Missing Values in R

Something interesting I found out about NA values in R. Typically, this won’t cause any bug or undesired results in data analysis, but there are certain cases that this should be aware of, especially when using rpy2 python package to call R code from a python script.

The type coercion

As opposed to python’s None is registered as a None type, R’s NA does not have a distinct type, and surprisingly it’s a logical type. The class is logical in a pure NA vector as well:

> class(NA)
[1] "logical"
> class(c(NA, NA, NA))
[1] "logical"

However, this logical type is not like TRUE OR FALSE. It can’t be evaluated:

> if(NA) print('!!!')
Error in if (NA) print("!!!") : missing value where TRUE/FALSE needed

And it will be coerced to any other non-NA types in the vector:

> class(c(2, NA))
[1] "numeric"

You can also coerce itself into other types:

> class(as.numeric(NA))
[1] "numeric"
> class(as.character(NA))
[1] "character"

This flexibility of NA’s typing is actually very useful and convenient in data analysis, where you do not need to worry about type conversion all the time.

A potential bug with rpy2

Normally, this NA type issue won’t cause any problems; however, I found a bug in using pure NA vectors with rpy2 package. When I use pandas2ri.ri2py() function to convert R a data frame to a pandas DataFrame, a pure NA logical vector/column would turn into a Series of min values of type int32, which is -2147483648.

However, if the NA value is pre-converted into numeric or character type in R, the converted pandas DataFrame will be normal, showing as NaN.

This is a caveat to note when using rpy2 to convert data frames from R. If there is a pure NA column, be sure to convert it to numeric or character type beforehand.

Also, there is another trick: R has a constant NA_character_ stored in the global environment, which is essentially a character typed NA. You can simply assign the column with this constant as df$MyNACol <- NA_character_ to replace the default logical NA.