33. SEMinR Lecture Series -Moderation Analysis with Categorical Moderator

Moderation analysis in SEMinR can handle categorical moderators by treating the category (e.g., bank type) as a single-item construct and then explicitly modeling the interaction between that category and the predictor(s). The practical takeaway is that once the measurement model and structural paths are set up, the moderation effect is tested through PLS estimation followed by bootstrapping, with significance judged using t statistics (threshold 1.96) and confidence intervals.

The workflow starts with the usual SEMinR setup: load the dataset, verify the data, then build a measurement model. In this session, the continuous constructs are defined using multiple indicators (e.g., OC1–OC8 for organizational commitment). A categorical moderator—bank type—is represented as a single item named as a construct in the model (private sector bank coded as 1, public sector bank coded as 0). With the measurement model in place, the key step is creating an interaction term for moderation. The interaction is formed using the categorical moderator construct (private) together with the predictor construct(s), using SEMinR’s interaction function and specifying the two-stage interaction effect.

Next comes the structural model, where the moderation paths are specified. The model includes direct effects and the interaction effect: the continuous predictor (OC) and the moderator (private) are linked to the outcome (CC, collaborative culture), and the interaction term (OC into private) is also linked to CC. The same pattern is repeated for the other predictor (POS, perceived organizational support), producing a second interaction term (POS into private). In the SEMinR model notation shown, the interaction terms are written in the structural relationships as “OC into private” and “POS into private,” with the moderator referenced by the construct name used in the model.

After defining the structural relationships, the analysis estimates the PLS model and stores results in a summary object. Bootstrapping is then run to test whether the moderation effects are statistically significant. The results show both interaction effects are significant because their t statistics exceed 1.96. The sign of the interaction coefficients matters for interpretation: a negative coefficient indicates the moderator reverses or changes the strength of the relationship in a way that differs between public and private sector banks. Concretely, the negative sign on the OC×private interaction is interpreted as public sector banks having the stronger POS/OC-to-CC relationship in one direction, while private sector banks show a stronger OC-to-CC relationship in the slope interpretation.

To make the moderation effect interpretable, the session performs slope analysis using an external Excel approach attributed to Jeremy Dawson for binary moderators. Values are extracted from the SEMinR output (including the path coefficient for OC→CC and the interaction coefficient for OC into private). The slope for each moderator group is then plotted: the steeper gradient corresponds to the group where the predictor has the stronger effect on collaborative culture. The same slope logic is applied to POS→CC moderation, again using the interaction coefficient and comparing gradients across public (0) and private (1) sector banks.

Finally, the session confirms significance for both moderation effects using t statistics and confidence intervals that do not cross zero. The overall conclusion is that bank type (categorical moderator) significantly moderates how organizational commitment and perceived organizational support relate to collaborative culture, with the direction and strength differing between public and private sector banks.