Accompanying note: Model-based Clustering with Missing Not At Random Data

Description

This document is the accompanying note of the main paper "Model-based Clustering with Missing Not At Random Data". We assume the data missing not at random (MNAR) values, i.e. the effect of missingness depends on on the missing values themselves.An example includes clinical data collected in emergency situations, where doctors may choose to treat patients before measuring heart rate: the missingness of heart rate depends on the missing heart rate itself. For such a setting, the observed data are therefore not representative of the population. The main paper focuses on the specific MNARz setting, for which the only effect of missingness is on the class membership; in this document, we give some details for other MNAR settings.

Additional details