🚀 go-pugleaf

Allen Herzog writes:
...
> is good at these but far from user-friendly; GLIM and Genstat provide
> great flexibility but require considerable statistical expertise.

This is part of folklaw. So many people say "Minitab is easy to start
using, Glim is difficult". Since the commands to read in a simple data
matrix, plot a scatterplot and fit a regression line are comparable in
number, amount of typing and complexity of syntax in both, I strongly
suggest that any differences are subjective and expressed by people who
have generally used only one. I've used both, many other people on the
list have used both, but worldwide reputations are built up on the opinions
of the wider population of users. The difference between Minitab and GLIM
is that Glim then doesn't restrict you to a few simple models with enforced
Normal error assumptions. Incidentally, at York University they have taught
(maybe still do) Genstat as a first computer language to economics under-
graduates. I've also recently promoted Genstat to a user who just wants to
multiply matrices together.

>            D. Collet (Dept. of Applied Statistics, Univ. of Reading, UK)
> has just published a book "Modelling Binary Data". London: Chapman and
> Hall, 1991 ISBN 0-412-38790-5. This book is solely devoted to modelling

Dave Collett used to be at Hull. I mention this only to pre-empt the
gutter press from dragging up our relationship at some irrelevant point
in the future. (That's a UK in-joke of the week)

> three different relatively simple problems.  Since the author is a
> statistician (and works in the UK) GLIM wins hands down, followed
> closely by Genstat.  He finds the 2 of the SAS PROCS (CATMOD and LOGISTIC)

For reasons above, I think the "and works in the UK" is an unfair slur
on the software. Without seeing the book, I suggest that the Glim
approach and notation IS a useful and usable approach to analysing
binary data. In particular, the specification of models in both Glim and
Genstat, in terms of an error distribution and a set of predictive
variates and factors - with notation for interactions - seems to me
far easier to explain that the Byzantine ramblings of the SPSS Manova
command ... and I say that as an SPSS user of 20 years sitting.

>       It would be helpful if more rigor were employed in expressing
> opinions on the matter of evaluation.

Flame off. Thanks for the stimulus Allen.

Allan (that's with A a)