04. SEMinR Lecture Series | Specifying the Measurement Model | PLS-SEM in R

Specifying the measurement model is the prerequisite step for reliable PLS-SEM results in R, because any structural relationships among constructs depend on whether the constructs themselves are measured with adequate reliability and validity. In SEMR/PLS-SEM workflows, the model is built from two parts: the measurement model (outer model), which links latent variables to their indicators, and the structural model (inner model), which tests relationships among latent variables. Before any hypothesis testing on the structural paths, the measurement model must be defined so the software can evaluate construct quality.

In SEMR, measurement and structural modeling are handled in tandem, but the program distinguishes between them through how the model is specified. The measurement model’s job is to establish reliability and validity criteria for each construct; without that, any conclusions about the paths between constructs are only as trustworthy as the underlying measurement. To define these constructs in R, SEMR uses the `constructs` function, which compiles the measurement model definitions for all constructs in the study.

The `constructs` function relies on the `composite` function to describe the measurement of individual constructs. For the common case of reflective and formative constructs, the key choice is the “mode” argument: reflective constructs use `mode a`, while formative constructs use `mode P` (as described in the lecture). The `composite` function takes the construct name, the indicator item names (often grouped as multi-item scales), and the mode that indicates whether the construct is reflective or formative. For example, a reflective construct such as “collaborative culture” might be represented by multiple indicators in the dataset—CC1 through CC6—where the construct name is “collaborative culture,” the indicator code is “CC,” the item numbers run from 1 to 6, and the mode is set to reflective.

The lecture also distinguishes between multi-item and single-item constructs. Multi-item constructs are defined with `composite` and a mode (reflective vs. formative). Single-item constructs are treated differently: they are not reflective or formative in the same way, so the model definition can simply reference the single indicator name from the dataset.

A practical workflow is laid out: first create a measurement model object using `constructs`, then define each construct inside it using `composite`, and finally run the code to confirm there are no syntax errors—especially around bracket placement. In the example, the measurement model is assembled by defining multiple constructs (three predictors and one dependent variable) within the `constructs` call, each with its own indicator set (e.g., V1–V4 for Vision, development items, rewards items, and CC1–CC6 for collaborative culture). Once the measurement model is successfully specified and executed, the next step is to define the structural model in the following session.