Impact of inconsistent imputation models in mediation analysis
Ye, Bo · 2021
In this dissertation, we study the impact of inconsistent imputation methods in mediation analysis and its application.We present the study in three papers.In the first paper, we investigate the impact of possibly incoherent imputation models on mediation analyses.In particular, we discuss commonly used joint modelling approach as well as variable-by-variable approach when the ultimate analytical goal in mediation analysis.Practical advantages of each approach along with the discussion on coherence of the imputation model is our focal point in our manuscript.A comprehensive simulation study is summarized to gauge the performance of widely utilized imputation models.In the second paper, we study the similar topic but in multi-level data.The choice of imputation model is often driven by the focus of the post imputation models.However, this is not necessarily the case in all applications of multiple imputation (MI) inference, especially when MI is used in large-scale surveys.Our work assesses the impact of an imputation model on the mediation analysis with clustered data.Specifically, we consider joint and variable-by-variable imputation models leading up to multi-level mediation analysis.We provide theoretical and analytical assessment of the bias under each imputation method.A comprehensive simulation study is conducted to under-stand the performance of imputation methods in a repetitive sampling framework.In the third paper, we study whether environmental programs affect student academic performance.We examined if the association between EPA Tools for School (TfS) policies or other environmental programs and student test scores were mediated by student attendance.We linked the 2015 School Building Condition Survey (BCS) with School Report data provided by the New