An R function for checking whether a variable is binary or continuous

Question

This function takes a numeric vector and returns "con" if it's continous and "bin" if it's binary.

It does not take the multinomial case into account, i.e. if a variable y has three possible values 0, 1, 2, it's treated like a continuous variable.

Code:

checkBinaryTrait = function(v, naVal="NA") {
    if(!is.numeric(v)) stop("Only numeric vectors are accepted.")
    vSet = unique(v)
    if(!missing(naVal)) vSet[which(vSet == naVal)] = NA
    vSet = vSet[which(!is.na(vSet))]
    if(any(as.integer(vSet) != vSet)) return("con")
    if(length(vSet) > 2) return("con")
    "bin"
}

Tests:

v = c(1, 1.1, 1, 1.1, NA)
checkBinaryTrait(v)
v = c(1, 2, 1, 2, NA)
checkBinaryTrait(v)
v = c(-9, 2.3, 4.1, -9, -9)
checkBinaryTrait(v, -9)
v = c(-9, 2, 4, -9, -9)
checkBinaryTrait(v, -9)

Answer 1

Using the return statement is not recommended. You can get the same effect by rewriting with else if and else, like this:

checkBinaryTrait = function(v, naVal="NA") {
  if (!is.numeric(v)) stop("Only numeric vectors are accepted.")
  vSet = unique(v)
  if (!missing(naVal)) vSet[vSet == naVal] = NA
  vSet = vSet[!is.na(vSet)]

  if (any(as.integer(vSet) != vSet)) "con"
  else if (length(vSet) > 2) "con"
  else "bin"
}

I also removed the unnecessary which calls. This code still passes all your tests.

Actually the last statement can be further simplified to:

  if (any(as.integer(vSet) != vSet) || length(vSet) > 2) "con"
  else "bin"

You might also want to change the return type to TRUE or FALSE, in which case the last statement would become simply:

  !(any(as.integer(vSet) != vSet) || length(vSet) > 2)

And then, how about renaming checkBinaryTrait to is.binary?

Finally, the <- operator is more common than =. For example Google's style guide explicitly forbids using =.

Answer 2

Let me address some portions of your code before providing an alternative implementation.

• missing(naVal)

  I would prefer not using this approach but an appropriate neutral default value for naVal. We can use NULL for this purpose.

• vSet[which(vSet == naVal)] = NA

  Replacing calues with NA before removing them is an unnecessary step. Furthermore, replacing values with NA is easier with the is.na<- function, for example, is.na(vSet) <- vSet == naVal.

• vSet[which(!is.na(vSet))]

  You can omit NA values with the na.omit function.

---

Here's an alternative implementation. For details, have a look at the comments.

checkBinaryTrait <- function(v, naVal = NULL) { 
  if( !is.numeric(v) ) stop("Only numeric vectors are accepted.")
  # remove NA's
  v2 <- na.omit(v)
  # get unique values
  v_unique <- unique(v2)
  # remove 'naVal's
  v_unique2 <- v_unique[! v_unique %in% naVal]
  # count number of unique values and check whether all values are integers
  if ( length(unique(v_unique2)) > 2L || 
         any(as.integer(v_unique2) != v_unique2) ) "con" else "bin"
}

Some tests:

> checkBinaryTrait(v, -9)
[1] "bin"
> checkBinaryTrait(c(1, 1.1, 1, 1.1, NA))
[1] "con"
> checkBinaryTrait(c(1, 2, 1, 2, NA))
[1] "bin"
> checkBinaryTrait(c(-9, 2.3, 4.1, -9, -9), -9)
[1] "con"
> checkBinaryTrait(c(-9, 2, 4, -9, -9), -9)
[1] "bin"

This implementations also allows multiple naVal values:

> checkBinaryTrait(c(1, 2, 2, 1, -9, -9.9), c(-9, -9.9))
[1] "bin"

Answer 3

I felt factors() was appropriate here. Assuming decimal values are also considered (nothing in the question related to this)

checkBinaryTrait = function(v,naVal = "NA"){
  if (!is.numeric(v)) stop("Only numeric vectors are accepted.")
  if(length(levels(factor(v[-which(v == naVal)]))) < 3) "bin" else "con"
}

If only integers are to be considered, coercing into integers:

checkBinaryTrait = function(v,naVal = "NA"){
  if (!is.numeric(v)) stop("Only numeric vectors are accepted.")
  if(length(levels(factor(as.integer(v[-which(v == naVal)])))) < 3) "bin" else "con"
}