Hi,
I'm looking for a way to bind specific model parameters to a constant
(in my case: 0) before fitting the model. I'm fitting multinomial
logistic regression models (via glm()).
Assuming a model with a categorial response variable c {c1,c2,c3}
that is a matrix like
c2: b2 b2.1 b2.2 b2.3
c3: b3 b3.1 b3.2 b3.3,
I would like to force, e.g. b2.1, b3.1 and b2.3 to 0 before
estimating the other parameters.
I'm aware of a similar - unresolved - question on list list back in
July:
https://stat.ethz.ch/pipermail/r-help/2005-July/075042.html
The "offset" in an interaction term doesn't seem to do what I want
-
as far as I can see, it manipulates the intercept, and it's additive
(i.e. an intercept is estimated anyways).
Thanks
David
--
David Reitter - ICCS/HCRC, Informatics, University of Edinburgh