count NAs with aggregate

Dear members of this list,

I'd like to count missing values using the aggregate function.

Something like this:

count_nas <- function(arg1) {
   return(sum(is.na(arg1)))
}

aggregate(cbind(var1, var2, var3) ~ subject + time, data = mydataset, 
count_nas)


It's not working: I end up with a matrix containing zeros, although 
there are missings in the data frame.
I'd highly appreciate a hint!

Thanks a lot,
Timo Stolz