Dear fellow list users, we would like to analyze three variables that are count data (no zeros, no integers) as responses to two categorical predictor variables using compar.gee. An example would be
compar.gee(FlowerNumber ~ sex + PollinationSystem + sex*PollinationSystem). One of our variables is normally distributed, the other two can be normalized by log-transformation. So far we have modelled our data using family=gaussian, but we would prefer to model untransformed data using a log-link function, i.e. family=poisson. However, several of our models with family=poisson crash R or result in the following warning messages: Warning messages: 1: In gee(noFH ~ poll * sex, c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, : Maximum number of iterations consumed 2: In gee(noFH ~ poll * sex, c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, : Convergence not achieved; results suspect 3: In gee(noFH ~ poll * sex, c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, : Cgee had an error (code= 104). Results suspect. Can any of you explain why are we getting this warning message? Is there a fix to any of this? Alternatively, if family=poisson really doesn't work for our data, is it possible to change the link function associated with family=gaussian from link=identity to link=log? We tried this by writing compar.gee(model, data, phy, family=gaussian(link="log")) and get the following error message: Error in switch(fname, gaussian = -sum((Y - MU)^2)/2, binomial = sum(Y * : EXPR must be a length 1 vector In addition: Warning message: In if (fname == "binomial") W <- summary(glm(formula, family = quasibinomial, : the condition has length > 1 and only the first element will be used Is it at all possible to change the link function in compar.gee? Any input would be greatly appreciated. Nina Hobbhahn and Megan Welsford _______________________________________________ R-sig-phylo mailing list - [email protected] https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/[email protected]/
