2018/10/04 by Soumen Dey, Dey, Soumen, Mohan Delampady +3
Environmental Science · Mathematics · #Applications (stat.AP) #Artificial intelligence #Bayes estimator #Bayes factor #Bayes' theorem #Bayesian hierarchical modeling #Bayesian inference #Bayesian probability #Census and Population Estimation #Computer science #Econometrics #FOS: Computer and information sciences #Machine learning #Mathematics #Model selection #Selection (genetic algorithm) #Statistics #Survey Sampling and Estimation Techniques #Wildlife Ecology and Conservation #stat.AP
paper · pdf · doi:10.48550/arxiv.1810.02397
published in arXiv (Cornell University) (Cornell University)
arxiv created 2018/10/04 · openalex publication_date 2018/10/04 · arxiv updated 2018/10/08 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
A vast amount of ecological knowledge generated recently has hinged upon the\nability of model selection methods to discriminate among various ecological\nhypotheses. The last decade has seen the rise of Bayesian hierarchical models\nin ecology. Consequently, popular tools, such as the AIC, become largely\ninapplicable and other tools are not universally applicable. We focus on a\nclass of competing Bayesian spatially explicit capture recapture (SECR) models\nand first apply some of the recommended Bayesian model selection tools: (1)\nBayes Factor - using (a) Gelfand-Dey (b) harmonic mean methods, (2) DIC, (3)\nWAIC and (4) the posterior predictive loss function. In all, we evaluate 25\nvariants of model selection tools in our study. We evaluate these model\nselection tools from the standpoint of model selection and parameter estimation\nby contrasting the choice recommended by a tool with a `true' model. In all, we\ngenerate 120 simulated data sets using the true model and assess the frequency\nwith which the true model is selected and how well the tool estimates N\n(population size). We find that when information content is low, no particular\ntool can be recommended to help realise, simultaneously, both the goals of\nmodel selection and parameter estimation. In such scenarios, we recommend that\npractitioners utilise our application of Bayes Factor for parameter estimation\nand recommend the posterior predictive loss approach for model selection when\ninformation content is low. When both the objectives are taken together, we\nrecommend the use of our applications of Bayes Factor for Bayesian SECR models.\nOur study reveals that although new model selection tools are emerging (eg:\nWAIC) in the applied statistics literature, an uncritical absorption of these\nnew tools (i.e. without assessing their efficacies for the problem at hand)\ninto ecological practice may mislead inferences.\n