Modeling confidential data via modified hurdle mixed models

In this work we analyze event count data that show zero inflation with non-ignorable missingness due to confidentiality. The emphasis is not on imputation methods but on the choice of a suitable model. As a motivating example, we analyze a dataset on University student’s mobility (SM) in Italy whose records are reported aggregately. In such data, records with less than three moving students are automatically removed. To detect the determinants of SM, similarly to a hurdle mixed model, we estimate two separate models, a binomial mixed model for the ”zero” part and a two-truncated negative mixed binomial for the ”non-zero” part.