Bayesian model checking, comparison and selection with emphasis on outlier detection for geotechnical reliability-based design

Abstract

Model checking, evaluation and comparison are critical steps in Bayesian data analysis but not routinely considered in statistical analysis of geotechnical data. This paper presents an overview of model checking and comparison concepts and techniques employed in modern Bayesian data analysis that are useful for analysis of geotechnical engineering data. These methods are applied to statistical analysis of load and pullout model bias data for steel strip mechanically stabilized earth walls. It is shown how model checking can result in statistical models with better fits. The practical implications of using improved statistical models are discussed in the context of reliability-based design and load and resistance factor design calibration of the internal stability pullout limit state. Finally, statistical model uncertainty and Bayesian model averaging are discussed and explored using an example of reliability-based design for the same limit state.