#SmartPLS4 Series 11 - How to Design a Measurement Model with Lower Order Constructs?

TL;DR

Start by adding and assessing every lower-order construct (including moderators and directly measured dependent variables) before analyzing higher-order constructs.

Briefing Cornell Notes

Briefing

Designing a SmartPLS measurement model that includes both higher-order and lower-order constructs starts with a simple rule: every lower-order construct must be assessed first, even when higher-order constructs will be analyzed later. In the example model, “Internal Marketing” is treated as a higher-order construct built from three lower-order dimensions—Vision Development and Rewards—while “Internal Service Quality” is another higher-order construct composed of four lower-order subdimensions: empathy, responsiveness, assurance, and reliability. By contrast, the two moderators (R ambiguity and role conflict) and the endogenous dependent variable are modeled as lower-order constructs measured directly by items, without additional dimensions.

The workflow begins by creating a PLS-SEM model canvas and placing all lower-order constructs as latent variables with their indicators. In practice, that means adding the indicators for Vision, Development, and Rewards; adding the indicators for empathy, responsiveness, assurance, and reliability; and also adding the indicators for the two moderators. The dependent variable is added as well, along with the mediators (the endogenous constructs in the study). Even though moderators will later be used to test interaction effects, they still must be included in the model at this stage because they are part of the measurement structure.

Next comes the structural linking needed to run the model algorithm. Internal marketing’s higher-order structure is handled by linking each independent dimension (the lower-order components) to the mediators and to the dependent variable. The model then links the mediators to the dependent variable as standard paths. For moderation, the two moderator constructs are connected to the endogenous variable in the specific relationship they moderate—here, R ambiguity and role conflict are set up as moderators between culture (CC) and the dependent variable (OP). Once these relationships are drawn, the model is ready for the first measurement-model check.

That first check is performed by running “PLS algorithm,” which produces path coefficients and—crucially for measurement-model evaluation—factor loadings for each lower-order construct. The transcript emphasizes that measurement-model assessment starts with loadings: factor loadings are inspected by selecting a latent variable and viewing its indicator values. This step is treated as the gateway to the next phase, where reliability and validity will be evaluated for the lower-order constructs before moving on to the higher-order measurement structure in later analysis.

Cornell Notes

The model-building process for SmartPLS with higher-order constructs begins by treating every lower-order construct as a measurement target first. In the example, Internal Marketing and Internal Service Quality are higher-order constructs built from multiple lower-order dimensions (e.g., empathy/responsiveness/assurance/reliability for Internal Service Quality), while moderators like R ambiguity and role conflict are added as lower-order constructs measured directly by items. After placing all lower-order constructs and indicators on the canvas, the next step is linking paths: dimensions connect to mediators and the dependent variable, and moderators are connected to the specific relationship they moderate (culture to OP). Running the PLS algorithm then yields factor loadings, which become the first measurement-model criterion to inspect before reliability and validity checks.

Why must lower-order constructs be analyzed before higher-order constructs in SmartPLS measurement modeling?

Lower-order constructs are the measurable building blocks. Even when higher-order constructs will be tested later, the workflow requires running the measurement-model assessment for all lower-order constructs first. In the example, Internal Marketing and Internal Service Quality are higher-order constructs, but their components (Vision Development and Rewards; empathy, responsiveness, assurance, reliability) are treated as lower-order constructs whose factor loadings must be checked first. The same principle applies to moderators and the dependent variable when they are measured directly by items.

How does the model represent a higher-order construct like Internal Service Quality?

Internal Service Quality is modeled as a higher-order construct with four lower-order subdimensions: empathy, responsiveness, assurance, and reliability. Those subdimensions are added as separate latent variables with their own indicators during the lower-order measurement-model stage. The higher-order structure is then handled later, after the lower-order constructs have been assessed for measurement quality.

What is the first measurement-model criterion checked after running the PLS algorithm?

Factor loadings. After calculating the PLS algorithm, the output includes path coefficients and factor loadings. The transcript highlights that measurement-model assessment starts with loadings: selecting a latent variable reveals the factor loading values for its indicators (e.g., by clicking the latent variable and viewing the indicator-level loading list).

How are moderators handled during the initial lower-order measurement-model stage?

Moderators are included as lower-order constructs with their own indicators during the initial stage. The moderation effect itself is not tested yet as an interaction; instead, the moderators are placed in the model so the measurement part can be assessed. Later, they are connected to the endogenous variable in the specific relationship they moderate (in the example, R ambiguity and role conflict moderate the link between culture (CC) and OP).

What structural links are created before measurement-model evaluation in this example?

Internal marketing’s dimensions are linked to the mediators and the dependent variable. Mediators are linked to the dependent variable as well. For moderation, the two moderator constructs are connected to the endogenous variable for the relationship they moderate—explicitly described as moderating between CC and OP. Once these paths are drawn, the PLS algorithm can be run to inspect factor loadings.

Review Questions

In what order should lower-order and higher-order constructs be assessed, and what measurement quality step is performed first?
How does the example connect moderators to the model before running the PLS algorithm?
Which measurement-model output is inspected first, and how is it accessed for a specific latent variable?

Key Points

1
Start by adding and assessing every lower-order construct (including moderators and directly measured dependent variables) before analyzing higher-order constructs.
2
Model higher-order constructs by decomposing them into their lower-order dimensions with their own indicators (e.g., Internal Service Quality from empathy, responsiveness, assurance, reliability).
3
Include moderators as lower-order constructs in the measurement stage, even if their interaction effect will be tested later.
4
Link lower-order dimensions to mediators and the dependent variable, then connect moderators to the specific path they moderate (e.g., CC → OP).
5
Run the PLS algorithm to generate factor loadings and use factor loadings as the first measurement-model criterion to inspect.
6
After loadings are checked, proceed to reliability and validity assessment for the lower-order constructs before moving deeper into the higher-order measurement structure.

Highlights

The workflow insists on assessing lower-order constructs first, even when higher-order constructs are part of the final model.

Internal Service Quality is treated as higher-order, built from four lower-order subdimensions: empathy, responsiveness, assurance, and reliability.

R ambiguity and role conflict are added as lower-order constructs with indicators, then connected to the specific relationship they moderate (culture to OP).

Factor loadings are the first measurement-model output inspected after running the PLS algorithm.

The model links internal marketing dimensions to mediators and the dependent variable before measurement evaluation begins.

Topics

Mentioned

PLS-SEM
PLS
DV
OP
CC
OP