مجله اپیدمیولوژی ایران (Dec 2017)
Studying the Performance of Multivariate Analysis of Variance (MANOVA) Models and Structural Equation Modeling on Complex Relationships between Variables
Abstract
Background and Objectives: We may sometimes measure the joint effect of correlated independent variables on several dependent variablesThe present study aimed to evaluate the performance of multivariate analysis of variance (MANOVA) and structural equation modeling (SEM) on complex relationships between variables. Methods: The present study evaluated the knowledge and attitude of 15-18 year-old individuals towards narcotics (glass, ecstasy). The effect of independent variables on two latent variables of knowledge and attitudes was studied using SEM and MANOVA modelingThe mean square error of methods were compared. Results: The direction of associations was similar in both methods but their coefficients and p-values were different. only the effect of gender (P-value= 0.007) on knowledge in both methods was significant. Nevertheless, gender (P-value < 0.001) and marital status (P-value< 0.001) were significantly associated with attitude in both methods. The mean square error of multivariate analysis of variance and structural equation modeling was 0.98 and 0.002 respectively. Conclusion: In the current studythe performance of SEM was better than MANOVA. Therefore, it is suggested that SEM to be used to study the multifactorial relationship between variables. In addition, only gender was effective on knowledge in both methods, while gender and marital status were effective on attitude in both methods.
