[S] "Sums of squares", the taxonomy issue again.

Bill Venables (wvenable@attunga.stats.adelaide.edu.au)
Wed, 29 Apr 1998 10:39:26 +0930


Folks,

Over the last few weeks I have been having a private exchange
with a most urbane and interesting person who works for, let's
say, another major commercial statistical software system (but
not SAS) who will have to remain anonymous, about various linear
model issues. (I propose to speak about related matters at the
Washington conference, if anyone is at all interested.)

Nevertheless the next paragraph from my correspondent in this
morning's email came as a real surprise to me, even though it
merely confirmed what I thought could only have been extreme
rumors.

> I definitely agree on the sums of squares issue. The Type IV
> stuff really gets me. In some situations, the results depend
> on the codings used to define variables! But we had to
> implement them in our [lm+aov] procedure, because
> U.S. government agencies mandate use of them. (They can't
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> specify software packages, but they can specify methods, and
> until we did that, specifying Type IV as the method for
> certain problems basically meant specifying SAS, unless you
> were willing to figure out how to get them with other
> software. I could make our MANOVA procedure match a given set
> of Type IV sums of squares, but it meant jumping through many
> more hoops than most people wanted to deal with.)

(Indignant ^^^^ pointing mine, of course.)

So, it really is the case that in the US the government holds the
statistical profession in such complete contempt that it feels it
can prescribe by regulation how we should do our job.

Does it also prescribe what drugs and clinical procedures a
medico should use to treat certain conditions, the way architects
should design buildings, or the way artists should depict the
landscape (or the president)? In some cases I think the answer
is an equally sorry `yes'.

I must say I find this completely outrageous, especially since
the profession knows that the analysis sanctioned by regulation
is WRONG, as my sad but resigned correspondent above plainly
acknowledges.

It now seems to be the case that S-PLUS is well on the way to
going down the same tacky road, at least in this issue.

Well I think the profession should simply refuse to be a party to
such right-wing practices and protest, (a) that the government
should even seek to intrude so deeply into fundamental matters of
professional practice, especially delicate issues like these in
which it clearly has no competence and very weak advice, and (b)
that software vendors should toe the line so meekly.

Bill V.

-- 
Bill Venables, Head, Dept of Statistics,    Tel.: +61 8 8303 5418
University of Adelaide,                     Fax.: +61 8 8303 3696
South AUSTRALIA.     5005.   Email: Bill.Venables@adelaide.edu.au

----------------------------------------------------------------------- This message was distributed by s-news@wubios.wustl.edu. To unsubscribe send e-mail to s-news-request@wubios.wustl.edu with the BODY of the message: unsubscribe s-news