Dealing with NA/NULL/NaN

1 In general

In R there are several type of special values such as NA, NaN and NULL. They are similar but at the same time different especially when directly tested. In general:

• NA is a missing information
• NaN means not a number and is the result of an not-defined math operation like \(\frac{0}{0}\).
• NULL is the absence of any information

2 Type

In terms of data type, NA is logical, NaN is numeric type and NULL do not have a specific type. A particularity of NA is that can be also adapted to the type of vector where is located. In R we cannot have a vector of numbers and logical values:

is.logical(NA)

## [1] TRUE

is.logical(NaN)

## [1] FALSE

is.logical(NULL)

## [1] FALSE

vec <- c(1,2,3,4, TRUE, 5, FALSE)
vec # TRUE is 1 and FALSE is 0

## [1] 1 2 3 4 1 5 0

But we can have a numeric, character or logical vector with NA values and the NA is forced to the specific type:

# NA alone is logical but

vec <- c(1,2,3,4,NA,5,6)
class(vec)

## [1] "numeric"

for(i in vec) print(class(i))

## [1] "numeric"
## [1] "numeric"
## [1] "numeric"
## [1] "numeric"
## [1] "numeric"
## [1] "numeric"
## [1] "numeric"

# now NA is numeric

This is possible because there are several type of NA values that are not usually employed directly but are used from R depending on the vector:

is.character(NA_character_)

## [1] TRUE

is.numeric(NA_real_)

## [1] TRUE

is.integer(NA_integer_)

## [1] TRUE

# This

vec <- c(1,2,3,4,5,NA)
vec

## [1]  1  2  3  4  5 NA

# is the same as

vec <- c(1,2,3,4,5,NA_integer_)
vec

## [1]  1  2  3  4  5 NA

The NaN is a numeric type because is the result of undefined operations:

is.numeric(NaN)

## [1] TRUE

0/0

## [1] NaN

vec <- c(1,2,3,4,NaN)
vec

## [1]   1   2   3   4 NaN

The NULL value is surely the strangest because represents the absence of any value and the class is NULL itself meaning that is different from NA and NaN values.

x <- NULL
class(x) # the class is NULL

## [1] "NULL"

Compared to NA and NaN that are associate with some information (absence of value and undefined number), NULL values are somehow transparent and cannot be stored in vectors:

vec <- c(1,2,3,4,5, NA, 8, NaN, 11, NULL)
vec # the null value is not present

## [1]   1   2   3   4   5  NA   8 NaN  11

3 The problem with NA

As Hadley suggest¹ the NA values infect all other computations. Practically many R functions such as mean() or sd() cannot return a value if there is a NA within the target vector.

x <- c(runif(10), NA)
mean(x)

## [1] NA

sd(x)

## [1] NA

In fact, these functions have a specific argument na.rm = TRUE for doing the computation without considering NA values. Of course we can remove NA values before the computation and the result will be the same:

mean(x, na.rm = TRUE)

## [1] 0.4320439

mean(x[!is.na(x)])

## [1] 0.4320439

• http://www.cookbook-r.com/Basics/Working_with_NULL_NA_and_NaN/
• https://www.projectpro.io/recipes/what-is-difference-between-na-and-nan-r

---

1. Sources

   ↩︎