[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Covariance Matrix transform in MLE estimation

From   Maarten buis <>
Subject   Re: st: Covariance Matrix transform in MLE estimation
Date   Wed, 29 Apr 2009 12:50:35 +0000 (GMT)

--- On Wed, 29/4/09, Malcolm Wardlaw wrote:
> I've got a mixed bernoulli type log-likelihood function
> that I'm using the inverselogit(-x-) trick in order to 
> keep Stata from searching for lambda values less than 0 
> or greater than 1.  Essentially telling it to estimate 
> x=logit(-lambda-) instead of -lambda- itself.
> I know how to ask Stata for the transformed parameter and
> variance estimate (That is, transforming the logit(-lambda-)
> estimate into - lambda-. -->nlcom invlogit([logitkappa]_cons).
> What I need are the covariance terms of the parameters. 
> Essentially the off diagonals of the Hessian matrix (as if I 
> was estimating -lambda- itself along with the other
> parameters.  How do I recover that transformation?

use -nlcom-:
nlcom (kappa: invlogit([logitkappa]_cons)) ///
      (x1: x1) (x2:x2), post

This will leave the variance covariance matrix between
kappa x1 and x2 behind in e(V).
> Oh, and on a side note, does anyone know why you have to
> use tricks like that instead of just telling Stata not to 
> search for parameters there?

I am speculating, but it might because this way you can
avoid estimates at the boundery of the parameter space. 

-- Maarten

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index