Predictive Quantitative Modelling Through SEM Approach

Structural Equation Modelling (SEM) as a two-step process combining Confirmatory Factor Analysis (CFA), used to validate the measurement model, with the evaluation of hypothesised structural relationships. He introduced key terminology, including latent constructs, manifest or observed variables, exogenous and endogenous variables, measurement error and residuals, and described SEM as a combination of exploratory factor analysis and multiple regression that subsumes related techniques such as path analysis, causal modelling and confirmatory factor analysis. He illustrated basic relationship types, including direct, indirect, spurious, correlational and moderating effects, and set out a step-by-step guideline for conducting SEM, from reviewing theory and specifying the model, through data collection and exploratory factor analysis, to confirmatory factor analysis and testing the structural model, including checks for multivariate normality, outliers, linearity, multicollinearity and homoscedasticity, and testing for mediation and moderation, before reporting findings. On the measurement side, he covered the distinction between exploratory and confirmatory factor analysis, model identification (under-identified, just-identified and over-identified models), classical test theory's decomposition of an observed score into a true score and error, and the difference between formative and reflective constructs. He discussed construct validity in detail, covering face validity, convergent validity (via Average Variance Extracted, expected above 0.5, and factor loadings), construct reliability (expected above 0.7), and discriminant validity via the Fornell-Larcker criterion and the newer HTMT approach, and recommended sample-size ratios of at least 1:10, and ideally 1:20, respondents per estimated parameter. He closed by explaining how to assess overall model fit using goodness-of-fit indices such as GFI, AGFI, NFI and CFI (each expected to exceed 0.9), RMSEA (expected below 0.08), and comparative indices such as AIC and BIC, while cautioning that SEM results can be unstable with samples smaller than 200 observations.