Principles and interest of GOF tests for multistate capture-recapture models

Abstract

Optimal goodness–of–fit procedures for multistate models are new. Drawing a parallel with the corresponding single–state procedures, we present their singularities and show how the overall test can be decomposed into interpretable components. All theoretical developments are illustrated with an application to the now classical study of movements of Canada geese between wintering sites. Through this application, we exemplify how the interpretable components give insight into the data, leading eventually to the choice of an appropriate general model but also sometimes to the invalidation of the multistate models as a whole. The method for computing a corrective overdispersion factor is then mentioned. We also take the opportunity to try to demystify some statistical notions like that of Minimal Sufficient Statistics by introducing them intuitively. We conclude that these tests should be considered an important part of the analysis itself, contributing in ways that the parametric modelling cannot always do to the understanding of the data.