Replace <NA> in a factor column

I want to replace <NA> values in a factors column with a valid value. But I can not find a way. This example is only for demonstration. The original data comes from a foreign csv file I have to deal with.

df <- data.frame(a=sample(0:10, size=10, replace=TRUE),
                 b=sample(20:30, size=10, replace=TRUE))
df[df$a==0,'a'] <- NA
df$a <- as.factor(df$a)

Could look like this

      a  b
1     1 29
2     2 23
3     3 23
4     3 22
5     4 28
6  <NA> 24
7     2 21
8     4 25
9  <NA> 29
10    3 24

Now I want to replace the <NA> values with a number.

df[is.na(df$a), 'a'] <- 88
In `[<-.factor`(`*tmp*`, iseq, value = c(88, 88)) :
  invalid factor level, NA generated

I think I missed a fundamental R concept about factors. Am I? I can not understand why it doesn't work. I think invalid factor level means that 88 is not a valid level in that factor, right? So I have to tell the factor column that there is another level?

Solution 1:

1) addNA If fac is a factor addNA(fac) is the same factor but with NA added as a level. See ?addNA

To force the NA level to be 88:

facna <- addNA(fac)
levels(facna) <- c(levels(fac), 88)

giving:

> facna
 [1] 1  2  3  3  4  88 2  4  88 3 
Levels: 1 2 3 4 88

1a) This can be written in a single line as follows:

`levels<-`(addNA(fac), c(levels(fac), 88))

2) factor It can also be done in one line using the various arguments of factor like this:

factor(fac, levels = levels(addNA(fac)), labels = c(levels(fac), 88), exclude = NULL)

2a) or equivalently:

factor(fac, levels = c(levels(fac), NA), labels = c(levels(fac), 88), exclude = NULL)

3) ifelse Another approach is:

factor(ifelse(is.na(fac), 88, paste(fac)), levels = c(levels(fac), 88))

4) forcats The forcats package has a function for this:

library(forcats)

fct_explicit_na(fac, "88")
## [1] 1  2  3  3  4  88 2  4  88 3 
## Levels: 1 2 3 4 88

Note: We used the following for input fac

fac <- structure(c(1L, 2L, 3L, 3L, 4L, NA, 2L, 4L, NA, 3L), .Label = c("1", 
"2", "3", "4"), class = "factor")

Update: Have improved (1) and added (1a). Later added (4).

Solution 2:

other way to do is:

#check levels
levels(df$a)
#[1] "3"  "4"  "7"  "9"  "10"

#add new factor level. i.e 88 in our example
df$a = factor(df$a, levels=c(levels(df$a), 88))

#convert all NA's to 88
df$a[is.na(df$a)] = 88

#check levels again
levels(df$a)
#[1] "3"  "4"  "7"  "9"  "10" "88"

Solution 3:

I had similar issues and I want to add what I consider the most pragmatic (and also tidy) solution:

Convert the column to a character column, use mutate and a simple ifelse-statement to change the NA values to what you want the factor level to be (I have chosen "None"), convert it back to a factor column:

df %>% mutate(
a = as.character(a),
a = ifelse(is.na(a), "None", a),
a = as.factor(a)
)

Clean and painless because you do not actually have to dabble with NA values when they occur in a factor column. You bypass the weirdness and end up with a clean factor variable.

Also, in response to the comment made below regarding multiple columns: You can wrap the statements in a function and use mutate_if to select all factor variables or, if you know the names of the columns of concern, mutate_at to apply that function:

replace_factor_na <- function(x){
  x <- as.character(x)
  x <- if_else(is.na(x), "None", x)
  x <- as.factor(x)
}

df <- df %>%
  mutate_if(is.factor, replace_factor_na)

Solution 4:

The basic concept of a factor variable is that it can only take specific values, i.e., the levels. A value not in the levels is invalid.

You have two possibilities:

If you have a variable that follows this concept, make sure to define all levels when you create it, even those without corresponding values.

Or make the variable a character variable and work with that.

PS: Often these problems result from data import. For instance, what you show there looks like it should be a numeric variable and not a factor variable.