28. SEMinR Series - Higher Order Construct Analysis - Reflective-Formative

Higher-order SEMinR modeling hinges on treating each construct type differently: reflective–reflective higher-order blocks get reliability/validity checks at the second-order level, while reflective–formative higher-order blocks require formative diagnostics (VIF, outer weights, and outer loadings) rather than the usual reflective criteria. In this walkthrough, internal marketing (IM) is modeled as a reflective–reflective higher-order construct, internal service quality (ISQ) as a reflective–formative higher-order construct, and organizational performance (OP) as a lower-order reflective construct—then the model is tested with mediation, using ISQ as the mediator between IM and OP.

The process starts by validating every lower-order reflective construct, regardless of whether it later feeds into a reflective–reflective or reflective–formative higher-order model. The lower-order reflective pieces are estimated first for reliability and validity using composite reliability (CR), Cronbach’s alpha, convergent validity, and discriminant validity. In the example, IM’s reflective dimensions—vision development and rewards—are validated, and ISQ’s reflective dimensions—assurance, reliability, empathy, and responsiveness—are also validated at the lower-order level. OP is validated as a separate lower-order reflective construct as well.

With the measurement models in place, the structural model is built by linking the independent variable (IM) to the mediator (ISQ) and then linking both IM and ISQ to the dependent variable (OP). The same path pattern is applied at the dimension level: IM dimensions connect to OP and to ISQ, while ISQ dimensions connect to OP. After estimating the model, the output is checked through reflective-model evaluation outputs such as loadings, reliability, HTMT, and cross-loadings.

Next comes the second-order assessment. For the reflective–reflective higher-order construct IM, reliability and validity are assessed at the higher-order level alongside its lower-order reflective components. Because IM is reflective–reflective, it uses the default mode (no explicit weights are needed). For ISQ, the higher-order construct is reflective–formative: the arrows run from the four reflective dimensions (assurance, reliability, empathy, responsiveness) into the formative higher-order composite ISQ, so weights matter. To specify ISQ as higher-order formative in SEMinR, the model requires setting the appropriate mode (via `wids mode SC B`).

ISQ’s validation follows formative rules. Reporting of reflective validity metrics for ISQ is skipped; instead, diagnostics are used: VIF values for the formative indicators must be below the threshold (VIF < 5), outer weights are checked for significance (some indicators can be insignificant), and outer loadings must exceed 0.50 and be significant to confirm indicator contribution. Bootstrapping then supports hypothesis testing and mediation. The mediation results show a significant specific indirect effect from IM to OP through ISQ, with t statistics exceeding 1.96 and confidence intervals that do not cross zero. The workflow ends with path coefficients, confidence intervals, and R-square values for the endogenous constructs, producing a complete reflective–reflective and reflective–formative higher-order SEMinR model ready for reporting.