disabling NA token as na.string in read.table

Vadim Ogranovich <vograno at arbitrade.com> writes:

Dear R-Users,

I have a csv file that has NA tokens and these tokens are perfectly good
values that need not to be converted to NA by read.table(). I tried to
prevent the conversion by specifying the na.strings arg., but this seems to
only add to the list of NA strings, not substitute.

system("cat foo")

system("cat foo")
1 foo
2 NA

read.table("foo", na.strings="foo")

read.table("foo", na.strings="foo")
  V1 V2
1  1 NA
2  2 NA


This is R1.6.0 on Linux.

What did I do wrong?

Hmm, this looks like a bit of a bug. read.table() ends up calling
type.convert() with its default "NA" na.string. Now, if "NA" was in
the na.string for read.table(), scan() would already have turned it
into <NA> at that point, so I suspect you might have preferred
na.strings=character(0), but that has the side effect of turning the
real NA into a factor level:

x <- c(NA,"NA","foo")
type.convert(x)

[1] <NA> <NA> foo
Levels: foo

type.convert(x,na.strings=character(0))

[1] <NA> NA   foo
Levels: NA foo NA

dput(type.convert(x,na.strings=character(0)))

structure(c(3, 1, 2), .Label = c("NA", "foo", NA), class = "factor")

I.e. it looks like the internals of type.convert needs some fixing up.

O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907