2.4 Coercion Rule

At this point, we have introduced three types of atomic vectors: numeric (Section 2.1) (containing double and integer types), character (Section 2.2), and logical vectors (Section 2.3). Recall that by definition, the atomic vectors always contain elements of a single type. You may wonder what will happen if we try to create a vector with values of mixed types and that is exactly what we are going to answer in this section.

2.4.1 Coercion with c()

When we supply arguments of different types in the c() function, R will unify all elements into the most complex type, which is usually called the coercion rule. Specifically, R uses the following order of complexity (from simple to complex). \[\mbox{logical} < \mbox{numeric} < \mbox{character}\]

Let’s see a few examples to learn how the coercion works. The first example mixes logical values with numbers.

mix_1 <- c(TRUE, 7, 4, FALSE)
mix_1
#> [1] 1 7 4 0
typeof(mix_1)
#> [1] "double"
class(mix_1)
#> [1] "numeric"

You can see that the logical values are converted to numbers, in particular, TRUE is converted to 1 and FALSE is converted to 0, when they mix with numbers. At the end, you can see mix_1 is a numeric vector with four numbers. The result of coercion can be confirmed with the typeof() and class() function.

The second example mixes numbers with strings.

mix_2 <- c("today", "is", "Jan", 15, "2022")
mix_2
#> [1] "today" "is"    "Jan"   "15"    "2022"
class(mix_2)
#> [1] "character"
typeof(mix_2)
#> [1] "character"

You can see that both 15 and 2022 are converted into strings since strings are more complex than numbers. Then, mix_2 will be a character vector.

The next example mixes logical values, numbers and strings.

mix_3 <- c(16, TRUE, "pig")
mix_3
#> [1] "16"   "TRUE" "pig"
class(mix_3)
#> [1] "character"

You can see in both mix_3 and mix_4, both 16 and TRUE are converted to strings! That’s because values of character type are the most complex among all values.

Next, let’s see an interesting example in which we have two layers of coercion.

mix_4 <- c(c(16, TRUE), "pig")
mix_4
#> [1] "16"  "1"   "pig"

Nested c() will collapse into a single vector recursively and during the process, the coercion rule will apply whenever needed. First, c(16, TRUE) will be converted to c(16, 1) since numbers are more complex than logical values. Then, the expression becomes c(c(16, 1), "pig"). Since characters are more complex than numbers, c(16, 1) will be converted to c("16", "1") when you combine it with "pig", leading to the results of mix_5. To help you understand the process, let’s look at another example.

mix_5 <- c(16, c(TRUE, "pig"))
mix_5
#> [1] "16"   "TRUE" "pig"

Here, the first layer is c(TRUE, "pig") which is coerced to c("TRUE", "pig"). Then, 16 will be coerced to "16" in c(16, c("TRUE", "pig")), leading to the final result. The difference between mix_4 and mix_5 reflects the sequential coercion process.

Lastly, let’s talk about the coercion within numeric values. In particular, we have learned that there are two kinds of types numeric values are stored: namely integers and doubles. In the coercion rule, we have \[\mbox{integer} < \mbox{double}.\]

Let’s see the following examples.

typeof(c(1, 5L))
#> [1] "double"
typeof(c(TRUE, 5L))
#> [1] "integer"

Let’s now summarize the coercion rule of all types we have learned. \[\mbox{logical} < \mbox{integer} < \mbox{double} < \mbox{character}\]

2.4.2 Cocercion in operators

In addition to appearing in the vector creation using the c() function, the coercion rule also applies when we apply operators with different types.

typeof(1L + 3)
#> [1] "double"
2 + TRUE + FALSE + TRUE
#> [1] 4
(TRUE * 30 + FALSE * 29)/2
#> [1] 15

R is very smart in whether to apply the coercion.

typeof(1L + 5L)
#> [1] "integer"
typeof(1L * 5L)
#> [1] "integer"
typeof(1L/5L)
#> [1] "double"

When comparing vectors of different types, the coercion rule also will apply. Take the following vectors as example:

a <- c(-1, 0, 1)
b <- c(TRUE, FALSE, TRUE)
a == b
#> [1] FALSE  TRUE  TRUE
# > [1] FALSE TRUE TRUE

When we are trying to figure out if a is equal to b, we get a length-3 logical vector. Elements in b is first converted into 1 and 0 based on the coercion rule we previously introduced, which is TRUE = 1 and FALSE = 0, and the result will be c(1, 0, 1). Thus, a comparison between a logical and a numeric vector is changed into comparing two numeric vectors.

You can make comparisons between vectors of other types, the following example shows that the classic transitive property in math (\(a = b\) and \(b = c\) imply \(a = c\)) doesn’t hold in R.

1 == TRUE
#> [1] TRUE
TRUE == "TRUE"
#> [1] TRUE
1 == "TRUE"
#> [1] FALSE

2.4.3 Explicit Coercion

Besides the coercion rule which automatically converts all elements into the most complex type, you can also use functions to do the conversion manually. In particular, as.numeric(), as.integer(), as.character(), and as.logical() convert its argument into numeric, integer, character, and logical, respectively.

as.numeric(c(TRUE, FALSE))
#> [1] 1 0
as.character(c(TRUE, FALSE))
#> [1] "TRUE"  "FALSE"
as.integer(c(TRUE, FALSE))
#> [1] 1 0
as.logical(c(1, 0))
#> [1]  TRUE FALSE
as.logical(c("TRUE", "FALSE"))
#> [1]  TRUE FALSE

2.4.4 Exercises

1. Looking at the following codes without running in R, what are the storage types of mix_1, mix_2, mix_3, mix_4, mix_5, and mix_6? Verify your answers by running the code in R and explain the reason.

int_1 <- 5L
int_2 <- 6L
num_1 <- 2
char_1 <- "pig"
logi_1 <- TRUE
mix_1 <- int_1 + int_2
mix_2 <- int_1 + num_1
mix_3 <- int_1/int_2
mix_4 <- c(num_1, char_1)
mix_5 <- c(num_1, logi_1)
mix_6 <- c(num_1, char_1, logi_1)

2. If logi_2 <- c(TRUE, FALSE, TRUE) and logi_3 <- TRUE, what are the values of 3 * logi_2 + logi_3 and logi_2 - logi_3?