s-news
[Top] [All Lists]

robust maximum likelihood estimation

To: s-news@lists.biostat.wustl.edu
Subject: robust maximum likelihood estimation
From: chris.2.wallace@gsk.com
Date: Wed, 23 Jul 2003 16:23:53 +0100
I am a relatively new user of Splus (experienced user of stata)

I would like to use maximum likelihood to fit a model which can be 
expressed as a function of a linear predictor:
L= A*(1-p)^2 + B*p*(1-p) + C*p^2

A, B and C are fixed constants, and p=1/(1+exp(-eta)) where eta=Xb.

I have found the function nlmin and can use this to obtain maximum 
likelihood estimates of b.  However, I would also like to obtain standard 
errors and believe asymptotic properties of MLEs may not hold.  Therefore 
I would like to use clustered robust estimation.

Can anyone point me in the direction of appropriate functions or some 
example code?  The robust library available in Splus v 6.1 does not appear 
to allow user-defined likelihoods.

Many thanks, Chris.



<Prev in Thread] Current Thread [Next in Thread>