Model selection with missing covariates for policy considerations in fox enclosures. Issue 15 (18th November 2017)

• Record Type: Journal Article
• Authors: Hoegh, Andrew; Crandell, Ian; Klopfer, Scott; Fies, Mike
• Is Part Of: Journal of applied statistics. Volume 44:Issue 15(2017)
• Page Start: 2645
• Page End: 2658
• Publication Date: 2017-11-18
• Subjects: Reversible jump MCMC -- MCMC -- data augmentation -- survival analysis -- missing data -- Bayesian modeling; 62-07 -- 62P12; Statistics -- Periodicals; 519.5
• DOI: 10.1080/02664763.2016.1259401
• Languages: English
• ISSNs: 0266-4763

Abstract

ABSTRACT: Foxhound training enclosures are facilities where wild-trapped foxes are placed into large fenced areas for dog training purposes. Although the purpose of these facilities is to train dogs without harming foxes, dog-related mortality has been reported to be an issue in some enclosures. Using data from a fox enclosure in Virginia, we investigate factors that influence fox survival in these dog training facilities and propose a set of policies to improve fox survival. In particular, a Bayesian hierarchical model is formulated to compute fox survival probabilities based on a fox's time in the enclosure and the number of dogs allowed in the enclosure at one time. These calculations are complicated by missing information on the number of dogs in the enclosure for many days during the study. We elicit expert knowledge for a prior on the number of dogs to account for the uncertainty in the missing data. Reversible jump Markov Chain Monte Carlo is used for model selection in the presence of missing covariates. We then use our model to examine possible changes to foxhound training enclosure policy and what effect those changes may have on fox survival.