I'm wanting to model binary data and count data as an interrelated process through hurdle models in the package pscl. My model setup is as follows;
Best.Fit <- hurdle(MaxN ~ Site * Julian_Date + offset(log(Effort)) | Site + offset(log(Effort)), na.action = "na.fail", data = my.data.counts, dist = "negbin", zero.dist = "binomial")
While I'm doing building models for multiple species, one species was recorded as present on each day of the study causing complete separation. I found a solution herefor only the binary data and built the following code for the 0's and 1's only.
Binary.Only <- glm( Presence ~ Site + offset(log(Effort)),
family = binomial (link = "logit"), method = brglmFit, data = my.data.binary)
It seems like I can't use the bias reduction method in the hurdle model like I can for the binary data. Does anyone have any suggestions as to how to deal with complete separation when applying binary data to the hurdle model?
I did try adding the argument
method = brglmFit
right into the hurdle model but that didn't work and then I later found that hurdle models aren't compatible with this method through the GitHub Issues page
Thanks in advance for your thoughts/help!