ddply with .drop=FALSE converts factor .variable to character

[BenBarnes]

When using plyr with .drop = FALSE, I find that factor vectors included in the .variables argument are (at least sometimes) converted to character. The R code below illustrates what I mean:

mydf <- data.frame(aa = factor(rep(1:3, each = 4), levels = 1:4), bb = rep(LETTERS[2:5],
  each = 3))
class(mydf$aa) # factor

ans <- ddply(mydf, .(aa), summarize, uLet = length(unique(bb)), .drop = FALSE)
class(ans$aa) # character

ansDrop <- ddply(mydf, .(aa), summarize, uLet = length(unique(bb)), .drop = TRUE)
class(ansDrop$aa) # factor

## End of R code

As is written in the package's NEWS file, as of version 1.5 empty factor levels are indeed preserved when .drop = FALSE. However, the class of the grouping variable(s) seems to vary depending on the value of this argument. It's no big deal to convert the data back to a factor after running ddply, but perhaps an additional modification to the split_labels function to restore factor classes after expand.grid (where stringsAsFactors = FALSE) would provide more consistent output.

All the best