Methods for investigating parameter redundancy

Abstract

The quantitative study of marked individuals relies mainly on the use of meaningful biological models. Classical inference is then conducted based on the model likelihood, parameterized by parameters such as survival, recovery, transition and recapture probabilities. In classical statistics, we seek parameter estimates by maximising the likelihood. However, models are often overparameterized and, as a consequence, some parameters cannot be estimated separately. Identifying how many and which (functions of) parameters are estimable is thus crucial not only for proper model selection based upon likelihood ratio tests or information criteria but also for the interpretation of the estimates obtained. In this paper, we provide the reader with a description of the tools available to check for parameter redundancy. We aim to assist people in choosing the most appropriate method to solve their own specific problems.