Truth, missing values and choices

Comparisons answer a question

n > 10 is not a statement about the program, it is a value: TRUE or FALSE. You can store it, print it and count it. Because TRUE counts as one and FALSE counts as zero, sum(warm) is the number of TRUEs, which is a trick worth knowing on day one.

R also has a third answer. NA means "not known", and it is not the same as FALSE or as zero. A comparison with a missing value comes back as NA, because the honest answer to "is this unknown number bigger than ten" is that nobody knows yet. NA is not the string "NA" and not NaN either: NaN is a number that could not be computed, while NA is a value that was never known.

One test, every value

Because a comparison works element by element, temps > 24 gives a vector of TRUE and FALSE, one per temperature. That vector is a tool: R can count it, find positions in it, or use it to pick rows out of a table, which is exactly what the next chapters do.

which(test) turns it into positions. any(test) and all(test) collapse it to a single answer. And %in% asks a different kind of question: is this value one of those, as in day %in% c("Mon", "Tue"). which drops the NA positions rather than reporting them, so the count it gives can be smaller than the vector suggests; sum(test, na.rm = TRUE) is the count of every TRUE.

Deciding with if and else

if (n > 10) { ... } else { ... } runs one block or the other. The braces hold a group of lines; the condition must be a single TRUE or FALSE, so a whole vector will not do. Add else if when there are three or more cases.

R will not police your style here, but the shape above is the one that scales: braces on the same line as the keyword, and the branch you expect most often first.

A test, a count, and a branch

temps <- c(19, 25, 31, 18, 27)
warm <- temps > 24

sum(warm)
which(warm)
any(warm)
all(warm)

if (sum(warm) > 3) {
  cat("a warm week\n")
} else {
  cat("a mixed week\n")
}

# Missing values, and the test for them.
ages <- c(10, NA, 30)
is.na(ages)
sum(ages, na.rm = TRUE)

Two kinds of and

R has two pairs of logical operators and they are not interchangeable. & and | work element by element, so temps > 24 & temps < 30 gives one flag per temperature. && and || look only at the first element and stop as soon as the answer is decided; they are for a single condition, such as the test in an if.

Confusing them is a quiet bug. A filter written with && keeps or drops the whole vector based on its first value, and recent versions of R warn about it. If the thing on either side has a length above one, use the single-character form.

There is one more spelling worth knowing: xor(x, y) is TRUE when exactly one side is true. It is rare in filters, but it is the honest operator for "one or the other, not both", and seeing it once saves you from building the same thing out of two & tests.

The third answer

NA makes logic three-valued. TRUE & NA is NA, because the result depends on the unknown; FALSE & NA is FALSE, because false and anything is false. TRUE | NA is TRUE for the same reason, while FALSE | NA is NA. That is why is.na exists: it is the only test that answers TRUE or FALSE about a missing value.

ifelse(test, yes, no) is the vectorised branch. It takes a logical vector and returns a value for every position, which is what you want for a column. A plain if is for one decision, and its condition has to be length one; when you have a vector, reduce it with any, all or sum first.

Try these next

• Count how many values in a vector are above their mean, using sum(x > mean(x)).
• Replace the missing values in a vector with zero using ifelse(is.na(x), 0, x).
• Write a branch that prints "all warm" only when all(temps > 20, na.rm = TRUE) is true.