Re: st: MLE est. linear form with two sets of parameters

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

--- On Sun, 30/5/10, Denis Kalugin wrote:
> The goal is to estimate two latent variables, let's say L1
> and L2. The variables are linked to variable y in the following
> manner:
> y=0 if L1<z<L2
> y=-1 if z>L2
> y=1 if z<L1
> 
> L1=exp(x*b1+error)/(1+exp(x*b1+error)
> L2=min[exp(x*b1+error)/(1+exp(x*b1+error)
> , exp(x*b2+error)/(1+exp(x*b2+error)
> 
> This is estimated via slightly altered probit (as far as I
> understand).
> 
> P(y=-1)=norm[(log(z/(1-z)) - xb1)/sigma(error1)]
> P(y=1)=(1-norm[(log(z/(1-z)) -
> xb1)/sigma(error1)])*(1-norm[(log(z/(1-z)) -
> xb2)/sigma(error2)])
> P(y=0)=1-p(y=-1)-p(y=1)
> 
> However, my problem right now is how to feed TWO sets of
> parameters (b1 and b2) to be estimated. If i write this
> as a linear form, then stata would generate only one 
> vector. 

Looks like your are mixing up logits and probits, either use -normal- or
use exp(.)/(1+exp(.)). The latter can be coded more briefly as 
-invlogit()- . Also note that coding normal(-xb) or invlogit(-xb) is much
more stable than (1-normal(xb)) or (1-invlogit(xb)).

Anyhow, for your immediate question: 

Your claim about how -lf- works is not correct. You really need to get 
your hands on: William Gould, Jeffrey Pitblado, William Sribney (2006) 
"Maximum Likelihood Estimation with Stata, 3rd Edition", College 
Station, TX: Stata Press.
<http://www.stata.com/bookstore/mle.html>

Your model looks very similar to -zoib-, so you can type -ssc install
zoib- and look at that program. 

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/