What is a higher-order construct?

A construct measured not directly by items but by several lower-order constructs, each with its own items. It models an abstract concept through more concrete sub-dimensions and adds a second layer to the measurement model.

Which estimation approach should I use?

The two-stage approach, and in particular the extended two-stage approach, is robust and flexible, and works with reflective and formative levels. Repeated indicators are simpler but can bias results when indicator counts are uneven.

How do I validate a higher-order construct?

Validate each level according to its type: reflective levels with reliability and discriminant-validity checks, formative levels with weight significance and collinearity checks. Always validate the lower-order constructs before the level above.

PLS-SEM METHOD

Higher-order constructs in PLS-SEM

A higher-order construct captures an abstract concept through several more concrete sub-constructs, adding a second layer to the measurement model. This page explains the types, the estimation approaches, how to validate them, a worked example, and the common pitfalls.

What a higher-order construct is

Most constructs are measured directly by their indicators. A higher-order construct (also called a hierarchical component model) sits one level up: it is measured by several lower-order constructs, each of which has its own indicators. For example, job resources might be a higher-order construct made up of autonomy, support and feedback.

Higher-order constructs reduce the number of structural paths and model an abstract concept at the right level of generality, as long as the extra layer is theoretically justified.

A worked example

Job resources is modeled as a higher-order construct with three lower-order components: autonomy, support and feedback, each measured by its own items. The higher-order construct then enters the structural model as a single driver, instead of three separate paths.

This keeps the structural model simple while preserving the detail of the three components underneath.

A higher-order construct (job resources) built from three lower-order components, each with its own indicators.

The four types

Higher-order constructs are classified by whether each level is reflective or formative. The type determines how you specify and validate the model.

Reflective-reflective: Both levels reflective: the higher-order construct reflects correlated lower-order components, each reflected by its items.
Reflective-formative: Reflective lower-order constructs form the higher-order construct. The most common type in practice.
Formative-reflective: The higher-order construct reflects formatively measured lower-order constructs.
Formative-formative: Both levels formative: formative components form the higher-order construct.

Pick the type from theory and the direction of the indicator relationships, before estimation. The type dictates which validity checks apply at each level.

Estimation approaches

Repeated indicators

The higher-order construct is measured by all the indicators of its lower-order components at once. Simple, but it can create bias when the number of indicators is uneven.

Two-stage

Stage one estimates the lower-order construct scores; stage two uses those scores as indicators of the higher-order construct. Robust and flexible.

Extended two-stage

A two-stage variant that carries the latent scores forward, which OpenPLS supports by adding latent-variable scores to a dataset so you can build the next level on them.

Disjoint two-stage

A two-stage variant that keeps the higher-order and lower-order parts in separate models across the two stages.

Building a higher-order construct step by step

Decide from theory which lower-order constructs make up the higher-order construct, and the reflective or formative type at each level.
Estimate the lower-order constructs and validate them like any construct (reliability and validity at that level).
Add the latent-variable scores to your dataset so they can serve as indicators of the higher-order construct.
Build the higher-order construct on those scores and estimate the structural model (the extended two-stage approach).
Validate the higher-order level according to its type, then interpret the structural paths.

Common pitfalls

Adding a higher-order layer without theoretical justification, just to simplify the diagram.
Using repeated indicators when the lower-order components have very unequal numbers of indicators, which biases the weights.
Applying reflective validity checks (like HTMT) to a formative level, or vice versa. The type decides which checks apply.
Forgetting to validate the lower-order constructs before building the level above them.

Higher-order constructs in OpenPLS

OpenPLS supports higher-order constructs. A dedicated helper sets up the hierarchy, and the Add latent-variable scores to a dataset action lets you carry scores forward for the extended two-stage approach, which also makes third-order models possible. Free and runs in the browser.

Build a higher-order construct in your own model

Model the lower-order constructs in the visual editor, carry their scores forward, and build the level above in a few clicks. Free, no install.

Open OpenPLS What is PLS-SEM?

Frequently asked questions

What is a higher-order construct?: A construct measured not directly by items but by several lower-order constructs, each with its own items. It models an abstract concept through more concrete sub-dimensions and adds a second layer to the measurement model.
Which estimation approach should I use?: The two-stage approach, and in particular the extended two-stage approach, is robust and flexible, and works with reflective and formative levels. Repeated indicators are simpler but can bias results when indicator counts are uneven.
How do I validate a higher-order construct?: Validate each level according to its type: reflective levels with reliability and discriminant-validity checks, formative levels with weight significance and collinearity checks. Always validate the lower-order constructs before the level above.