New Diagnostic Tools for Capture-Recapture Models

Capture-recapture models have increased in complexity over the last decades and goodness-of-fit assessment is crucial to ensure that considered models provide an adequate fit to the data. In this thesis, my primary emphasis is to develop new diagnostic tools for capture-recapture models in order to target possible reasons of lack-of-fit, which might provide biological insights and point towards better-fitting candidate models.

Starting with the basic Cormack-Jolly-Seber model, I develop a new tool for detecting heterogeneity in capture. I then progress to the more complex multi-state models, for which I propose a test for detecting a mover-stayer structure within the population. Finally, I move on to more general models presenting additional levels of uncertainty: first partial observations and then unobservable states. In the presence of partial observations, part of the observations are assigned to states with certainty whereas others are not. I develop a new test for the underlying state-structure of the partial observations, this test detects that the partial observations are not generated by the observable states defined in the experiment. In the presence of unobservable states, the additional level of uncertainty relates only to the non-captures. I present a procedure to test whether one or two unobservable states need to be defined in order for the model to provide an adequate fit to the data.

Lastly, I explore the use of multi-state models to incorporate individual time-varying covariates, based on a fine discretisation of the covariate space.