On 10/24/2014 7:01 AM, F. Thomas wrote:
On 24/10/2014 11:28, John Darrington wrote:
...
SPSS does not permit you to examine their code, so nobody can know whether of
not
there are inaccuracies in it. I would be suspicious of any research
publication which
quotes statistical results produced by software which cannot be audited.
Which errors did you find ?
I work with both, SPSS and PSPP on different PCs. I have worked with
SPSS since 1977, then on punch cards.
Never any of my reports or publications were rejected because of
errors in SPSS procedures.
An industrial product such as SPSS that is used by research,
industry, and administration since several decades and in particular
by numerous researchers in social science and survey _methods _, such
as the National survey data archives, like ROPER, ICPSR, NDA, GESIS,
ZUMA, a program packet which is one of the standard research tools in
survey research, is fraught with error ?
And all those specialists , for decades, do not see errors in their
calculation in SPSS ???
How to say - don't exaggerate.
You're not addressing John's point, which is philosophical. He says that
IF there were an error in SPSS, you would never be able to tell from an
audit of their code.
Furthermore, I question your faith. How would NDA (whatever that is) or
any of the other organizations or individuals know that there was an error?
But, what the hack, let's answer YOUR question: I HAVE FOUND ERRORS IN
BOTH SPSS and SAS on a few occasions. Just recently I used SPSS 21 to
calculate means and SD's (yes, means and SD's!). First I used the
descriptives subcommand of CORR (or RELIABILITIES, I cannot recall) and
then later I re-calculated them using descriptives and they differed in
the second or third decimal. This happens all the time when there is
missing data because the samples differ and some routines (RELIABILITY,
for example) force listwise deletion, but THERE WAS NO MISSING DATA in
this dataset. So, that's one, and I was in EXACTLY the situation that
John described because I was doing this work for hire. Luckily the
client did not care about small differences.
Recently I was in a research meeting teaching some colleagues how to run
a complex bit of SPSS syntax. We had three different versions of SPSS
running on the same syntax using the same data on four workstations.
Three of us were able to reach agreement. The fourth colleague (whose
results could not be made to reconcile) was running SPSS on her Windows
8 laptop and she's now in a multivariate class and her version of SPSS
doesn't reproduce the instructor's logistic regression output. Looks to
me like her version of SPSS has bugs...
Back around 1990 when PROC CALIS was new and I was taking an SEM class,
I found a bug in Hartmann's work. I cannot recall now what it was but it
must have been glaring because I wasn't comparing software, I was just
reading the output and it popped out.
There was a fairly recent version of SPSS that had a bug that prevented
output from appearing. I think it was only on Windows 7 but how the hell
does a version of the software slip out the door with THAT bug? That's
like shipping a car from an auto factory without an engine. In fact, as
far as I know, SPSS, Inc. has had to release patches for every version
of SPSS to fix bugs in their software -- sometimes more than one patch.
So, those patches are absolute proof that their software has bugs of
some kind.
So, let's review: John's point is PHILOSOPHICAL, and he's right. Your
faith is, frankly, quite childish. I have personally found bugs and
furthermore it's also silly to argue that any large software package
does not have bugs. If YOU haven't found errors, it's because you
haven't looked hard enough.
So, to use your own words: don't exaggerate the safety of closed software.
-Alan
--
Alan D. Mead, Ph.D.
President, Talent Algorithms Inc.
science + technology = better workers
+815.588.3846 (Office)
+267.334.4143 (Mobile)
http://www.alanmead.org
Announcing the Journal of Computerized Adaptive Testing (JCAT), a
peer-reviewed electronic journal designed to advance the science and
practice of computerized adaptive testing: http://www.iacat.org/jcat
_______________________________________________
Pspp-users mailing list
Pspp-users@gnu.org
https://lists.gnu.org/mailman/listinfo/pspp-users
