Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: GLAMM and nested logit/probit


From   "Kristian Karlson" <kristian.karlson@gmail.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   Re: st: GLAMM and nested logit/probit
Date   Tue, 18 Nov 2008 22:05:44 +0100

Dear Maarten,

Thank you ever so much for your answers. I am well acquainted with the Holm
& Jaeger paper, the selection problem in educational transition models
(sequential logit), and will look into the simulated likelihood routines. In
fact, a previous draft of the Holm & Jaeger paper is my main inspiration,
but the authors are looking at only two transitions. The nlogit suggestion
is also brilliant (also for eventual starting values in later models).

However, I believe I haven't fully explained my project (here goes). The
main point of the study is to delve into what the unobserved components in
educational decisions consist of, i.e. describing the latent classes that
account for unobserved heterogeneity with variables that indicate different
forms of motivation. This strategy might seem a bit confusing or
unconventional, because I could control for these variables in the main
equation. However, the idea was to include concomitants for the latent class
memberships and, in doing so, try to delve into the "contents" of the
unobserved heterogeneity. This is possible in -gllamm-.

Nonetheless, I am still very interested if you or anyone else could help me
with the syntax on doing a more complex transition models in -gllamm-, i.e.
a model as the one I have stated in the main thread (with a nested
structure). The main point is that the more complex a model (i.e. the more
realistic model of the educational system), the better the identification of
the unobserved heterogeneity influencing educational decisions net of
respondent background characteristics (family background, gender, ability,
etc.).

All the best,
Kristian


-----Oprindelig meddelelse-----
Fra: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] På vegne af Maarten buis
Sendt: 18. november 2008 18:58
Til: statalist@hsphsun2.harvard.edu
Emne: Re: st: GLAMM and nested logit/probit

I have three comments that might or might not be useful:

1) I would start with -nlogit- (see: -help nlogit-) if only to get a
baseline model.

2) This looks very much like a model discussed in a working paper by
Anders Holm and Mads Jaeger (2008). This model can be generalized to
more complex transition models using -gllamm-, but using maximum
simulated likelihood routines may be easier and/or quicker. For those
see: http://www.stata-journal.com/sj6-2.html

3) One thing that may help is to do a sensitivity analysis, you are
after all trying to control for stuff you haven't observed. I have
created some tools for doing a sensitivity analysis for this type of
model (and for this type of application). These are discussed in Buis
(2008). However, I still need to update the help-files, so the
procedure discussed in that paper is right now only available on my
computer... The hard stuff is done though, so it should not take long
to get that finished. (For those who are worried about my dissertation
progress: this is a chapter in my dissertation)

Hope this helps,
Maarten

Buis, Maarten L. (2008) The consequences of unobserved heterogeneity in
a sequential logit model. http://home.fsw.vu.nl/m.buis/

Holm, Anders and Mads M. Jaeger (2008), Selection bias in educational
transition models: Theory and empirical evidence. Social Policy and
Welfare Services Working Paper 11:2008. http://www.sfi.dk/sw60623.asp

--- Kristian Karlson <kristian.karlson@gmail.com> wrote:
> Can GLLAMM perform nested multinomial logit models? I would like to
> estimate a nested logit (or probit) model with unobserved
> heterogeneity captured by a number of latent classes. My project is
> about educational decisions in Denmark.
> 
> My nested choice structure looks like this (a simple model of the
> Danish educational system):
> 
> 1A. No further education
> 1B. Vocational
> 1C. High School
>    2C1. No further education
>    2C2. Short education
>    2C3. Medium education
>    2C4. Long education
> 
> The first choice is among 1A-1C. Contingent on 1C the next choice is
> between 2C1-2C4. This model can be understood as a form of duration
> model in the guise of a continuation ratio model with multiple exits.
> I would like to include timevarying covariates, i.e. covariates that
> change over the nested choices. I had an idea about the model being a
> form of multi-process model. First process is 1A-1C, second process ?
> contingent on 1C ? is 2C1-2C4. That is, modelling 2C1-2C4 is only
> possible when the selection process into 1C has been modelled.
> Does anyone have a suggestion as to how to model this with GLLAMM? My
> alternatives are non-Stata programs (such as LEM or aML), but I fancy
> GLLAMM and Stata.
> 
> If the information in this thread is too sparse, please let me know -
> I will then provide some more detailed information.
> 
> Thank you, 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/
> 



-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room N515

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------


      
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index