#SmartPLS4 Series 17 - How to Report Reflective Reflective Higher Order Construct

Reflective higher-order constructs in PLS-SEM should be reported using the same reliability and validity checks used for reflective lower-order constructs—then summarized with a small number of tables rather than separate sections for every statistic. In this session, the focus is on a reflective–reflective higher-order construct called Internal Service Quality (ISQ), built from four reflective lower-order dimensions: Reliability, Assurance, Empathy, and Responsiveness.

After the model is re-estimated with the lower-order dimensions treated as indicators at the higher level, the reporting process starts with the measurement results for the higher-order construct. The outer loadings (reported as factor loadings) are checked first; they should look acceptable before moving on. Next come the reliability and convergent validity metrics for the higher-order construct itself—reported in the same way as for lower-order constructs. The session emphasizes that ISQ’s higher-order validity is established by assessing factor loadings, reliability, and convergent validity using the same logic applied earlier in the series.

For reliability, the session points to the standard PLS-SEM outputs: Cronbach’s alpha (α) and composite reliability (CR). For convergent validity, it uses the Average Variance Extracted (AVE). These values are pulled directly from the PLS algorithm output and formatted into a single summary table.

Discriminant validity is then reported using two common criteria: the HTMT ratio and the Fornell–Larcker criterion. The key reporting idea is to compare the higher-order construct (ISQ) against other constructs in the model, not to re-validate the lower-order constructs again. In the example workflow, the HTMT results are copied into a “Table 9” style discriminant validity table, and the Fornell–Larcker and HTMT tables are formatted for readability. The session highlights the interpretation: within-construct variance for ISQ is higher than shared variance with other constructs, which supports discriminant validity under the Fornell–Larcker criterion.

A practical takeaway is how to present results efficiently. For lower-order constructs, the series used separate reporting structures (factor loadings, indicator multicollinearity, reliability, convergent validity, and discriminant validity). At the higher-order level, the session recommends consolidating these into one section and using one or two tables to summarize the key statistics, since most estimates repeat the same measurement-model logic and only the higher-order construct changes.

Overall, the reporting template for a reflective–reflective higher-order construct boils down to: identify the higher-order construct and its subdimensions, report outer loadings, report reliability (α and CR) and convergent validity (AVE), then report discriminant validity (HTMT and Fornell–Larcker). The session’s workflow is designed to keep the write-up consistent with earlier lower-order reporting while avoiding unnecessary duplication of headings and tables.