Re: st: xttobit and xtprobit missing standard errors

["Guillaume Frechette" <gurst1@hotmail.com>]

contribute to this list are more competent than I am) but I would say that 
in this type of estimation, there can be a lot of path dependence. For 
instance, for some optimization methods the hessian "accumulates" as a 
result of each step in the iteration. Thus, you can sometimes get different 
results as a function of where the process started.

Happy to see you resolved your problem.

Cheers,

g


> From: "Alejandro Lopez-Feldman" <alelopez@ucdavis.edu>
> Reply-To: statalist@hsphsun2.harvard.edu
> To: <statalist@hsphsun2.harvard.edu>
> Subject: Re: st: xttobit and xtprobit missing standard errors
> Date: Fri, 19 Nov 2004 11:00:34 -0800
>
> Guillaume,
>
> Thanks for the help. I used superxt and it worked out well.
> I just have a quick question. Convergence wasn't achieved with the default
> options for superxt so I tried type(2) but it didn't work so I changed rho
> from the default of .5 to the value that I was getting from the xttobit
> (rho=.01), that is convergence was achieved with type(2) rho(.01).  The
> weird thing (at least to me) is that after convergence the result of rho 
> was
> .57. My question is why I didn't achieve convergence using type(2) and the
> default rho that is close to the "true" value?
>
> Thanks,
> Alejandro
>
>
> ----- Original Message -----
> From: "Guillaume Frechette" <gurst1@hotmail.com>
> To: <statalist@hsphsun2.harvard.edu>
> Sent: Friday, November 19, 2004 6:28 AM
> Subject: RE: st: xttobit and xtprobit missing standard errors
>
>
> > You could try superxt (http://homepages.nyu.edu/~gf35/html/sg158.htm) or
> > just try different starting values or different number of quadrature
> points.
> >
> > g
> >
> > >From: "Alejandro Lopez-Feldman" <alopezfeldman@ucdavis.edu>
> > >Reply-To: statalist@hsphsun2.harvard.edu
> > >To: statalist@hsphsun2.harvard.edu
> > >Subject: st: xttobit and xtprobit missing standard errors
> > >Date: Wed, 17 Nov 2004 16:33:58 -0800 (PST)
> > >
> > >Hi Stata users,
> > >
> > >I am using xtprobit and xttobit to estimate a couple of models with
> random
> > >effects but there seems to be a problem with my results.  When I run 
> the
> > >xtprobit command everything seems to be fine (no warning messages,
> > >convergence achieved after 5 iterations and many significant
> coefficients)
> > >but no standard errors are estimated for lnsig2u and therefore none 
> s.e.
> > >for sigma and rho, although the likelihood ratio test for rho= appears
> > >(Prob >= chibar2 = 0.029) . I have a similar problem when I run 
> xttobit,
> I
> > >get s.e for sigma_e but not for sigma_u and therefore none for rho.
> > >
> > >It seems to me that stata is trying to tell me that there is something
> > >wrong with my results but I don't know how bad the problem is. Are my
> > >estimates valid at all? Are the s.e for the coefficients meaningfull? 
> Is
> > >there a way to get around this problem? I am interested in getting s.e
> for
> > >rho in both cases because in both cases I want to test the hypothesis
> that
> > >the panel estimator isn't different from the pooled estimator.
> > >
> > >Thanks,
> > >Alejandro
> > >*
> > >*   For searches and help try:
> > >*   http://www.stata.com/support/faqs/res/findit.html
> > >*   http://www.stata.com/support/statalist/faq
> > >*   http://www.ats.ucla.edu/stat/stata/
> >
> >
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/support/faqs/res/findit.html
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/