Reflective-Reflective HOC using #SmartPLS4: Reliability, Validity, and Hypotheses Testing

Reflective–reflective higher order constructs in SmartPLS4 can be handled cleanly by validating the lower-order dimensions first, then treating those validated dimensions as “items” for the higher-order construct, and only after that running hypothesis tests. The practical payoff is that reliability, convergent validity, and discriminant validity get checked at both levels—so any mediation or structural claims rest on measurement quality rather than assumptions.

The workflow starts with a model where an independent variable is a higher order construct built from multiple reflective dimensions. In the example, “Internal Service Quality” is measured through four reflective dimensions—reliability, assurance, empathy, and responsiveness—and it influences “Organizational Performance,” with “Organizational Commitment” (or collaborative culture) acting as a mediator. Instead of combining the four dimensions into a single block immediately, each dimension is first modeled as its own lower-order construct and validated on its own.

After the lower-order constructs are specified, the next step is to run the PLS algorithm and inspect outer loadings, reliability, and validity. Items are not deleted if reliability and convergent validity are acceptable (reliability above 0.70 and convergent validity above 0.50 in the example). Discriminant validity is also checked; values slightly below 0.90 are treated as acceptable when the constructs belong to the same higher-order structure. Once all four lower-order dimensions meet the criteria, the analysis moves up one level.

To validate the higher-order construct, SmartPLS4 is used to create a higher-order reflective–reflective construct by exporting scores (creating a new model/duplicate) and then re-specifying the measurement model. The four validated lower-order constructs are removed from the structural links and reintroduced as the indicators of the higher-order construct “Internal Service Quality.” In this reflective–reflective setup, the arrows point in the correct direction for a reflective–reflective specification, and the PLS algorithm is run again.

At this stage, the higher-order construct’s measurement quality is assessed using the same reliability and validity logic. Because the lower-order constructs were already validated, the analysis focuses reporting on the higher-order construct’s reliability, convergent validity, and discriminant validity rather than re-listing every lower-order result.

With measurement quality confirmed at both levels, hypothesis testing proceeds via bootstrapping. The example uses 10,000 bootstrap samples with bias-corrected confidence intervals (One-Tailed, since the direction of relationships is known). Path coefficients are checked for significance, and mediation is evaluated through indirect effects. Significant path coefficients and significant indirect effects indicate that the hypothesized relationships—including mediation—are supported. The result is a step-by-step, measurement-first approach for reflective–reflective higher order constructs in SmartPLS4 that culminates in reliable hypothesis testing.