[R] Convergence problem with zeroinfl() and hurdle() when interaction term added

[Sarah Valencia]

Hello,

I have a data frame with 1425 observations, 539 of which are zeros. I
am trying to fit the following ZINB:

f3<-formula(Nbr_Abs~ Zone * Year + Source)
ZINB2<-zeroinfl(f3, dist="negbin", link= "logit", data=TheData,
offset=log(trans.area), trace=TRUE)

Zone is a factor with 4 levels, Year a factor with 27 levels, and
Source a factor with 3 levels. Nbr_Abs is counts of a species that
shows a high level of aggregation. These counts are offset by the area
searched per transect.

The trace output and error message are as follows:

Zero-inflated Count Model
count model: negbin with log link
zero-inflation model: binomial with logit link
dependent variable:
   0    1    2    3    4    5    6    7    8    9   10   11   12   13   14
 539  125   72   41   33   35   35   31   15   22   22   11   13   16   13
 .... (truncated for brevity)...
 285  286  287  288  289  290  291  292  293  294 <NA>
   0    0    0    0    0    0    0    0    0    1    0
generating starting values...done
calling optim() for ML estimation:
Error in optim(fn = loglikfun, gr = gradfun, par = c(start$count,
start$zero,  :
  non-finite value supplied by optim
In addition: Warning message:
In glm.fit(Z, as.integer(Y0), weights = weights, family =
binomial(link = linkstr)) :
  fitted probabilities numerically 0 or 1 occurred


I get the same optim error when I run a similar call using hurdle
instead of zeroinfl. However, both commands work fine when the
interaction terms is removed ( Nbr_Abs~ Zone + Year + Source). Is this
a case of some kind of linear relationship between my covariates? In
addition, I can run a negative binomial glm with the interaction term,
which I didn't think would be possible if that were the case.

Any help would be much appreciated!

Thanks,
Sarah

--
Sarah Valencia, PhD student

Bren School of Environmental Science and Management

University of California Santa Barbara, CA 93106

Lab: (805) 893-5054