Objective Bayes Variable Selection for Site-Occupancy Models Using Latent Normal Mixtures

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Daniel Taylor Rodriguez*+ and Claudio Fuentes and Andrew Womack and Nikolay Bliznyuk
• Companies: University of Florida and and University of Florida and University of Florida
• Keywords: occupancy model ; data-augmentation ; intrinsic priors ; binary regression ; model priors ; model uncertainty

Abstract

The last two remaining jaguar populations in the world are in South and Central America. To ensure an active pathway between them, a first step is to produce distribution maps for jaguars and their prey in Colombia -the necessary passage between the two continents. Site-occupancy models are well suited for this purpose, these estimate the percentage of area occupied by a species adjusting for imperfect detection. This model is used extensively to inform population management decisions, however, Bayesian variable selection methods using suitable testing priors (both on the parameter and model spaces) are yet to be put forth. By specifying the occupancy model as a mixture of latent normal variables, we build an Objective Bayes variable selection procedure based on intrinsic priors, suggesting different penalization strategies that automatically correct for the multiplicity of the testing problem, given the cardinality of the model space considered. We use our methods to identify the drivers of jaguar population occupancy in Colombia, and provide credible intervals for all model parameter estimates that accommodate for model uncertainty.