Re: st: duration analysis in gllamm

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Mon, 12/7/10, Melaku Fekadu wrote:
> I want to estimate a duration model (time-to-first-employment)
> through gllamm with unobserved heterogeneity. 

Steve has given you some additional references, I am just going
to give you a warning on this question:

>    How should look like a code in gllamm
> unobserved heterogeneity (parametric and non-parametric)? 

non-parametric models for unobserved heterogeneity are just 
divide the population up into groups. The argument is that
such a categorical distribution is an approximation of the
continuous unobserved variable. I have recently done some
simulations where I found that especially the two class
case tends to exhibit behaviour that is unique to a grouped
distribution: over time the negative correlation between 
the observed and unobserved variables first increased and
than decreased again. The latter decrease was due to the 
fact that at later points in time "frail" class became almost
empty. I am not that comfortable with an approximation of
a continuous distribution that on important characteristics
shows behaviour that is unique to a categorical distribuiton.
Especially since the information content in the data on this
unobserved variable is very small (obviously), so any goodness
of fit statistic used to select the number of classes might
easily let you choose a model with too few classes that would
show the behaviour I just mentioned.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   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/