Re: Statistica

Robin J Hyndman (hyndman@athena.unimelb.edu.au)
Tue, 30 Aug 94 21:11:18 EDT


From: reece@eos.xx.rmit.EDU.AU (John Reece)

> Please, some facts, some facts. It's all very well saying that this issue
> was done to death in 1991, but many subscribers, including myself, have
> signed on since then and have neither the time nor the energy to search the
> archives for the 1991 postings.

> So, _please_, can someone summarise this issue for me, privately if
> necessary. I have the official line from the Australian Statistica reps
> and I am more than prepared to confront them with any claims given to me.

The following potted history from an Australian perspective may be of
some use.

Statsoft advertising tends to involve detailed comparisons of their
product (Statistica) with other packages. In 1991, Statsoft released
advertising material comparing Statistica with Systat. Systat
responded in October 1991 with a lengthy document claiming:
1. that the comparisions were inaccurate and misrepresentative
2. that some of the code in Statistica had been copied directly
from Systat
3. that there were serious statistical errors in Statistica
Statsoft responded quickly with an even longer document full of counter
allegations. Then, in late November 1991, Systat produced a further brief
document.

At the time, both packages were distributed by the same company in
Australia (Hearne Software). They were placed in a dilemma since both
Systat and Statsoft asked them to ditch the other. Therefore Hearne
Software approached me as an independent statistical consultant to look
at the problem.

I didn't look at the issues of plagiarism or false advertising, but I
did look at statistical and numerical accuracy in both packages. I
found that Statistica had some very serious statistical problems. For
example, it produced an ANOVA table to two-way factorial experiment
with one of the four combinations missing. The problem was Statistica
happily estimated interaction when the interaction is not estimable in
this case. Another example was that Kolmogorov-Smirnov tests used the
wrong significance level because they assumed the distribution
estimated from the data was exact. In other cases, Statistica used
misleading terminology and misapplied statistical concepts. This
indicated to me that the developers at Statsoft had a rudimentary
understanding of statistics at best and the package was very
unreliable. There were also some more minor problems with numerical
accuracy.

I found the allegations against Systat largely unfounded apart from one or
two small bugs.

As a result, Hearne Software stopped distributing Statistica. (They
have since started distributing it again.)

Statistica threatened me and my employer (Uni of Melbourne) with a law
suit. They also made deliberately misleading claims about me and my
association with Systat. (I have never had anything to do with Systat
and have never used the package except when carrying out these tests.)
I ignored the threats and the allegations and haven't heard from
Statsoft since.

I believe that all of the problems documented at the time, both by me
and by others, have been corrected in later versions of the software,
but I have never checked this out.

It now appears that Statsoft have started a new round of comparisons
with other packages including Data Desk. The developer of Data Desk is
Paul Velleman and he is upset at some of the things stated, claiming
unfair comparisons, misleading claims and outright lies. I have no
experience of Data Desk and have not investigated the allegations this
time round. But given my experience in 1991, I am inclined to be
sympathetic to anyone who contends with Statsoft.

In 1991 I concluded that Statsoft had limited statistical knowlege and
was a company happy to slander others for their own benefit. I doubt
that anything has changed.

Rob Hyndman
(Uni of Melbourne, Australia)