Statistical methods for analysis under the presence of missing data

Abstract

Missing data are a recurring problem which can cause bias or lead to inefficient analysis, no matter how well a survey questionnaire is designed and no matter how effective is the data collection. These data need a special and meticulous handling in analysis. This is why so many statistical methods have been proposed and developed to address missingness. Some of them are based on deletion of incomplete cases, others try to predict each missing value and then to include the filled in value in analysis, these are called Simple Imputation Methods. Additionally, there is another method, known as Multiple Imputation, which is based on the creation of many imputed data sets by using Data Augmentation. In this thesis, each of these methods will be mentioned. Specifically, the Multiple Imputation method will be the main topic that will monopolize the interest and will be given special emphasis. In the context of this thesis included and an application of Linear Mixed Models in repeated measurements with data that are not complete. Applying different mixed effect models on these data we reach in the appropriate model through the Bayesian Information Criterion. In continue, we apply multiple imputation in our data and then fit the same models in the imputed data this time. Our main goal is to examine the similarities or differences that may have these two data sets