# Zero-inflated model underestimates the number of zeroes due to “near zero values”

I am using GAMLSS. The data looks like this:

So, I tried a zero-inflated beta model (beta because data lies in the [0,100] interval, and so I just divided by 100).

The estimates I got using GAMLSS showcase a very low $$\nu$$ estimate, i.e., the probability of a zero occuring. So, essentially, it just fitted a standard beta fit, a bit like this one on the top right corner:

Why does it do this? How can I fix it? What other distributions are there?

My data clearly DOES contain zeroes, many of them, so a standard beta is not a good fit.

• Do you have a small data set that shows the same problem? (possibly a subset of your data, if your own data set is large) – Glen_b Apr 27 '17 at 4:55