I am looking for some clarification on specifying ziformula= in the glmmTMB package. I understand if you have zero-inflation then ziformula=~1 if inflation is independent or any predictors (or alternatively you can specify which terms) BUT can you specify your response term for zero-inflation or is that invalid? As my response term is the term with the many zeros!
For example, I am looking at proportion of seeds that have germinated and as there are many plots with zero germination I have ALOT of zeros in my response term.
for example my model script:
m3 <- glmmTMB (germ/total~ suit + cage + soilstd+ gapstd + cage:suit
+cage:soilstd + cage:gapstd + (1|site/plotID),ziformula=~germ,
weight=total,family=binomial,data=data)
Where my response= germ (# seeds germinated) / total (25 planted)
When I specify the ziformula=~germ I run the model with no errors but if I change it to ziformula=~1 I get the "Model convergence problem; non-positive-definite Hessian matrix" warning. The only way I found to by-pass this warning it to not include interactions but obviously I want a model with interactions as I think they are ecologically valid. For reference I have standardize the continuous variables (soilstd, gapstd) and the others are factors (suit, cage).
If anyone can provide insight on if specifying a response variable for ziformula is valid? Also if not (if it should indeed be 1) other suggestions for eliminating the error? I did see this reference: https://cran.r-project.org/web/packages/glmmTMB/vignettes/troubleshooting.html BUT I cannot understand some of the stat processes outside of standardizing and trying a different optimizer (which only eliminated one convergence error). Thanks!
