R help - Feb 2003 - rbind.data.frame: character comverted to factor

Dear All,

on rbind:ing together a number of data.frames, I found that
character variables are converted into factors. Since this
occurred for a data identifier, it was a little inconvenient
and, to me, unexpected. (The help page explains the
general procedure used. I also found that on forming 
a data frame, character variables are converted to factors.

The help page on read.table has the 'as.is' argument, which
I suppose kind of suggests that character variables  tend to
get converted into factors. Is there such a "preference" for 
factors and should this behaviour be expected?

Example code

 d1 <- data.frame(id =letters[1:20], x = runif(20))
d2 <- data.frame(id =paste(letters[1:20],letters[1:20], sep = ""), 
x rexp(20))
d3 <- rbind(d1, d2)
str(d1) # <- id is factor
str(d2) # <- id is factor
str(d3) # <- id is factor
d1[["id"]] <- as.character(d1[["id"]])
d2[["id"]] <- as.character(d2[["id"]])
d3 <- rbind(d1, d2)
str(d1) # <- id is character
str(d2) # <- id is character
str(d3) # <- id is factor

Regards,

Markus
-- 
Markus J?ntti <markus.jantti at iki.fi>
Statistics Finland

Read ?data.frame: that tells you to use

d1 <- data.frame(id =I(letters[1:20]), x = runif(20))
d2 <- data.frame(id =I(paste(letters[1:20],letters[1:20], sep =
"")),  x
=rexp(20))
d3 <- rbind(d1, d2)

which of course works!

On 12 Feb 2003, Markus [ISO-8859-1] Jäntti wrote:
> Dear All,
>
> on rbind:ing together a number of data.frames, I found that
> character variables are converted into factors. Since this
> occurred for a data identifier, it was a little inconvenient
> and, to me, unexpected. (The help page explains the
> general procedure used. I also found that on forming
> a data frame, character variables are converted to factors.
as documented in many places, including ?data.frame.
>
> The help page on read.table has the 'as.is' argument, which
> I suppose kind of suggests that character variables  tend to
> get converted into factors. Is there such a "preference" for
> factors and should this behaviour be expected?
It's as documented.
> Example code
>
>  d1 <- data.frame(id =letters[1:20], x = runif(20))
> d2 <- data.frame(id =paste(letters[1:20],letters[1:20], sep =
""),  x > rexp(20))
> d3 <- rbind(d1, d2)
> str(d1) # <- id is factor
> str(d2) # <- id is factor
> str(d3) # <- id is factor
> d1[["id"]] <- as.character(d1[["id"]])
> d2[["id"]] <- as.character(d2[["id"]])
> d3 <- rbind(d1, d2)
> str(d1) # <- id is character
> str(d2) # <- id is character
> str(d3) # <- id is factor
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Brian Ripley wrote:
> d1 <- data.frame(id=I(letters[1:20]), x = runif(20))
> d2 <-
data.frame(id=I(paste(letters[1:20],letters[1:20],sep="")),
>                   x=rexp(20))
> d3 <- rbind(d1, d2)
> 
> which of course works!
Why ``of course''?  It seems to me that there is no ``of
course''
about it.  It is completely counter-intuitive.  What appears
to be going on is that the I() operator NOT ONLY puts its argument
into the data frame ``as is'', but it ALSO tacks a class
``AsIs''
onto that argument which prevents it from being mucked around with
thereafter.

This is a neat trick, but is fairly mysterious --- and could have
intricate ramifications.  How can one discern all the impacts of
an object's having ``AsIs'' as a class?  (It would appear that
objects of any structure and class can ``inherit from'' AsIs.)

It would be highly preferable not to have to use the I() operator, or
the ``AsIs'' class at all.  I.e. to have character vectors stay
character vectors unless the user explicitly asks them to be
converted to factors.  However Splus introduced the contrary policy
years ago, and R is stuck with it for compatibility reasons.


					cheers,

						Rolf Turner
						rolf at math.unb.ca