Re: SPSS file record layout type 2 vs type 3

[Ben Pfaff]

Ben Pfaff <address@hidden> writes:

> John Darrington <address@hidden> writes:
>
>> On Mon, Mar 03, 2008 at 09:50:29AM -0800, Ben Pfaff wrote:
>>
>>      I hope that you have found the PSPP documentation on the system
>>      file format.  This is drawn from the wotsit documentation plus
>>      pretty extensive experience with actual SPSS files around the
>>      web.  The source for the latest version of the PSPP documentation
>>      can be found at:
>>              
>> http://cvs.savannah.gnu.org/viewvc/pspp/doc/dev/system-file-format.texi?root=pspp&view=markup
>>      
>>
>> Now that I look at this, I notice that PSPP always writes the value 2.
>> The spss documentation suggests that this implies that the file can be
>> read by versions prior spss v7.5; I've never tested any of our files
>> with such an old version. So shouldn't we be writing 3 instead ?
>
> I have access to a very old version of SPSS (older than 7.5).  If
> you like, I'll try out some .sav files with it to see whether the
> "version number" here makes any difference.

I tried a few experiments this evening.  First, I went through my
whole library of .sav files (about 200 unique files, from all
over the web) to ferret out the ones that had layout code 3.  To
my surprise, there was only one: the 1991 U.S. General Social
Survey .sav file that comes with SPSS.

I tried loading this into the very old SPSS version.  It refused,
saying that it came from a newer version of SPSS that had an
incompatible .sav file format.  At this point it looked like you
were right and that we should use 3 instead of 2.

But then I used a hex editor to change that byte from 3 to 2.
And the old version of SPSS then read it in with no complaints at
all.  The dictionary was fine and so was the data.

So I suspect that layout code 3 was an aberration introduced into
some development version of SPSS that never really escaped
internal product development.  Otherwise such a layout code would
not just exist in files supplied by SPSS itself.

Actually it appears that we discussed this back in 2006:
        http://www.mail-archive.com/address@hidden/msg01285.html
I don't have a copy of survey_sample.sav mentioned in that
message.  If anyone can send me a copy or point me to one, I'd be
much obliged.
-- 
"GNU does not eliminate all the world's problems,
 only some of them."
--Richard Stallman