[Limdep Nlogit List] The Latent Class Ordered Choice model estimation

[William Greene]

Simona. Before each one, every time, fit the same model without the
LCM specification.  You are probably not getting the right starting
values. Also, do not use ALG=BHHH. Let it use BFGS.
/Bill Greene

----- Original Message -----
From: "Simona Rasciute" <S.Rasciute at lboro.ac.uk>
To: limdep at limdep.itls.usyd.edu.au
Sent: Friday, May 15, 2009 5:38:02 AM GMT -05:00 Columbia
Subject: [Limdep Nlogit List] The Latent Class Ordered Choice model estimation

Dear All, 

I am trying to estimate the Latent Class Ordered choice model and I would appreciate your comments on the following issue. 

When the model is estimated the following error is reported:
Initial iterations cannot improve function.Status=3
  Error   805: Initial iterations cannot improve function.Status=3
Function=  .10932395206D+06, at entry,  .16597492911D+06 at exit

The dependent variable has 10 categories. I am aware that the latent class estimator is unable to resolve more than one class with nice threshold parameters. Therefore, I have experimented with combining the categories. The simplest model that has been estimates is with 3 categories and 2 latent classes. The strangest thing is that initially none of the specifications worked, even if I included explanatory variables one by one. I have also tried different algorithms and number of iterations. This morning, however, the first estimation worked but afterwards identical estimation reported the same error above. 

Here is the syntax used:
ordered ;lhs = depvar
        ;rhs = List of independent variables
        ;LCM = with or without class membership variables
        ;pts=2
        ;alg=bhhh$ also other algorithms have been used

The data have also been carefully checked.

Thank you very much in advance.

Regards, 
Simona



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