Re: st: Sequential Probit

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

--- On Thu, 17/3/11, S.Jenkins@lse.ac.uk wrote:
> I disagree with Maarten. I think a form of multivariate
> probit model with selection can account for the sequential
> aspect he refers to.  Suppose there are 3 education
> levels: A, B, C
> 
> The model I have in mind is:
> Pr(reach A) as function of stuff, where risk set is all
> starting education
> Pr(reach B) as function of stuff, where risk set is those
> finishing A
> Pr(reach C) as function of stuff, where risk set is those
> finishing B
> And the errors of the latent outcomes are trivariate
> normal.

The difference is I think more a matter of terminology than 
of substance: I interpreted multivariate probit to mean a
model where the dependent variable is the highest level of
education. The interpretation of the marginal effects do 
then no longer have the sequential interpretation. Stephen's 
use of terminology actually makes more sense / is more 
consistent with use in other Stata commands in particular 
multinomial probit (mprobit) versus bivariate probit 
(biprobit).

> The standard bivariate probit model with selection
> (-heckprob- in Stata) could be applied to the 2 level case.
> 
> A paper that applies this approach is "Selection Bias in
> Educational Transition Models: Theory and Empirical
> Evidence" by Anders Holm and Mads Meier Jæger, unpublished
> paper, Center for research in compulsory schooling,
> University of Aarhus, Tuborgvej 164, DK-2400 Copenhagen
> NV  Denmark (email: aholm@dpu.dk or
> email: mmj@dpu.dk) 

This paper is about to appear in Research in Social 
Stratification Mobility.

-- Maarten

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

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



 
> One is asking a lot from these models given the data
> available, so they can be hard to fit and might be rather
> fragile.
> 
> Maarten's own approach to examining the impact of
> unobserved heterogeneity in these models in his -seqlogit-
> (on SSC) assumes a different type of model (I think). That
> is (I think) he assumes the same univariate normal frailty
> distribution for the errors of the latent outcomes in each
> equation (i.e. perfect correlation across equations). [His
> likelihood function accounts for the selection arising from
> the sequential nature of the problem.]  However, rather
> than /estimating/ the variance of the normal frailty
> distribution, he allows the researcher to assume different
> values for that variance and to inspect what happens to the
> estimates of the other model parameters with different
> assumptions. So, note: compared to the MV probit with
> selection approach, there is a different underlying model
> (for the frailty and link functions).  His approach
> might be more robust -- I don't know -- but that is partly
> bought by his assumptions implicit in his
> specification.  
> 
> Stephen
> ------------------
> Professor Stephen P. Jenkins <s.jenkins@lse.ac.uk>
> Department of Social Policy and STICERD
> London School of Economics and Political Science
> Houghton Street, London WC2A 2AE, UK
> Tel: +44(0)20 7955 6527
> Survival Analysis Using Stata: http://www.iser.essex.ac.uk/survival-analysis
> Downloadable papers and software: http://ideas.repec.org/e/pje7.html
> 
> 
> 
> Please access the attached hyperlink for an important
> electronic communications disclaimer: http://lse.ac.uk/emailDisclaimer
> 
> *
> *   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/