[R-sig-ME] glmmTMB convergence with unique outcome count 0 for a factor regressor category

[IAGO GINÉ VÁZQUEZ]

Dear all,

I am trying to fit a Negative Binomial mixed model, possibly zero-inflated with glmmTMB. I tried multiple offsets, removing covariates, including and removing the ziformula option, but it didn't converge, usually with output message
Error in (function (start, objective, gradient = NULL, hessian = NULL,  :
  NA/NaN gradient evaluation
In addition: Warning messages:
1: In (function (start, objective, gradient = NULL, hessian = NULL,  :
  NA/NaN function evaluation
2: In (function (start, objective, gradient = NULL, hessian = NULL,  :
  NA/NaN function evaluation
etc.

Now I found a fact on the data. For a factor regressor, the observations restricted to one of its values (in fact, the reference value) have a unique value for the outcome count, which is 0.

Is it the reason why the models don't converge?, so any model with a factor regressor with a category for which the unique outcome is 0 will never converge?

In that case:
Is relevant that this category for which the outcome is exclusively 0 is the reference one? If I change the reference category, could the model converge?
Is there any (other) solution to this, keeping the factor in the model?

Thank you in advance!
Iago


	[[alternative HTML version deleted]]