Re: st: mlogit, bootstrap, mfx: "no observations"

["Stas Kolenikov" <skolenik@gmail.com>]

Is there any reason that you don't like the standard errors supplied
by MLE (other than advisor's/reviewer's suggestion)? Technically the
full bootstrap for the categorical dependent variable should give you
infinite standard errors, as for some bootstrap samples, all the
responses will be identical, and thus the estimated parameters will go
to infinity. To apply the bootstrap to situations like that, certain
tweaks may be necessary, such as constraining the proportion of each
category not to go to zero.  You may be running into something similar
with -mfx-, although I don't use the command often enough to know
precisely what it does and why it breaks down.

Also, the bootstrap is good for smooth functions of data (or at least
smooth functionals of the distribution function), and categorical data
are obviously jumpy; I don't think the latter is much of a problem
here, but just to warn you that the bootstrap is not a device like a
magician's hat, it has its limits of applicability that one may hit
surprisinlgy often.

On 8/24/06, Guido Heineck <Guido.Heineck@gmx.net> wrote:
> Dear all,
>
> I have cross-sectional data and run a multinomial logit model for a dependent var with 4 categories (by gender). Following the model, I calculate marginal effects for a subset of covariates.
> I have to bootstrap the standard errors and here's where my problem comes in: Without the bootstrap option, everything runs fine. Bootstrapping the s.e. also works fine - without marginal effects. Everything together however generates the following error message:
>
> no observations
> r(2000)
>
> Here's the relevant part of my code:
>
> foreach c of local foo {;
> mlogit yvar `xvars' `c' if male==0, vce(bootstrap, reps(100) seed(123));
> mfx , predict(p outcome(1)) var(`c');
> mfx , predict(p outcome(2)) var(`c');
> mfx , predict(p outcome(3)) var(`c');
> mfx , predict(p outcome(4)) var(`c');
> };
>
> Any ideas on what I am missing here? Thanks for any helpful suggestions.
>
> Guido
>
> --
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Guido Heineck
> University of Erlangen-Nuremberg
> Department of Economics
> Lange Gasse 20
> 90403 Nuremberg
> Germany
> fon: +49-(0)911-5302-260
> fax: +49-(0)911-5302-178
> http://www.lsw.wiso.uni-erlangen.de
> http://www.8ung.at/heineck
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> ENER - European Network
> on the Economics of Religion
> http://www.ener-online.org
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
>
>
>
>
>
>
>
>
>
>
>
> Echte DSL-Flatrate dauerhaft f�r 0,- Euro*. Nur noch kurze Zeit!
> "Feel free" mit GMX DSL: http://www.gmx.net/de/go/dsl
> *
> *   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/

--
Stas Kolenikov
http://stas.kolenikov.name

*
*   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/