[R-sig-ME] (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate

[Philipp Singer]

After the discussions over the previous few days regarding my count data 
modeling, I am currently following a two-step hurdle model approach 
where my first goal is to model the binary data with a binomial model. 
However, when I incorporate one of my explanatory variables I end up 
with the following error:

Error in eval(expr, envir, enclos): (maxstephalfit) PIRLS step-halvings 
failed to reduce deviance in pwrssUpdate

For the other one, it works fine. Strangely, if I include it also as 
random effect, it also appears to work.

I have also tried to use glmmTMB, and also glmmADMB.

I have also some troubles when modeling the non-zero data with nbinom or 
truncated nbinom.

I have summarized most in a notebook:
http://nbviewer.jupyter.org/gist/psinger/4ec0a84e4c0d73be5add1c9b3109a020

Data can be found here:
https://www.dropbox.com/s/mlwopl18a1heb48/r_sample_1000.csv.gz?dl=1 
<https://www.dropbox.com/s/kqqxmc3wp225lug/r_sample.csv.gz?dl=1>

	[[alternative HTML version deleted]]