Bayesian Workflow for Geotechnical Engineering Data Analysis

Compared to disciplines such as medical science or economics where statistics has been an integral part of research and practice for long, the historical literature on statistical analysis of geotechnical engineering data is thin and simplistic. In more recent years, because of the shift towards probabilistic engineering design, including reliability-based design and load and resistance factor design, as well as the widespread and general interest in learning more from data, there has been an increase in using more advanced statistical and other data analysis methods in geotechnical engineering. Bayesian data analysis has been a particularly popular choice, because it offers a formal framework for combining information from other sources with current data in the form of prior distributions with the inherent capability for uncertainty quantification.So far, in this shift to using Bayesian methods in geotechnical data analysis, most of the effort has been devoted to either novel applications of existing models or the development of new models. Less attention has been paid to other aspects of data analysis, what is referred to as "workflow": an iterative and rigorous process of model building, model checking/evaluation, computational diagnosis, model comparison, selection and reporting.Following recent suggestions in the Bayesian data analysis literature, this paper explores workflow for geotechnical data analysis applications. The focus will be on both identifying current shortcomings in the geotechnical literature/practice and what aspects of the above-mentioned seminal workflow seem more relevant and crucial to geotechnical applications.We suggestthat adopting a tailored variation of Bayesian workflow would be straightforwardfor geotechnical engineering research and practice because its iterative nature resonates well with geotechnical engineers.

Keywords: Bayesian statistics, Statistical workflow, Rigorous data analysis.