Statalist


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

RE: st: gradient for each observation from ml


From   "Michael Anthony Carlton" <carlto12@msu.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: gradient for each observation from ml
Date   Sat, 19 Jul 2008 08:52:22 -0400

Unfortunately the multivariate normal does not have a closed form for the
CDF, so I am using the mdraws, mvnp plug-in from Capalleri and Jenkins
(2006). This is an lf method, and I know for certain that Stata calculates
the OPG for each observation before summing it and displaying the variance
matrix...I am just not sure how to get it to print these observation level
OPGs either into the dataset or a matrix. There has to be a simple way to
get these since it is a necessary calculation in the estimation of the ml
problem.

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Stas Kolenikov
Sent: Friday, July 18, 2008 1:21 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: gradient for each observation from ml

If you are coding this as -d1- or -d2- method, you have to code the
derivatives, anyway. If you are using -lf-, then there should be some
technical temporary variables created by Stata, but you should not
mess with them (I tried for some of my projects, and it never worked
as intended).

You might want to switch to Mata and -optimize- where a lot of things
come more naturally in terms of matrix computations.

On 7/18/08, Michael Anthony Carlton <carlto12@msu.edu> wrote:
>  Does anyone know if there is a way to get the calculated gradient (or
outer
>  product of the gradient) for each observation used in an ml command? I am
>  estimating a multivariate fractional response model where the first step
is
>  to estimate a multivariate "probit" for each time period and the second
step
>  is to estimate the final parameters using minimum distance. I need to use
>  the observation level OPG to estimate the correlation between time
periods
>  in order to create the optimal weighting matrix for the MD estimation.

-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: Please do not reply to my Gmail address as I don't check
it regularly.
*
*   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/


*
*   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