Re: st: Sequential Probit

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

--- On Fri, 4/3/11, elin.vimefall@oru.se wrote:
> I still have one question. I want to take into accont that
> there is some unobserved heterogenity term for the
> individual, which is constant across the different schooling
> decisions. 
> Can I do that in this setting?

That is a tough question. First of all there are a number of
subtly different outcome measures one could be theoretically
interested in when it comes to these types of models, and 
some of these are just un-estimatable with this type of data,
for example:

<http://www.maartenbuis.nl/wp/holm.html>

I have used two ways of getting a handle on this problem. 

The simplest is to define my outcomes of interest as the 
proportion of successes at each transition while controling 
for the variables in my model. That way the problem is 
defined away. This sounds like a cheap trick, but there is 
more to it than that. I tend to prefer to think in terms of 
probabilities and odds rather than in terms of latent variables,
and a probability or odds is not defined independent of what
variables you choose to control for. Probabilties and odds
include a degree of uncertainty, and uncertainty is just a
set of variables we chose to define as non-systematic or
"(bad) luck". So when we want to think in terms of 
probabilities or odds we need to divide the set of all 
possible explanatory variables into two parts, at least some
of these variables needs to be assigned to the "luck" 
category otherwise the probabilties would be trivially either
0 or 1. This is a conceptual point that has to do with the 
way the dependent variable is defined and it is different
from the fact that it is impossible to control for 
everything. So what I did using this strategy is to assign
all variables not in my model to the "luck" category.

The second strategy I have used for this is to create a set
of scenarios on what that unobserved heterogeneity might look
like, and estimate a set of models, each assuming that one of
these scenarios is true. That way you can get a feel for what
this unobserved heterogeneity might do. I discussed that 
strategy in:
<http://www.maartenbuis.nl/publications/uh.html>

Other strategies exist, and some of these will be discussed 
in a forthcomming special issue of "Research in Social 
Stratification and Mobility". But keep in mind that there can
be no easy solution to this problem: there is an inherrit 
contradiction in trying to do empirical research on something 
we have not seen. This is why I am actually more and more 
inclined to like my first solution, as it bypasses the whole
problem.

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/