Many thanks Nick, you are right I managed to somehow check whether my models
are stable or not. Still, when extracting the marginal effects from my ivprobit
with values away from the mean in my main variable of interest (i.e., adopting
values towards the very left of the distribution of my explanatory variable) I
get probabilities beyond 1, reaching up to 2 at the very bottom. I wonder if
this sort of outcomes are possible and more importantly, not signaling that
something is going wrong. Am using the following command:
mfx compute, predict(p) at(mean main_depvarlist=...)
Any help would be very appreciated.
Alejandro
In message
<031173627889364697C50B3B266CBB8A01C07ED0@GEOGMAIL.geog.ad.dur.ac.uk>
[email protected] writes:
> I am not quite sure what kind of answer you want,
> as you seem to answer your own question.
>
> In essence, -ivprob- is an user-written command
> that will run in Stata 8.2 up, while -ivprobit-
> is part of official Stata introduced in Stata 9.
> (That makes what you say a little surprising, but
> I guess a typo somewhere.)
>
> The acknowledgements indicate that StataCorp looked
> carefully at -ivprob- when writing -ivprobit-, but
> the relationship is not much closer than that,
> I guess. It's not as close -- as does sometimes
> happen -- as when a user-written command
> is adopted directly as official. Even in that
> case the code and help file can get rewritten
> and it can be a case of "I have my grandfather's axe.
> My father replaced the blade and I replaced the handle".
> (Sorry for any implicit testosterone here. No doubt there
> are female equivalents.)
>
> More specifically, as estimation methods differ
> you have something of a check on whether models
> are stable.
>
> If you want more on -ivprob- than is given in
> the help, look inside the program or contact
> the author (who may not be a member of Statalist).
>
> Nick
> [email protected]
>
> Alejandro Delafuente
>
> > Am trying to estimate a probit model with endogenous
> > regressors, but can't
> > really understand the difference between the ivprob command
> > and the ivprobit
> > command.
> > I read that ivprob implements Amemiya Generalized Least
> > Squares (AGLS)
> > estimators whereas ivprob uses Maximum-Likelihood estimation
> > procedures. Is the
> > difference then that AGLS carries out a sort of two-step
> > procedure but
> > correcting standard errors that would otherwise
> > (asymptotically) be less
> > efficient than maximum likelihood estimates?
> > Or is ivprob the same as ivprobit, but under different Stata versions?
> > I could not run ivprob in Stata 8.2 nor ivprobit in Stata
> > 9.0. I tried both
> > commands under each respective versions and the results were
> > very similar, but not the same.
>
> *
> * 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/
>
--
Alejandro de la Fuente
Department of International Development/QEH
University of Oxford, Mansfield Road, Oxford OX1 3TB
Tel: 01865 281836
*
* 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/
