Re: st: can Stata be used for this problem

[Hey Sky <heyskywalker@yahoo.com>]

Dear Kolenikov

thanks for your response and thanks for your advise last time you gave to me.

the model is about choice among edu, wrkwt (work with training), wrknt (work without training)
and home(stay at home) by using panel data. edu/wrknt/wrkwt/home are dummy variables.
 
Vijt = X*beta + alpa_j + gama_j*eta + sigma_t                (1)
Vijt is the utility get from choice on edu/wrknt, etc. and yes. it is multilogit model for panel data.

the hard part for me is the latent variable. it has one factor stucture, that is, mu_j=alfa_j +gama_j*eta. 
there is different parameters for different choice, alfa and gama (Ham and Lalonde 1996). 
eta has two discrete value which represents there have two types people. I assume the 
one eta =0 and gama_edu= 1 for identification. 

that is, for people who choose edu, the structure of the latent variable is 
mu_edu = alfa_edu + gama_edu*mu

for wrknt: mu_wrknt = alfa_wrknt + gama_ wrknt * mu. 
and the same for the other choices.

 
if I set up the following code for gllamm:  
 
gen cons =1
eq cons: cons
eq mu1: edu
eq mu2: wrknt
eq mu3: wrkwt

gllamm choice indep, i(id) ip(f) eq(cons sita1 sita2 sita3) noconstant nrf(4) family(binom) ///
   link(mlogit) base(4) trace dots

 
base on my understanding(it maybe wrong, plz remind me if so), the model will be: 
Vjt = X*beta + alpa*1 + gama_edu*eta  + gama_wrknt*eta +  gama_wrkwt*eta + sigma_t        (2)
 
the model would be different with my original model, since the intercept, alfa, in eq(1) is different 
with choice but (2) are all the same for any choice. 
 
and the model does not converge. any suggestions? thanks for them in advance.
 
 
Nan
from Montreal

 

 


----- Original Message ----
From: Stas Kolenikov <skolenik@gmail.com>
To: statalist@hsphsun2.harvard.edu
Sent: Mon, June 7, 2010 6:47:49 PM
Subject: Re: st: can Stata be used for this problem

On Mon, Jun 7, 2010 at 12:20 PM, Hey Sky <heyskywalker@yahoo.com> wrote:
> Vijt = X*beta + alpa_j + gama_j*eta + sigma_t
>
> alpa, beta, gama are parameters that need to be estimated.

I don't see any problems fitting this with -gllamm-. You treat choices
as fixed, then alpha_j is a part of the fixed effects, and gamma_j is
the choice-specific coefficient (loading). But, in all likelihood, you
don't observe utilities, so in the end it will boil down to some sort
of -mlogit- link. You need to describe the model more fully.

-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
*  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/