I have a situation where an experiment is being run in the following manner:
A one stage cluster sampling (I think this is accurate description ) is conducted whereby there are multiple organizations selected (10 of them)and within each of these organization, there are sub-organizations, whose members are all assigned as either treatment or control. So, within the larger selected organizations, the members of the smaller organizations are all treatment or all control. An analogy could be school districts being the top level selected organizations, whereby a school is selected within the district and their students are all treatment or all control.
The variable being measured is a continuous variable with a mass at zero.
I need to determine a confidence interval on the treatment effect taking into account the clustering and the zeros in the target variable.
What are my best options?
I have thought about maybe some permutation test where all combinations of treatment and control are scrambled , where all treated sub organization become control and vice versa.
The other thought was a tweedie regression with a random intercept. The cplm package in R seems to do this, but a confidence interval needs to be a wald CI ( estimate +/- 1.96 * SE).