Message-ID: <Pine.LNX.4.44.0402190727440.19592-100000@gannet.stats>
Date: 2004-02-19T07:30:56Z
From: Brian Ripley
Subject: Generalized Estimating Equations and log-likelihood calculation
In-Reply-To: <1077125378.4033a10213b47@webmail.telus.net>
On Wed, 18 Feb 2004 treebc at telus.net wrote:
> I'm working with clustered data sets and trying to calculate log-likelihood
> (and/or AIC, AICc) for my models. In using the gee and geese packages one
> gets Wald test output; but apparently there is no no applicable method
> for "logLik" (log-likelihood)calculation.
>
> Is anyone aware of a way to calculate log-likelihood for GEE models?
No (as with GLM quasi- models, it is not defined in general). Even if
there were, you would have find the maximized log-likelihood to find AIC,
and by definition GEE is not ML fitting except in a few special cases.
--
Brian D. Ripley, ripley at stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595