Introduction to structural equation modeling and Bayesian networks in statistics

Abstract

This thesis focuses on analyzing Structural Equation Modeling (SEM) and Bayesian Networks from a theoretical and practical approach. In the first chapter, basic knowledge and concepts about statistics, data and data types are introduced. This chapter is dedicated to familiarize readers who are beginners with the scientific field of data science. In the second chapter, the remedy and components of a SEM analysis are explicitly analyzed along with R, the computer program used to apply SEM and Bayesian Networks with real world data. Essential SEM parts such as correlation, covariance, Principal Component Analysis (PCA), latent variables, path analysis and SEM graph theory are introduced in the specific chapter of the paper. In the third chapter, the concept of Bayesian Statistics and Networks is analyzed. In the beginning of this chapter, the theoretical justification of critical properties of Bayesian Networks such as maps, d-separation Markov blanket, exact inference, approximate inference and algorithms which conduct the functions of a Bayesian Network are introduced. In the end of the second and third chapter, corresponding R labs for SEM and Bayesian Networks, respectively, demonstrate the usage of the two techniques with a real world example. During the R labs, each step and result is accompanied by the corresponding part of code which produces it. The input dataset of the R labs consists of information from 1470 employees of a company regarding demographic and company-related data. The R source code of this thesis is available on Github and the link is provided to the extended abstract which is located at page 7 and 8 of the paper.