1. The Concept of Higher Order Constructs (HOC)/Second Order/Hierarchical Components Models (HCM)

Higher order constructs—also called hierarchical component models—in PLS provide a way to measure a broad, abstract concept using multiple more specific subdimensions, while keeping the structural model simpler. Instead of linking every lower-order dimension directly to an outcome, researchers can connect only the higher-order construct to the dependent variable. In the example given, CSR is the higher-order construct, built from four lower-order dimensions (economic, legal, ethical, and discretionary). The model then tests how CSR relates to organizational performance, without forcing separate paths from each subdimension to that performance outcome.

This approach brings several practical advantages. First, it reduces the number of path relationships, improving model parsimony—an important hallmark of scientific modeling. If the research goal is the overall effect of CSR on organizational performance, then making separate links from each CSR dimension to performance becomes unnecessary. By collapsing those relationships into a single higher-order path (CSR → organizational performance), the model becomes leaner while still representing the underlying multidimensional structure.

Second, higher order constructs help address the “bandwidth–fidelity dilemma.” Bandwidth reflects the variety and complexity of information, while fidelity reflects how thorough and accurate the testing is. When each lower-order dimension is linked to the outcome, the model can generate more variety but may not deliver the focused thoroughness needed for the specific research question. Framing the analysis around the higher-order construct aims to keep the information targeted: variety is minimized to what matters for the higher-level concept, while the relationship being tested remains precise (CSR as a whole linked to organizational performance).

Third, higher order constructs can reduce collinearity among formative indicators. By reorganizing indicators into a higher-order structure, researchers can rearrange how formative components contribute across subdimensions, which can lessen redundancy and stabilize estimation.

Using higher order constructs also requires careful decisions before analysis. The higher-order construct must be grounded in well-developed measurement theory, and researchers must determine whether the lower-order components and the higher-order construct are reflective or formative—choices that depend on conceptualization and should align with theory-specific literature. Four types are described based on combinations of reflective and formative at the first and second-order levels (e.g., reflective–reflective, reflective–formative, formative–reflective, formative–formative). A key implication of reflective versus formative directionality is whether the higher-order construct would still exist if one lower-order dimension were removed.

Methodologically, two prominent estimation strategies are highlighted: the repeated indicators approach and the two-stage approach (with two-stage being the usual choice). The transcript notes that simulation work (citing Becca, 2012, and coauthors) finds extended repeated indicators can reduce bias in estimating the measurement model linking lower to higher constructs for reflective–formative types, while two-stage approaches can improve parameter recovery for paths from exogenous constructs to the higher-order construct and from the higher-order construct to endogenous outcomes. Performance differences tend to shrink when sample sizes are large.

Finally, several common mistakes are flagged: failing to assess measurement quality for higher-order constructs (not just lower-order components), misreading higher–lower relationships as structural paths rather than measurement relationships, and checking discriminant validity only for lower-order constructs while neglecting discriminant validity at the higher-order level. The session ends by pointing to upcoming guidance on PLS implementation and validation, alongside references including a paper in the Australasian Marketing Journal by Saiset and others.