[Rd] Problem with read.xport() from foreigh package (PR#7389)

[Werner Engl]

Dear R-devel list,

This is to confirm Prof. Ripley's analysis of the
read.xport issue.

The section on missing data in TS140 is pertinent
to numeric variables only. In SAS, character 
variables are of fixed length (between 1 and 200 
for the xport format). Shorter strings are padded 
with trailing blanks when assigned to a variable.

An uninitialized character variable is stored as 
all blanks in the xport format file. This is the 
only representation of 'missing' data for SAS 
character variables. 'Special missing' codes 
(.A to .Z and ._) are available for numeric 
variables only.

Please find enclosed a patch to the 
R-2.0.1/src/library/Recommended/foreign/SASxport.c
file and a xport file that I used for testing. The
xport file was created by SAS V8.2 on Linux, but 
should be plattform and version independent (except
for the header information). I have simply commented
out the code lines that try to detect missing character
values.

The code in SASxport.c already does a good job in 
removing trailing blanks from character values. 
For missing character data (all blanks) the result 
is the empty string (""), which is fine for me. 
There is no equivalent to the R missing character 
representation in SAS (as far as I know). 

The enclosed gzipped tar file contains:

diff_SASxport_c.txt	diff for SASxport.c
xptchar1.xpt	test file in xport format
xptchar.sas	trivial SAS program used to 
	generate xptchar1.xpt
xptchar_SAS_System_Viewer9_1.csv	xptchar1.xpt 
	converted to comma separated file using SAS 
	System Viewer 9.1 (on Win XP)

With the patch applied, read.xport produces the same 
data frame from xptchar1.xpt as read.csv does from 
xptchar_SAS_System_Viewer9_1.csv (tested on i386 Linux 
with R Version 2.0.1) except that read.csv converts empty 
strings to NAs. As explained above, the empty string is
closer to the meaning of an all-blanks value in SAS.

There is renewed interest in this old data format in 
the pharmaceutical industry, because the US Food and 
Drug Administration requests clinical and 
pre-clinical data to be submitted in this format. I 
spent some time analyzing the xport file format to 
be sure of what is actually submitted to FDA with 
these files.

Thank you for considering this patch (and for the 
great R system, of course)!


Best regards,

Werner Engl



_____________________________________
Werner Engl, PhD, CStat            
Senior Manager, Biostatistics                             
Baxter AG, Vienna, Austria
e-mail: werner_engl at baxter.com
--- Please disregard any text below this line ---

-- 

GMX DSL-Netzanschluss + Tarif zum supergünstigen Komplett-Preis!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PR7389_we20041209.tar.gz
Type: application/x-gzip
Size: 1727 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20041209/0f297125/PR7389_we20041209.tar.gz