UC Berkeley

Many methodological advances in the statistics and machine learning literature are not accessible to substantive researchers and applied statisticians because of the highly technical and abstract writing and the lack of user-friendly tools for implementing the methods in practice. This dissertation providespractically feasible solutions for three challenges in applied statistics ranging from diagnosing and handling violations of the missing at random (MAR) assumption typically needed for valid statistical inference, evaluating multilevel models, and estimating standard errors for Bayesian quantile regression. All three solutions are generally applicable, even beyond the situations considered in the dissertation, all are presented in a relatively non-technical, intuitive way, and all are easy to implement with minimal modification of existing data analytic workflow and with available statistical software.