RE: st: nlcom, "Maximum number of iterations exceeded."

[Mauro.PISU@oecd.org]

Dear Maarten,

Thanks again for your reply, but I do not think the problem I have with nlcom has anything to do with the missing standard errors. As the examples below show, slightly modifying the nonlinear expression of the marginal effect (changing the exponent of _b[/Umi]), and using exactly the same parameters as before, produces sensible results. 

. nlcom ME0 : _b[/Uci]^2 + _b[/Uxi]^2 + _b[/Umi]^2 + _b[/Uii]^2 ,

         ME0:  _b[/Uci]^2 + _b[/Uxi]^2 + _b[/Umi]^2 + _b[/Uii]^2
Maximum number of iterations exceeded.
r(498);


If I slightly modify the marginal effects changing the exponent of _b[/Umi] I get the following


. nlcom ME1 : _b[/Uci]^2 + _b[/Uxi]^2 + _b[/Umi] + _b[/Uii]^2 ,

         ME1:  _b[/Uci]^2 + _b[/Uxi]^2 + _b[/Umi] + _b[/Uii]^2

------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         ME1 |   .4094367   .0654285     6.26   0.000     .2811993    .5376741
------------------------------------------------------------------------------

. nlcom ME3 : _b[/Uci]^2 + _b[/Uxi]^2 + _b[/Umi]^4 + _b[/Uii]^2 ,

         ME3:  _b[/Uci]^2 + _b[/Uxi]^2 + _b[/Umi]^4 + _b[/Uii]^2

------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         ME3 |   .4098511   .0654577     6.26   0.000     .2815562    .5381459
------------------------------------------------------------------------------


So the problem must be in nlcom  and not in the estimates produced by nlsur (which I agree need to be fixed anyway). 

Cheers,

Mauro


*****

I have understood your situation to be the following:
You estimate a model with -nlsur-, "but for some standard errors 
that are missing for unknown reason." After that, you ignore this
problem and still try to use -nlcom-, and -nlcom- won't converge.

-nlcom- is a postestimation command, it will use whatever model
you estimated before it as its input, in your case -nlsur-. If
there is a problem in your previous command then there will
also be a problem in your -nlcom- command. Missing standard 
errors is a clear sign that there is something seriously wrong
with your -nlsur- command. So my comment was that you should 
first fix your -nlsur- command before you start using -nlcom-,
and the problem with your -nlsur- command is probably the reason
that -nlcom- won't converge.

The comment about perfect colinearity was just one possible
thing that could cause your problems with -nlsur-.

-- Maarten

******


*
*   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/