This appears to be related to hurdle models (or two-part models) in
which one models the dichotomy and the continuous parts separately and
combines the test statistics. When you have a probability mass and
continuous part of a distribution, I'm not sure it makes sense to
consider this as a mixture since the probability of 0 on the continuous
part is 0.
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Kristian
Karlson
Sent: Wednesday, January 21, 2009 9:59 AM
To: [email protected]
Subject: st: Mixing mlogit and gaussian in -gllamm-
Dear all,
Is anyone familiar with how to mix a multinomial logit and a ordinary
linear model in -gllamm-? More specifically, I want to mix a multinomial
and a gaussian distribution family (with links mlogit and identitity
respectively).
My problem is both at the data preparation stage and at the
specification of -gllamm-. Usually I expand the data to model the
mlogit, but when I have to combine this with a linear model as well, I
can't figure out exactly how to prepare the data and do the consequent
modeling in -gllamm-.
Generally, I think that my problem is how to use lv() and fv() - which
identifies the outcomes of interest - in combination with an expanded
data set, i.e., in combination with expand() in -gllamm-.
If anyone has tried such a thing in -gllamm-, I would very much like to
see their do-file. (I have spent a long time trying with either reshape
or expand, but I simply cannot get the basic clue as how to set up the
do-file).
All the best,
Kristian
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
