Multistage Hierarchical Capture-Recapture Models

TL;DR

This paper introduces a multistage hierarchical Bayesian approach for capture-recapture models that improves computational stability and efficiency, especially with latent random effects, by using a conditional model structure and parallel processing.

Contribution

It proposes a novel multistage hierarchical model specification that enhances computational stability and allows parallel fitting of capture-recapture models with latent effects.

Findings

More stable computation with the new model specification

Successful application to simulated and real data sets

Efficient use of parallel computing resources

Abstract

Ecologists increasingly rely on Bayesian methods to fit capture-recapture models. Capture-recapture models are used to estimate abundance while accounting for imperfect detectability in individual-level data. A variety of implementations exist for such models, including integrated likelihood, parameter-expanded data augmentation, and combinations of those. Capture-recapture models with latent random effects can be computationally intensive to fit using conventional Bayesian algorithms. We identify alternative specifications of capture-recapture models by considering a conditional representation of the model structure. The resulting alternative model can be specified in a way that leads to more stable computation and allows us to fit the desired model in stages while leveraging parallel computing resources. Our model specification includes a component for the capture history of detected individuals and another component for the sample size which is random before observed. We demonstrate this approach using three examples including simulation and two data sets resulting from capture-recapture studies of different species.