Penalized likelihood methods applied to age-period-cohort analysis

schedule le lundi 25 février 2019 de 17h00 à 18h00

Organisé par : A. Lefebvre, N. Meyer, O. Safsafi, T. Touati

Intervenant : Vivien Goepp (Paris Descartes - LPSM)
Lieu : Jussieu, salle 209, couloir 16-26

Sujet : Penalized likelihood methods applied to age-period-cohort analysis

Résumé :

In many epidemiological studies, the incidence of a time-dependent event is estimated along different variables. In the case of cohort data, the age-period-cohort model jointly estimates three time-dependent effects, but it makes too restrictive assumptions on the shape of the incidence. I will expand on two simple generalizations of the age-cohort model that can infer more general shapes of the incidence. The inference is made using a fused regularization method with approximate L0 penalty. We provide some illustrations with real-world data.