A robust P-spline approach to closed population capture–recapture models with time dependence and heterogeneity

We extend the conditional likelihood approach to the analysis of capture–recapture experiments for closed populations by nonparametrically modeling the relationship between capture probabilities and individual covariates using P-splines. The model allows nonparametric functions of multivariate continuous covariates as well as categorical covariates and time effects, greatly enhancing the techniques available to an analyst. To implement this approach in practice, we found it necessary to develop a robust modification of the Horvitz–Thompson estimator. The method is illustrated on several data sets and a small simulation study is conducted.