The variance of the conditional Poisson is the mean, not the inverse of the mean (using standard parameterization). You can use sub=intercept when you want all observations to be correlated, not just within plots (special syntax to show that there is one big subject). You are modeling spatial covariance as a G-side effect, not an R-side one. Thus, the variance of the G side term is not related to the conditional variance of the Poisson. These are different things.
... View more
