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   Mon, 31 Aug 2009 15:11:32 +0000 (GMT)

--- On Mon, 31/8/09, amatoallah ouchen wrote:
> I'm using a tobit model with friction, and the parameters
> have to be estimated by the maximum likelihood procedure,
> the number of iterations are exceeded, because the
> starting points i'm using are not good, is there is any
> way to choose consistent starting points?

I am a bit confused about what you mean with "a tobit model
with friction". Is this a special type of tobit model, or 
does that mean that you are haveing difficulty with 
estimating a regular tobit model (using -tobit-). 

In the former case: can you tell us exactly what commands
you used?

In the latter case: -tobit- is a reasonably well behaved
model, where the starting values provided by Stata should
work. If it doesn't converge than that indicates that 
there is something fundamentally wrong with your model,
e.g. (virtually) all your observations are censored, a 
(set of) variable(s) perfectly predcits the probability 
of being consored, etc. In this can easily remain 
unnoticed during univariate or bivariate inspection of
your data, as missing values on other variables in the
model will cause these observations to be dropped from
the analysis.

I would do the following. Say that your dependent 
variable is y and censoring happens at y=0, and your
independent variables are x1 and x2.

First create a dummy indicating whether an observation
is used in the analysis:

gen byte touse = !missing(y, x1, x2)

Check the number of censored and non-censored obs:
gen byte censored = y == 0 if touse == 1
tab consored if touse == 1

Check if there is perfect prediction of probability
of being censored:
probit censored x1 x2 if touse == 1

Do other checks to see if your model can be estimated
with your data.

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/



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