is that the strictness enforces good behavior and programming practices. Base ifelse will not. For example, you are not allowed to pass inputs that are a different length than the condition you are checking:

dplyr::if_else(TRUE,c(1,2,3),c(4,5,6))
  # Error: `true` must be length 1 (length of `condition`), not 3
  # Call `rlang::last_error()` to see a backtrace
  
ifelse(TRUE,c(1,2,3),c(4,5,6))
 # [1] 1

or for mismatched types for args:

dplyr::if_else(c(TRUE,FALSE),"1",2)
  # Error: `false` must be a character vector, not a double vector
  # Call `rlang::last_error()` to see a backtrace
  
ifelse(c(TRUE,FALSE),"1",2)
  # [1] "1" "2"

Conclusion

Hopefully this little demonstration helped illustrate why you get “interesting” values when you are performing some ifelse’s in your code, using tidyverse or not. Being critical of our code and the results help bring clarity to why certain behaviors exist.

Luckily for us, ifelse usually works within the tidyverse pipe workflow in our mutates without the strange quirks that I outlined above. However, if you do come across this issue, try the preferred if_else!

ret<-function(x,bools){
  if(length(x)==1){
    x
  }else if(length(x)==length(bools)){
    x[bools]
  }else{
    stop("inputs need to be the same length as `test`")
  }
}

ife<-function(test,yes,no){
  ans<-vector("list",length(test))
  ok<-!is.na(test)
  if(any(test[ok]))
    ans[test & ok]<-as.list(ret(yes,test & ok))
  if(any(!test[ok]))
    ans[!test & ok]<-as.list(ret(no,!test & ok))
  if(all(lengths(ans)==1)){
    ans<-do.call('c',ans)
  }
  return(ans)
}

ife(date1>date2,date1,date2)

## [1] "2019-01-02" "2019-01-02" "2019-01-03"