Dear helpers, in course of my master thesis I trie to program a GLM to model the number of seeds produced per flower relates to the time they were exposed to pollinators (categorical with two levels), the altitude of population (categorical with two levels) and the species (categorical with six levels). Since I have lots of zeros in my data and detected overdispersion when simply using family=poisson, I choose the quasi family. With the according default link option the homogeneity of variance in the residuals is violated and so I tried to set link=sqrt and the variance=mu^2. This combination should work, but unfortunately it does not. I first got the warning message to specify starting values which I hopefully solved right and I got the summary output and residuals plots. But the anova output still do not work by printing the following warning message: step size truncated out of bounds. I searched in the R helping mail list and also tried to contact people that once had similar problems, but got no reply. I would be glad if anybody cound explain what happens, and point out a solution for me to get the anova(model) output. Thank you a lot in advance. Sincerely, Judith Trunschke