Title: Information Theoretic Approaches for Quantifying Model Misspecification

Author: Charles Driver

Affiliation: Max-Planck-Institut für Bildungsforschung Berlin

Abstract:
Inferential statistics and interpretations of parameters are generally
dependent on a 'correctly specified' model, although most of us would
acknowledge that specifying the true model for real world phenomena is
generally impossible. I will present an approach using information theoretic
measures to try and better quantify model misspecification, guide model
improvement, and thereby hopefully improve predictive and inferential capacity
for theory based modelling approaches.