Bayesian predictive model selection in circular random effects models with applications in ecological and environmental studies

Abstract

In this paper we present a detailed comparison of the prediction error based model selection criteria in circular random effects models. The study is primarily motivated by the need for an understanding of their performance in real life ecological and environmental applications. Prediction errors are based on posterior predictive distributions and the model selection methods are adjusted for the circular manifold. Plug-in estimators of the circular distance parameters are also considered. A Monte Carlo experiment scheme taking the account of various realistic ecological and biological scenarios is designed. We introduced a coefficient that is based on conditional expectations to examine how the deviation from von Mises (vM) distribution, the standard choice in applications, effects the performances. Our results show that the performances of widely used circular predictive model selection criteria mostly depend on the sample size as well as within-sample-correlation. The approaches and selection strategies are then applied to investigate orientational behaviour of Talitrus saltator under the risk of dehydration and direction of wind with respect to associated atmoshperic variables.