Zero bound data


New Member

I have perhaps a very simple question but here it goes:

I am working with very zero bound health care services data (measured in hours per month). I have panel data over 9 months (I could use it as cross-sectional but the problem stays) with three measuring points, very zero bound in all time points. The main objective is to find out which clinical factors affect service use over 9 months period (so to exclude the zero user group is not an option). Obviously, GLM is not an option because of the distribution, Generalized linear model with gamma distribution is not an option either, as the model does not converge (I have read about adding some noise, but I don't know if this is OK if the subject actually doesn't use the service). Poisson regression would probably be fine if I would use days services used, but I have hours. Non-parametric models in kind of a "better not" in my field. To make things worse, the data is clustered and it would be very fine if I could take this to account.

Any ideas?

Best regards, M.


Less is more. Stay pure. Stay poor.
What does the distribution of the dependent variable look like? Do the estimates or precision estimates get any where near zero?