Rencontres Statistiques Lyonnaises

Methodological developments around causal inference and the analysis of high-dimensional data

par Lola ETIEVANT (ICJ)

Europe/Paris
112 (Bâtiment Braconnier)

112

Bâtiment Braconnier

Description

In my thesis project, I explored some of the problematics raised by the practical application of causal inference in cancer epidemiology. I will quickly present the four distinct projects I have been working on, and will then detail two of them. In the first one, we will investigate conditions ensuring that estimates derived under over-simplified causal models, where the longitudinal nature of the variables have been neglected, relate to causal quantities of interest under the true longitudinal causal model. We will then focus on the probabilistic formulation of partial least squares proposed by el Bouhaddani et al. (2018), to describe a limitation we have identified in several models proposing probabilistic formulations of dimension-reduction techniques. We will further illustrate the limitation through simulated examples.