Explained Variation in Models for Mortality Data

Mortality data on an aggregate level is characterized by very large sample sizes. For this reason, uninformative outcomes are evident in common goodness-of-fit measures. In collaboration with Maria L. Durban Reguera from the Universidad Carlos III de Madrid I proposed a new measure that compares different mortality models, including those with large sample sizes. Specifically, I developed a measure which uses a null model specifically designed for mortality data. Several simulation studies and actual applications demonstrated the performance of this new measure with special emphases on demographic models and P-spline approaches.

A manuscript about this measure has been published in Statistics and Econometrics Working Papers collection of the Universidad Carlos III: Goodness of Fit in Models for Mortality Data (link).