Question

I’m having a problem with a data.frame. To make it very simple I start with

    test<-data.frame(char=character(10), numr=numeric(10))
    test$char[1]<-"ery"

The result is

    Warning message:In `[<-.factor`(`*tmp*`, 1, value = c(NA, 1L, 1L, 1L, 1L, 1L, 1L,  :
   invalid factor level, NAs generated

If I do mode(test$char) I get [1] "numeric"

If I do mode(test$numr) I get [1] "character"

I can also do test$numr[1]<-"fjfj" without an error and the data is stored in that particular place.

If I instead of setting the data.frame with character(10) I just do everything as numeric then as in the previous example it will allow me to change the numeric to character simply by storing a string to something in a column even though it was previously defined as numeric.
Why does R treat character differently than I expect as in my example?

Answer 1

I’m a little suspicious of your results posted above.

>     test<-data.frame(char=character(10), numr=numeric(10))
> str(test)
'data.frame':   10 obs. of  2 variables:
 $ char: Factor w/ 1 level "": 1 1 1 1 1 1 1 1 1 1
 $ numr: num  0 0 0 0 0 0 0 0 0 0
> mode(test$char)
[1] "numeric"
> mode(test$numr)
[1] "numeric"

This is telling me that char is a factor, numr is numeric, and both are stored as numeric (factors have an additional attribute that maps the numeric level codes to labels). You’re getting an error because you’re trying to set a value in char that isn’t included in the list of levels (which includes only the blank string ""). As @GSee says in the comments, you probably wanted stringsAsFactors=FALSE:

> test<-data.frame(char=character(10), numr=numeric(10), 
                   stringsAsFactors=FALSE)
> str(test)
'data.frame':   10 obs. of  2 variables:
 $ char: chr  "" "" "" "" ...
 $ numr: num  0 0 0 0 0 0 0 0 0 0
> test$char[1] <- "ery" ## no error

You can set options(stringsAsFactors=FALSE) to make this your global default behaviour. There is a tradeoff here between convenience for yourself and confusion the next time you forget that you have this option set globally, ask a question on StackOverflow, and have everyone wonder why you’re getting different answers …

Finally, as you mentioned above, if char starts out as numeric, R will silently coerce it to a character string when you try to set an element to a character value. I think this is actually pretty bad design, but it’s too deeply built into R’s behaviour to change now …

> test<-data.frame(char=numeric(10), numr=numeric(10))
> test$char[1] <- "ery"
> str(test)
'data.frame':   10 obs. of  2 variables:
 $ char: chr  "ery" "0" "0" "0" ...
 $ numr: num  0 0 0 0 0 0 0 0 0 0