The (in)stability of bayesian model selection criteria in disease mapping
- Autores: Maren Vranckx, Thomas Neyens, Christel Faes
- Localización: Proceedings of the 35th International Workshop on Statistical Modelling : July 20-24, 2020 Bilbao, Basque Country, Spain / Itziar Irigoyen Garbizu (ed. lit.), Dae-Jin Lee (ed. lit.), Joaquín Martínez Minaya (ed. lit.), María Xosé Rodríguez Álvarez (ed. lit.), 2020, ISBN 978-84-1319-267-3, págs. 446-449
- Idioma: inglés
- Enlaces
-
Resumen
Several model comparison techniques exist to select a best model from a set of candidate models. This study explores the performance of model comparison statistics among several Bayesian software packages that are often used for spatially discrete disease modelling: the deviance information criterion (DIC), the Watanabe-Akaike information criterion (WAIC) and the log marginal predictive likelihood (LMPL). We focus on the software packages CARBayes, OpenBUGS, NIMBLE and Stan, in which we fit Poisson models to disease incidence outcomes with intrinsic conditional autoregressive, convolution conditional autoregressive and log-normal error terms. From data studies, we learn important disparities in model selection. Based on these conclusions, we provide recommendations on the optimal use of model comparison statistics for all kind of applications.
