[R] GEE code

[Thomas Lumley]

On Fri, 20 Jul 2007, Dimitris Rizopoulos wrote:

> from your description I think you need a random-effects model.

Since he specifically said "fixed effects" I would have assumed from the 
description that he wanted a fixed-effects model.

>								 Since
> you assume normality the parameter estimates from a random-effects
> model (in your case `GDP.log2') have a marginal interpretation (in
> fact they have both conditional and marginal interpretation).

But they still aren't the fixed effects estimates, and they behave 
quite differently under model misspecification (and under confounding).

A fixed effects linear model with GEE just needs the formula
   ineq~GDP.log2+country.
to specify an indicator variable for each country.

If he had a logistic regression model things would be more complicated, 
but for a linear or log-linear model it is just a matter of adding 
predictors.

Now, I might well use a linear mixed model in this context, but he did 
fairly clearly indicate that wasn't he was looking for.

 	-thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle