[R] naive variance in GEE

[Prof Brian Ripley]

On Mon, 8 Sep 2008, Qiong Yang wrote:

> The standard error from logistic regression is slightly different from the 
> naive SE from GEE under independence working correlation structure.
>
> Shouldn't they be identical? Anyone has insight about this?

They are computed quantities from iterations with different stopping 
criteria.  The coefficients are not 'identical' either.

Your example is incorrect (the first line) and not reproducible (no seed 
is set, no library gee), so we don't know what you saw.  But with

set.seed(1)
a <- rbinom(1000, 1, 0.2)
b <- rbinom(1000, 2, 0.1)
c <- rbinom(1000, 10, 0.5)
library(gee)
summary(gee(a ~ b, id=c, family="binomial", corstr="independence"))$coef
summary(glm(a ~ b, family="binomial"))$coef

the differences I see are negligible.  I suggest you talk to your 
supervisor about some courses on numerical methods.

>
> Thanks,
> Qiong
>
> a<-rbinom(1000,1)
> b<-rbinom(1000,2,0.1)
> c<-rbinom(1000,10,0.5)
> summary(gee(a~b, id=c,family="binomial",corstr="independence"))$coef
> summary(glm(a~b,family="binomial"))
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

-- 
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