Concept of Higher Order Constructs

Concept of Higher Order Constructs

Higher-order constructs (also known as hierarchical component models in the context of PLS-SEM) provide a framework for researchers to model a construct on a more abstract dimension (referred to as higher-order component) and its more concrete subdimensions (referred to as lower-order components).

HCM refers to a more general construct that is measured at a higher level of abstraction, while simultaneously assessing several sub-components (dimensions). Hence, by specifying lower-order components, HCM cover concrete traits of a more general conceptual variable of interest (Hair et al. 2018).

Advantages

Higher-order constructs help to reduce the number of path model relationships, thereby achieving model parsimony. Instead of specifying relationships between multiple independent and dependent constructs in a path model. 

Researchers can summarize the independent constructs in a higher-order construct, making the relationships from the (then) lower-order components to the dependent constructs in the model obsolete.

Higher-order constructs also help to overcome the bandwidth-fidelity dilemma. According to which there is a tradeoff “between variety of information (bandwidth) and thoroughness of testing to obtain more certain information (fidelity).”

Typically, bandwidth is used as a synonymous of complexity and variability. The opposite of bandwidth is narrowness. On the other hand, fidelity is a synonymous of accuracy and specificity. The opposite of fidelity is impreciseness.

Finally, higher-order constructs provide a means for reducing collinearity among formative indicators by offering a vehicle to re-arrange the indicators and/or constructs across different concrete sub-dimensions of the more abstract construct.

Types of Higher Order Constructs

To reap the benefits of higher order constructs, researchers must address at least three concerns.

First, the higher-order construct’s conceptualization and specification needs to be grounded in well-developed measurement theory. In fact, this step can be as challenging and tedious as developing a new measurement scale.

Specifically, when implementing a higher-order construct, researchers have to decide on

  1. The measurement model specification of the lower-order components, and
  2. The relationship between the higher-order component and its lower-order components, both of which can be reflective or formative in nature.

As a result, research has proposed four types of higher-order constructs (see Figure): reflective-reflective, reflective-formative, formative-reflective, and formative-formative.

Prior studies on higher-order constructs in PLS-SEM have shown that reflective-reflective and reflective-formative higher- order types feature prominently in different fields.

Types of Higher Order Constructs

Second, researchers can choose among different approaches to identify the higher-order construct. Prominent approaches are the repeated indicators approach or the two-stage approach (Hair et al., 2018).

Third, evaluating the measurement quality of higher-order constructs is highly challenging. For example,

  1. Some authors do not assess the reliability and validity of the lower-order components and higher-order components,
  2. Others erroneously interpret the relationships between higher- and lower-order components as structural model relationships—instead of assessing the lower-order components as elements of the higher-order construct’s measurement model.
  3. Researchers frequently analyze the discriminant validity of the lower-order components, they neglect the discriminant validity assessment of the higher-order construct as a whole.

References

Salgado, J. F. (2017). Bandwidth-fidelity dilemma. Encyclopedia of Personality and Individual Differences, eds Zeigler-Hill V., Shackelford TK, editors.(Berlin: Springer International Publishing, 1-4.

Sarstedt, M., Hair Jr, J. F., Cheah, J. H., Becker, J. M., & Ringle, C. M. (2019). How to specify, estimate, and validate higher-order constructs in PLS-SEM. Australasian Marketing Journal (AMJ), 27(3), 197-211.