# Can I use negative binomial regression for non-count data?

I have a set of data and I want to model a response variable which is "amount of money paid by the participants". I have data of about 600 people, who chose to pay between 0 and 100 (only integers). These data do not follow a normal distribution, and there are many zeros.

Using a linear model does not seem appropiate because residuals are not normal and model plots have strange patterns. I thought I could use a zero-altered negative binomial (or hurdle model, I don't know which term is more appropriate), but all the literature I find about negative binomial regressions refers only to count data.

Would it be possible to use negative binomial regression even if my data does not refer to counts, given that my response variable cannot take values under zero and that only integers are allowed?

Thank you!

• A quantile regression (presumably predicting the median/50th percentile) sounds like it might be a better option for you. – TPM May 7 '18 at 15:15
• Negative binomial is unbounded at the high end, but maybe a zero-inflated binomial? – HStamper May 7 '18 at 20:52
• Thank you for your comments! I had not thought about quantiles regression, I will try to learn how to do that in R. Regarding zero-inflated binomial, I understand that this model separates zeros from actual counts, as if zeros were "false zeros", and that is why I discarded it (my zeros are real zeros) in favour of zero-altered binomial, which models zeros with the counts; however, I don't know if I am mistaken at some point. – ClaudiaRR May 8 '18 at 6:45