🚀 go-pugleaf

>I have been poring over Chapter 9 in the SAS/STAT Guide for Personal
>Computers (ver 6.0) to try to figure out the difference(s) between
>these four types of estimable functions, and I am beginning to
>doubt what I thought I once understood.
>
>Can somebody explain in "plain english" the differences between these
>four types of SS?
>

  Without a doubt I am biased, but one reference that I think explains
the differences in the different types of sums of squares and the
hypotheses that are tested is a book that I co-authored.  The book is:
        Analysis of Messy Data - Vol. I: Designed Experiments  by
George A. Milliken and Dallas E. Johnson.  It is published by Van
Nostrand Reinhold and can be obtained by calling 606-525-5183.  Last
Thursday, they had 140 copies left of the third printing.  It sells
for $58.95 I think.  The next printing will be in paper back and it
will sell for about $40, but will not be available until the fall.
We are planning a revision, but it will not be finished for 2-3 years.

  I think that this book can also help to answer a couple of other
questions that appeared recently on this list.  Particularly the
one about how contrasts are computed and books which are useful for
the analysis of clinical trials.  The book was written for applications
oriented statisticians and data analysts.  The topics in the book were
primarily those that never seemed to get talked about in traditional
statistics courses.  Some data is too "messy" to analyze, but if data comes
from designed experiments, I think that this book can help data analysts to
obtain correct analyses.  There are several chapters on split-plot experiments
and repeated measures experiments, although when the book was originally
written the REPEATED option in SAS GLM did not exist.  Those who are faced
with analyzing unbalanced split-plot and repeated measures experiments should
read Chapter 28 very carefully.  Those who do will discover that the standard
errors computed by the GLM-LSMEANS option are often wrong and the p-values
computed for pairwise comparisons are also often wrong.  If there are
specific questions about portions of the book, I would be happy to try to
answer them.  George and I have also been teaching public seminars
out of the book for the Institute of Professional Education since 1979.
Well that is enough! I told you at the beginning that I was
biased!  Dallas E. Johnson, Professor of Statistics & Statistical
Consultant, Kansas State University.