[Limdep Nlogit List] Random effects tobit

Thornburg, Steven sthornburg at wsu.edu
Tue Nov 17 11:23:29 EST 2009


I can sympathize.
If the covariance matrix is singular your model may be over
parameterized.  Try reducing it and adding in variables at each step.
Critically evaluate the variables that cause the problem for possible
multicollinearity

Per Limdep version 9 manual page R18-2

Error 805 - Initial iterations cannot improve function
Exit status will be 3. This may be due to unusable starting values or
the starting values are already in the solution

Error 806 - Line search does not improve fn.  Exit iterations
Exit status will be 3. This is usually not an error. If the likelihood
function is fairly flat near the maximum, this will occur.  However, for
a very badly behaved log likelihood, this error will occur and will give
you warning about the function being optimized. If this happens in the
first iteration, the model is probably inappropriate  for this data set.

Steve Thornburg
sthornburg at wsu.edu

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Thomas
Broberg
Sent: Monday, November 16, 2009 1:23 PM
To: limdep at limdep.itls.usyd.edu.au
Subject: [Limdep Nlogit List] Random effects tobit

Hi!

I have tried to estimate a RE-tobit in Limdep 9.0, so far with limited
success. Most times the variance matrix turns out to be singular or I
get error 805 or 806, which is not described in the manuals list of
error-codes. 

I have estimated the same models in Stata (xttobit) and got no error- or
warning-messages and the results differed from the Limdep-results (even
for models which Limdep accepted without warnings) . I have tried
different algorithms in Stata, but that did not change the results
substantially. 

Are the someone that can give me some feedback on this issue? 

Models estimated: Dependent variable with lower limit = 1. Independent
variables including both dummies and continuous variables. Unbalanced
panel (1-4 periods).


Thanks in advance!

Thomas Broberg
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au


More information about the Limdep mailing list