Problems with zinb

[rcoulter]

I am currently running a zero-inflated negative binomial model to examine count data in a dataset of 20,000 participants. I have a high amount of zero values, and the data is overdispersed -- thus zinb is appropriate.

My model is not converging though -- specifically when fitting the full model it says "not concave."

I think it's because of the 'excess zero' part of the model. Below is my code. I am curious to know how you usually decide what covariates go into the "inflate" part. Do you add all covariates? What's the thinking behind it? Any info is appreciated.

zinb month_GREEK_DRINK i.sexor age21 dlatino dasian dblack dother unemployed home havecar greeklife fresh soph junior intervention Camp2 Camp3 Camp4 Camp5 Camp6 Camp7 Camp8 Camp9 Camp10 Camp11 Camp12 Time2 Time3 Time4 Time5 Time6 Time7 Time8 Time9 if male==1, irr exposure(f1) inflate(i.sexor dlatino dasian dblack dother intervention fresh soph junior Camp2 Camp3 Camp4 Camp5 Camp6 Camp7 Camp8 Camp9 Camp10 Camp11 Camp12, offset(lnf1))

[Maarten Buis]

The trick is to start with a much much simpler model, and gradually work your way up to more complex models.

As an aside, I looks like you have a lot of indicator (dummy) variables. To make full use of the facilities available in Stata you need to use the factor variable notation.