# How to model zero-inflated continous (from negative to positive) data

We have a dataset of around 20k observations. The dependent variable is the change (i.e. delta) on the amount of a common resource (e.g. land) of individual households in a year, so:

1. It has negative and positive values (some increased their land, while others decreased it)
2. Around 70-80% of observations are 0s (most of them did not exchange land)
3. It's a common and static resource (the sum of the dependent variable for the whole dataset is 0)
4. The independent variables are continuous or categorical

We tried multiple transformations but it doesn't really work. I was reading about the zero-inflated approaches, but most of them are non-negative count variables. We're trying quantile regression, but not yet sure if that would be the best approach.

Would you have any suggestions?

• It depends on the purpose of the analysis but you could consider having two models (a) model the probability of being a zero, (b) conditional on not being a zero model the change score. Nov 12 '20 at 15:40
• Thank you. So first a binomial logistic regression to understand the 0s and non-0s, and then a linear regression excluding all the 0s? Nov 19 '20 at 15:20