PRIORS COMPARISON IN BAYESIAN MEDIATION FRAMEWORK WITH BINARY OUTCOME

Jean-Michel Galharret (1) , Anne Philippe

1 LMJL - Laboratoire de Mathématiques Jean Leray

In human sciences, mediation designates a particular causal phenomenon where the effect of a variable X on another variable Y passes (par-tially or entirely) through a third variable M. The parameters of interest in mediation models are the direct effect of X on Y and the indirect effect of X on Y through M. We use a Bayesian framework to estimate these parameters and we compare different construction of prior distribution. The results show that providing information improves the quality of the estimation. We also propose a Bayesian procedure to test the absence of direct and indirect effect. Results of simulations show that this procedure works as well as the usual frequentist test for the direct effect and better than the usual bootstrapping test for the indirect effect. Finally, we apply our approach to real data from a longitudinal study on the well-being of children in school.

Mots clés: mediation analysis; Bayesian estimation; direct and indirect effect

Domaines: Statistiques [math.ST]

Jean-Michel Galharret, Anne Philippe. PRIORS COMPARISON IN BAYESIAN MEDIATION FRAMEWORK WITH BINARY OUTCOME. 2019. ⟨hal-02070053⟩