Bayesian Networks and analysis with incomplete data
Stephanie Ranft · 2020
A statistical pipeline for the analysis of incomplete data is theorised and applied, with missing data entries substituted for using multiple imputation by chained equations (MICE). Bayesian Networks are constructed for the purposes of multivariate analysis, and providing insight into conditional (in)dependencies between variables. Partial correlation is used predominantly, as both an investigative and diagnostic tool. The theories explore the merits of frequentists and Bayesian approaches, with the ensuing application conducted in a Bayesian framework.