[R] Quasipoisson with geeglm
Soren.Hojsgaard at agrsci.dk
Thu Apr 7 17:05:22 CEST 2011
In gee there is no quasipossion, because gee is in a way already quasi.
With GEE we do not fit a poisson glm, but use in the construction of the sandwich covariance matrix the variance function of the poisson family. In Gee always an 'overdispersion' is estimated.
Fra: r-help-bounces at r-project.org [r-help-bounces at r-project.org] På vegne af JANSEN, Ivy [Ivy.JANSEN at inbo.be]
Sendt: 7. april 2011 13:32
Til: r-help at r-project.org
Emne: [R] Quasipoisson with geeglm
I am trying to use the GEE methodology to fit a trend for the number of butterflies observed at several sites. In total, there are 66 sites, and 19 years for which observations might be available. However, only 326 observations are available (instead of 1254). For the time being, I ignore the large number of missing values, and the fact that GEE is only valid under MCAR. When I run the following code
geeglm(SumOfButterflies ~ RES_YEAR, family = poisson, data = ManijurtNoNA, id = RES_ROTE_ID, corstr = "ar1")
I obtain "normal" output. Not surprisingly, overdispersion is present (Estimated Scale Parameters:  185.8571), so changing to quasipoisson is needed. However, the code below
geeglm(SumOfButterflies ~ RES_YEAR, family = quasipoisson, data = ManijurtNoNA, id = RES_ROTE_ID, corstr = "ar1")
produces the following error
Error in geese.fit(xx, yy, id, offset, soffset, w, waves = waves, zsca, : variance invalid.
Other correlation structures encounter the same problem. I also tried adding "waves = RES_YEAR" (although I am not sure how waves should be used correctly), but it does not change anything.
Any suggestions what might be wrong?
[[alternative HTML version deleted]]
R-help at r-project.org mailing list
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
More information about the R-help