2. Reflective-Reflective Higher Order Construct/Second Order Analysis and Reporting in SmartPLS

Reflective–reflective higher order constructs (HOCs) in SmartPLS are treated as a workable, literature-supported configuration—even though critics argue they may be conceptually “meaningless.” In this setup, a higher order construct (HOC) is measured reflectively by indicators that are themselves the indicators of multiple lower order constructs (LOCs). The practical implication is that researchers can model a single overarching concept (the HOC) while still retaining distinct subdimensions (the LOCs), then assess measurement quality and structural relationships in a disciplined two-stage workflow.

The core specification relies on the repeated indicators approach. All items belonging to each reflective lower order construct are simultaneously assigned to the reflective measurement model of the higher order construct. In SmartPLS terms, indicators from the LOCs are “loaded onto” the HOC, with arrows reflecting the reflective–reflective logic: the HOC points to its indicators (which are the LOC items), and the LOCs are treated as subdimensions contributing to the HOC. For structural modeling, antecedent constructs connect directly to the HOC rather than to the LOCs, and the HOC connects directly to the criterion variable.

Because reflective–reflective HOCs are debated, the workflow emphasizes measurement validity rather than assuming it. Quality assessment starts with the LOCs: reliability and convergent validity are checked using standard reflective criteria such as Cronbach’s alpha, composite reliability (CR), and average variance extracted (AVE), alongside discriminant validity checks (including HTMT). After LOCs pass these checks, the analysis moves to the HOC itself.

SmartPLS estimation for reflective–reflective HOCs uses the disjoint two-stage approach. Stage 1 estimates a model containing only the LOCs (including exogenous and endogenous constructs) without specifying the HOC. Stage 2 then uses latent variable scores (LVS) generated from Stage 1 as the indicators for the HOC, while the LOCs remain as constructs in the structural model. This separation is crucial: it ensures the HOC’s measurement properties are evaluated using the LVS-based representation rather than mixing estimation steps.

For reporting, the transcript lays out a results chapter structure: introduce the analysis technique and hypotheses, report the measurement model, then report structural model outcomes. Measurement reporting begins with factor (outer) loadings, then multicollinearity via VIF (acceptable when below 5), followed by reliability (Cronbach’s alpha and CR) and convergent validity (AVE, with the guidance that CR above 0.70 can support validity even when AVE is slightly low). Discriminant validity is reported using Fornell–Larcker criterion, cross-loadings, and/or HTMT (with HTMT used as a key check).

Finally, structural evaluation uses path coefficients and significance testing via bootstrapping (with 5,000 recommended). In the worked example, the higher order construct CSR shows a significant positive effect on organizational performance (organizational performance as the criterion). Mediation through team identity is assessed via indirect effects: the transcript notes that mediation depends on p-values (e.g., no mediation when p > 0.05, but partial mediation can appear under a looser significance threshold such as 0.10). The overall takeaway is a repeatable method: validate LOCs first, generate LVS for the HOC through disjoint two-stage estimation, then report measurement and structural results consistently in SmartPLS.