I've installed and have been using --margin-- and it appears to work well for
my purposes. Where ---mfx compute-- took approximately 25 hours of CPU time
to deliver a single result, --margin-- gives me about 24 results in less than
12 hours. Since my variable of interest is a dummy, I feel the average
marginal effect produced by --margin-- is a cleaner result than the marginal
effect at the mean produced by --mfx compute--.
David Clingingsmith <[email protected]> wrote
> I am running a set of regressions using mlogit with mfx to get
> marginal effects. Each regression has between 1.5 and 3.5 million
> observations and estimates 5 coefficients for three values of the
> independent variable.
>
> As I've learned, mfx with standard errors takes a very long time
> to run. So far its used 41 hours of CPU time on a Solaris machine (dual
> 800 Mhz processors) without giving me one marginal effect, and when I've
> asked it to run without computing the errors it takes 10 hours or so per
> maringal effect.
>
> My questions are:
>
> 1. Is there another way to get marginal effects with standard
> errors other
> than mfx?
I wrote a program called margin that computes the marginal effects and the
standard errors analytically. It should produce the effects much quicker.
It is available at the website http://www.bkae.hu/bartus/stata
You can also install it using the net from "http://www.bkae.hu/bartus/stata";;
command.
> 2. Am I much better off using a Windows box with a 2.4Ghz
> processor? I
> would need to add memory to it to get it to hold the dataset.
I do not know the answer but mfx also runs slowly on Windows platforms,
provided
the dataset is huge enough.
Best wishes,
Tamas
------------------------------------------------
Tamas Bartus, PhD
Assistant Professor, Department of Sociology and Social Policy
Budapest University of Economic Sciences and Public Administration
1093 Budapest, Fovam ter 8.
Phone: +36-1-217-51-72 Fax: +36-1-217-44-82
Homepage: http://www.bke.hu/bartus
--
David Clingingsmith
Department of Economics
Harvard University
�
email: [email protected]
office: 617-588-1473
*
* 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/
