Laura,
I have a panel data on firms and my dependent variable is a discrete choice variable with four
alternatives. I have used multinominal logit model to analysis factors affecting firms’ choice
behaviour. However, I would like to profit more from cross-sectional time-series data and thought
that this is possible using a random effects model.
I am just starting to learn Gllamm and would like to ask if anyone knows about estimating mlogit
in the framework of Gllamm. I have being reading the Gllamm manual but I am not sure how should I
do it. My explanatory variables do not vary across alternatives and the alternative set is the
same for each firm. In order to include random effects do I need to have random effects varying
between alternatives?
Yes, any random effect that is constant across alternatives will cancel out
in the multinomial logit probabilities. One set-up is to have alternative-specific
random intercepts for all but one alternative and allow them to be correlated.
Examples of this are given in the manual in Sections 9.3 (continuous random effects) and
9.4 (discrete random effects).
If this is the case, could I use the expanded() option although the
explanatory variables are not varying across alternatives?
Yes
I would also like to ask whether I have understood the meaning of levels right. If the panel
includes information on the firm, industry in which the firm is operating and year(s) when the
firm exists in the panel, is it possible to say that these are different levels and to have
random effects varying between firms, industries and years?
gllamm can only handle nested clustering variables or "factors" such as
firms nested in industries. Years appear to be crossed with firms (within industries)
so you cannot include random effects for year as well as firm and industries.
Finally, could anyone recommend a good reference on mlogit within Gllamm?
In addition to the manual, we have a paper:
Skrondal, A. and Rabe-Hesketh, S. (2003). Multilevel logistic regression for polytomous data and
rankings. Psychometrika 68 (2), 267-287.
and a less technical version from last year's proceedings of the Joint
Statistical Meetings which is downloadable from
http://www.gllamm.org/pub.html
The data and do-file for these models are available at:
http://www.gllamm.org/examples.html#rankings
Best regards,
Sophia
I would really appreciate any advice.
Thank’s in advance.
Laura
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
