[R-sig-ME] Errors in computing GEE models

[Ken Beath]

On 02/10/2008, at 5:39 AM, david dav wrote:

> Hi,
> I posted this on the general R-announce list a few weeks ago. I was
> suggested to try it here :
>
> I would like to fit a GLM model with GEE on clustered data.
> I tried to use gee in the GEE package on a twin data set. All cluster
> are of size 2. I removed the missing data first.
>

Trying geepack may be the best solution.

Check what is actually returned from na.omit and that there is not  
perfect correlation between clusters. Changing to "exchangeable"  
corstr shouldn't make any difference but maybe worth a try.

Ken



> library(gee)
> mod.Y <- gee(Y ~ X1 + X2 , id = IDENTIF2,  family = binomial, corstr
> = "unstructured", data = na.omit(df))
>
> Beginning Cgee S-function, @(#) geeformula.q 4.13 98/01/27
> running glm to get initial regression estimate
> (Intercept)         X1M  X2Sans AMP      X2Stim
> -2.7756284   0.4658861  -1.6508288  -1.5059518
> Erreur dans gee(Y ~ X1 + X2, id = IDENTIF2, family = binomial, corstr
> = "unstructured",  :
> NA/NaN/Inf dans un appel à une fonction externe (argument 2)   #
> meaning in a call to an external function
> De plus : Warning message:
> In gee(Y ~ X1 + X2, id = IDENTIF2, family = binomial, corstr =
> "unstructured",  :
> NAs introduits lors de la conversion automatique   # meaning NAs
> introduced during the automatic conversion
>
> The regression gives results with SAS version 8 on the same dataset
> (results are pretty close from the results of a naive logistic
> regression).
> I tried to change the explanatory variables (X1 alone, X2 alone, other
> possible explanatory variables : no result). I do not find where the
> error could come from.
> Thanks
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>