Comparison between multiple regression and structural equation modelling in identifying influential factors in academic performance

• Main Author: Dg Siti Nurisya Sahirah Ag Isha
• Format: Thesis
• Language: English; English
• Published: 2024
• Subjects: QA299.6-433 Analysis
• Online Access: https://eprints.ums.edu.my/id/eprint/40801/1/24%20PAGES.pdf; https://eprints.ums.edu.my/id/eprint/40801/2/FULLTEXT.pdf; https://eprints.ums.edu.my/id/eprint/40801/
• Institution: Universiti Malaysia Sabah

Multiple regression (MR) and structural equation modelling (SEM) are statistical techniques frequently used in various fields. Despite the popularity of both methods, limited studies have discussed and highlighted the modelling process of MR and SEM in detail, including their underlying assumptions and procedural steps, as well as comparing the findings for both statistical analyses, especially in the education context. Therefore, this study is conducted to address this gap by presenting a clear and detailed procedure for both MR and SEM. Besides that, this study compares the findings of both methods in identifying the significant factors in students' performance and examining the role of academic motivation as a mediator. A total of 533 undergraduate students from Universiti Malaysia Sabah participated in this study and selected through stratified random sampling. Perception of academic achievement, grade point average (GPA), and cumulative grade point average (CGPA) are used to measure academic achievement. Five factors are included in the model as the independent variables: personal, psychological, demographic, socioeconomic status, and institutional. This study adopted the standard instruments to measure personal, psychological, and institutional factors such as the Big Five Inventory Personality Traits, Rosenberg Self-Esteem, Vallerand Academic Motivation, Schutte Self-Report Emotional Intelligence, Eysenck General Intelligence, and Course Experience Questionnaire. Three types of analyses are employed to identify significant factors: MR, SEM with composite variables (SEMc), and SEM with measurement indicators (SEMm). The findings reveal that MR and SEMc yield similar findings in terms of significant factors identified and values of coefficient of determination (R2), standardized beta coefficient (β), and standard error. In contrast, SEMm obtained less significant factors as compared to MR and SEMc, but the values of coefficient of determination (R2), standardized beta coefficient (β), and standard error are higher in SEMm. In conclusion, this study suggests that MR is preferable to SEM in identifying significant factors when using composite variables. However, SEM is superior to MR in assessing mediation effects since it can examine the influence of each variable in the model simultaneously.