CB-SEM using #SmartPLS4 - 2 - An Introduction to Structural Equation Modeling (SEM)

Structural equation modeling (SEM) is positioned as the right tool for testing complex causal frameworks—especially when multiple mediators, indirect effects, and higher-order constructs appear in the same model—because it can evaluate measurement quality and structural relationships simultaneously. The session frames a common research scenario: antecedent variables influence “KM processes” and “organizational performance” indirectly through “creative organizational learning.” That kind of multi-path, multi-construct setup can be difficult to handle in SPSS, particularly when the design requires more than one mediator or serial/compound mediation within a single integrated model.

With SmartPLS 4, the workflow expands beyond variance-based SEM by adding an option for covariance-based SEM. The series outline lays out a progression from basics to advanced applications: building measurement and structural models, checking reliability and validity, improving model fit, and then moving into mediation, moderation, and higher-order construct analysis. Before any modeling begins, the session emphasizes that constructs must be measured with multiple questionnaire items—typically at least four to six per construct—because SEM relies on indicators to represent latent (unobserved) variables, and item deletion can occur during assessment.

SEM itself is defined as a family of techniques that examines relationships among many variables at once, using a confirmatory approach. Rather than running separate regressions, SEM tests whether observed data provide evidence for hypothesized directions and significances within a proposed theoretical framework. The method is described as combining elements of factor analysis (to assess measurement properties like reliability and validity) with regression-like modeling (to test how predictors influence outcomes). A key advantage highlighted is the ability to model multiple independent and dependent variables, incorporate error terms, and handle constructs that can act as predictors, mediators, moderators, or even controls within the same system.

The modeling process is split into two core sub-models. First comes the measurement model, assessed through confirmatory factor analysis (CFA), which checks whether items (indicators) properly measure their intended latent constructs and whether constructs are distinct enough to show discriminant validity. Second comes the structural model, assessed by testing the significance of hypothesized paths—often using t statistics to determine whether relationships exceed a threshold for statistical significance. In this framework, constructs are represented as latent variables (circles/ovals), questionnaire items as manifest variables (rectangles), and arrows indicate directional influence.

The session closes by clarifying SEM terminology—exogenous versus endogenous latent variables, factor loadings as the strength of association between constructs and indicators, and t statistics as the criterion for path significance—then sets expectations for subsequent theoretical and practical sessions focused on using SmartPLS for covariance-based SEM.