Informative missing data.

schedule le lundi 21 octobre 2019 de 17h00 à 18h00

Organisé par : F. Bechtold, W. Da Silva , A. Fermanian, S. Has, Y. Yu

Intervenant : Aude Sportisse (Sorbonne Université)
Lieu : Jussieu, salle Paul Lévy, couloir 16-26, salle 209

Sujet : Informative missing data.

Résumé :

This talk will be devoted to the problem of missing data, which is ubiquitous in the practice of data analysis. Theoretical guarantees of estimation strategies or imputation methods rely on assumptions regarding the missing data mechanism, i.e. the cause of the lack of data. We will focus on Missing Not At Random data (MNAR) case, when the probability of being missing depends on unobserved data such as its value itself and  which is extremely frequent in practice. A classic exemple of MNAR data is surveys where rich people would be less willing to disclose their income. We propose two different imputation and estimation methods: one relies on the modeling of the missing data mechanism, the other on graphical models and causality.