Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: How to choose consistent starting values


From   Maarten buis <maartenbuis@yahoo.co.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: How to choose consistent starting values
Date   Tue, 1 Sep 2009 14:48:38 +0000 (GMT)

--- On Tue, 1/9/09, amatoallah ouchen wrote:
> the  tobit model with friction is based on the limited
> dependant variable (LDV) model of Tobin (1958) and Rosett
> (1959). here is the model
>
> y*t = xtb + et
>
>
> yt = y*t - a1  if    y*t< a1
> yt=0           if    a1<y*t< a2
> yt = y*t- a2   if    y*t> a2
>
> y*t   is the latent variable
>
> with a1< 0  and   a2>0
> and e i.i.d, residuals of the estimation with
> variance  sj^2. The parameters b, a1 a2 and s are
> solved by maximising a log-likelihood function.

I have two thoughts:

1) If the a's are known than you should be able to
use -intreg- to fit this model. I am a bit worried
about estimating the a's. First, it seems to me
highly improbable that the data contains enough
information to reliably estimate these parameters
(other than a1 is largest observed y less than 0
and a2 is the smallest observed y more than 0).
Second, in many cases we already have a good idea
about what these values may be, either from the
questionair design or by looking at the observed
y-s, so why not use this information. So I would
just fix these at reasonable values, and use
-intreg-.

2) When I am creating a program using -ml- and it
doesn't converge the problem has always been an
error my likelihood function (a bracket in the
wrong place, a lost minus sign, etc.), and never
in the starting values. I realize that not
everybody is as sloppy as I am, but it is an
indication about what may be going on.

Hope this helps,
Maarten

Ps. The Statalist FAQ explicitly askes everybody
to give full references and not just author-year
references. The reason is that this list is read
by people from many different disciplines, and
most references that are world famous in one's 
own sub-sub-discipline (and can thus be dealt
with with a name-year reference) are completely
unknown to the rest of the world.

--------------------------
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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index