Mediation Analysis in Structural Equation Modeling (Sem): Theoretical Foundations, Statistical Methods and Practical Implications

Abstract

This study offers a comprehensive investigation of mediation analysis in Structural Equation Modelling, highlighting its theoretical basics, statistical practices, and real-world applications. It differentiates mediation from moderation, explaining how mediation helps in understanding indirect relationships between latent variables. Various proposed mediation models, including simple mediation, multiple mediation, and moderated mediation, are discussed in detail. The study also analyses statistical methods such as the Causal Steps Approach (Baron & Kenny, 1986), the Product-of-Coefficients Method (Sobel Test), Bootstrapping, the Bayesian Estimation Method, and Monte Carlo Simulation, each with its respective advantages and limitations. Additionally, advanced Structural Equation Modelling techniques, such as multigroup mediation, longitudinal mediation, and latent variable mediation, are examined to address complex research scenarios. Employing a literature review-based methodology, the study synthesizes existing knowledge on best practices for estimating mediation effects using Structural Equation Modelling. Software tools like AMOS, Mplus, LISREL, and SmartPLS are discussed in the context of model specification, estimation, and evaluation. Real-world applications in business, psychology, human resource management, and marketing are illustrated, including customer trust mediating the relationship between service quality and purchase intention, employee engagement mediating the effect of transformational leadership on job performance, and social media engagement mediating brand trust and purchase intention. Key findings highlight bootstrapping as a better method for estimating indirect effects due to its non-reliance on normality of the data assumptions and Bayesian SEM as a robust substitute for handling small sample sizes and incorporating preceding knowledge. The study also discusses crucial challenges such as measurement error, model misspecification, the need for longitudinal data to establish causal inference, and comparisons between Structural Equation Modelling-based mediation and regression-based mediation using the PROCESS macro. By presenting a structured framework for mediation analysis in Structural Equation Modelling, this current study contributes to advancing causal modelling methods across various disciplines and provides directions for future research.