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

eric.gibson@pharma.Novartis.com
Wed, 29 Apr 1998 15:58:42 -0400


Hello Splus Users,

Everytime I hear this back and forth arguement about the Type I through
Type IV sums of squares
I start asking people whether or not they have ever read Milliken and
Johnson's book
"The Analysis of Messy Data Volume I: Designed Experiments" (1992).
Invariably the answer is no. Since they are both students of Franklin
Graybill ( argueably
the father of linear models) and they have an impressive record of
publication in the field of linear
models, I would highly reccomend it. Especially the discussion of Type I
through Type IV sums
of squares, which is a constant theme throughout the text.

I don't even bother to enter into discussions on the topic of Type I
through Type IV sums of squares
anymore. I just refer the discussants to the Messy Data text, which
thoroughly lays the issue to rest.

I agree with Frank Harrell, that pharmaceutical companies have done a great
deal of damage on this topic
over the years. Largely because they live in a vacuum and focus too much
on the pharmaceutical literature
instead of the statistical literature. I tend to give the advice of
Milliken and Johnson far more weight than the
advice of someone like Stephen Senn (no offense meant).

Kind Regards,
Eric Gibson

owner-s-news@wubios.wustl.edu on 04/29/98 01:42:39 PM

Please respond to fharrell@virginia.edu

To: s-news@wubios.wustl.edu
cc:
Subject: Re: [S] "Sums of squares", the taxonomy issue again.

Alan Zaslavsky and Bert Gunter made some wonderful points. They made me
think of a few other things to add:

1. I believe that SAS Institute and pharmaceutical companies have driven
the use of improper F-tests much more than the FDA

2. As Bert hints at, I believe that we teach many incorrect approaches in
graduate schools. One of my pet peeves is the failure of most programs
to address overfitting and model validation, for example.

3. I've been a consultant to FDA for many years and have found that FDA
statisticians are flexible and current in their knowledge of methodology.
They know a good argument for using a different
statistical technique when they see one. Frequently, sponsors propose
very standard, vanilla methods in fear of what FDA statistical reviewers
may think. These fears are largely unjustified in my experience. At any
rate, sponsors can have their proposed methods pre-approved in most
cases.

4. Paul Gallo, Assistant Director of Biostatistics at Novartis
Pharmaceuticals has put together a really nice paper that addresses
center effects in multi-center randomized trials. He is presenting the
paper at the "Understanding, Applying and Not Misusing the Mathematical
and Statistical Techniques Used in Clinical Trials" conference in Reston,
Virginia tomorrow and Friday. Paul addresses some long-ignored issues
such as loss of power when using type III sums of squares. Paul's E-mail
address is paul.gallo@pharma.novartis.com.

---------------------------------------------------------------------------
Frank E Harrell Jr
Professor of Biostatistics and Statistics
Director, Division of Biostatistics and Epidemiology
Dept of Health Evaluation Sciences
University of Virginia School of Medicine
http://www.med.virginia.edu/medicine/clinical/hes/biostat.htm

-----------------------------------------------------------------------
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
RFC-822-Headers:
X-Priority: 3
X-Msmail-Priority: Normal
X-Mailer: Microsoft Outlook Express 4.72.2106.4
X-Mimeole: Produced By Microsoft MimeOLE V4.72.2106.4
Received: from localhost (majordom@localhost) by wubios.wustl.edu
(8.8.6/8.8.6) with SMTP id MAA15419

-----------------------------------------------------------------------
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