Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: mvprobit and changing results


From   "Stephen P. Jenkins" <stephenj@essex.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   Re: st: mvprobit and changing results
Date   Tue, 22 Jul 2008 12:24:19 +0100

=====================================

Date: Mon, 21 Jul 2008 17:33:56 -0500
From: "Stas Kolenikov" <skolenik@gmail.com>
Subject: Re: st: mvprobit and changing results

- -mvprobit- is a simulation-based estimator. You need to fix the
random
number generator seed, see the corresponding option of -mvprobit-.
However if you see drastic differences between runs, that probably
means that the simulation variability dominates the underlying
sampling variability, so you would want to increase the number of
- -draws-. I remember having a discussion with Stephen Jenkins about
the
number 5 which I found ridiculously low, but he in the end convinced
me that it works OK in most applications. Yours might be an exception
- -- increase that number until the results with different -seed-
values
become more stable.
=====================================

You /must/ set the seed in advance using the seed() option.  See also
the Cappellari-Jenkins 2003 Stata Journal article which discusses this
issue. (You can download the article for free from the SJ website)

My recollection of the conversation Stas refers to differs from his.
The point of setting the default # of draws at 5 is that is obviously
too low to use for real life applications. However, given that
-mvprobit- is a simulation-based estimator and numerical intensive, it
can takes a long time to converage. So, using a small # of draws can
save time /when testing syntax/.  A very crude rule of thumb is that #
draws should be at least as large as the square root of the sample
size. Again, see the Cappellari-Jenkins 2003 Stata Journal article 

See also the special issue of the Stata Journal, 6(2), on maximum
simulated likelihood (MSL). Including the Cappellari-Jenkins article,
which discusses way to speed up estimation of MV Probit and related
models, and provides code. (Pre-print is a working paper on ISER
website)

Please do not use -mvprobit- without having first read some literature
on MSL (it's in standard texts like Greene's Econometric Analysis now)
and the help files in detail.

Stephen
-------------------------------------------------------------
Professor Stephen P. Jenkins <stephenj@essex.ac.uk>
Director, Institute for Social and Economic Research
University of Essex, Colchester CO4 3SQ, U.K.
Tel: +44 1206 873374.  Fax: +44 1206 873151.
http://www.iser.essex.ac.uk  
Survival Analysis using Stata:
http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/ 
Downloadable papers and software: http://ideas.repec.org/e/pje7.html

Learn about the UK's new household panel survey, the United Kingdom
Household Longitudinal Study: http://www.iser.essex.ac.uk/ukhls/ 


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index