I am trying to model a dataset with GLMs but I am wondering how to start with the first step of fitting a maximal model that tests all covariates and their interactions when there are simply not enough combinations of observations to do so? I am getting error messages when I try to fit a full model with all interactions but not if I take out the interactions. Since my statistics texts recommend the approach of starting with a 'full model'.. how do you start from full when you can't? These are types of errors I am getting:
"Error: no valid set of coefficients has been found: please supply starting values"
"Error in glm.fit(x = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, : NA/NaN/Inf in 'x' In addition: Warning message: step size truncated due to divergence "
When I try to do the following models, for example: Glm1 <- glm(DPMSpeciesA ~ DPMSpeciesB* Diel* Tidal_phase* Tidal_cycle * Month* Julian_Date, data=data1, family=poisson).
(the first error)
Glm2<-glm(DPMSpeciesA ~ DPMSpeciesB * Diel * Tidal_phase * Tidal_cycle * Month, data=data1, family=poisson)
(the second error) Where DPM is detection positive minutes. However, when I change the format to: Glm3 <- glm(DPMSpeciesA ~ DPMSpeciesB + Diel + Tidal_phase + Tidal_cycle + Month + Julian_Date, data=data1, family=poisson), it works fine.
I am assuming the reason for the errors is because there are not enough combinations of observations to test for all the interactions... so where would be appropriate place to start the model from so that I can use stepwise regression?