SmartPLS3 - Analyze, Interpret, and Report Higher Order Reflective-Formative Construct (Uptd)

Validating a higher-order reflective–formative construct in SmartPLS requires a two-stage workflow: first prove the lower-order reflective measures are sound, then test whether the higher-order formative composite (and its indicators) reliably and validly represent the construct. In the example, “internal marketing” is modeled as a higher-order construct that is formative at the second level, built from lower-level dimensions—vision development, rewards, and organizational performance—each measured reflectively. The process starts by validating every lower-order reflective construct because the higher-order model cannot be assessed directly without first establishing measurement quality at the indicator level.

For the lower-order reflective blocks, the workflow checks outer loadings, reliability, and validity. Outer loadings are expected to be acceptable (in the example, they were all “good”), followed by reliability using Cronbach’s alpha and composite reliability, both reported as above the 0.70 threshold. Convergent validity is assessed through average variance extracted–type criteria (the example uses the “over 0.5” rule for convergent validity), and discriminant validity is evaluated primarily with HTMT (all values green and below a conservative threshold of 0.85; a more liberal 0.90 cutoff is also mentioned). The Fornell–Larcker criterion is also referenced as a cross-check: within-construct variance should exceed shared variance with other constructs.

Once the lower-order constructs pass, the higher-order formative construct is validated using a disjoint two-stage approach. First, latent variable scores for the lower-order constructs are generated in stage one (the example notes a sample size of 341). These scores are exported to Excel and re-imported into SmartPLS as indicators for the higher-order construct in stage two. Because the higher-order construct is formative, the arrow directions are set so that the lower-order latent variable scores form “internal marketing.”

Higher-order convergent validity is then tested via redundancy analysis, a method associated with Chin (1998). This requires a global single-item measure capturing the overall essence of internal marketing—here, an item like “the organization has proper vision, has development initiatives and rewards its employees.” The correlation between the formative higher-order construct (internal marketing) and this global reflective measure is assessed through a path coefficient; the example reports a value of 0.8, comfortably above the 0.708 benchmark, indicating convergent validity.

Next comes collinearity diagnostics for the formative indicators using VIF values; the example reports VIFs around 2.65 and below 5, indicating no multicollinearity concerns. Then the model tests the significance and relevance of the formative “outer weights” through bootstrapping (the example uses 500 resamples for demonstration, noting 5,000–10,000 as typical). Two of the formative indicators show significant outer weights, while “rewards” is insignificant. Crucially, insignificance does not automatically mean deletion: the decision should consider outer loadings, and the example keeps the indicator because its outer loading remains above 0.5 and is significant. Finally, the relationship between internal marketing (higher-order) and the dependent variable (organizational performance) is assessed using bootstrapped path coefficients, with internal marketing showing a significant influence.

In short: validate reflective lower-order blocks first (loadings, reliability, convergent and discriminant validity), then validate the formative higher-order construct with redundancy analysis, VIF checks, bootstrapped outer weights, and outer loadings—before interpreting the structural link to organizational performance.