06. SEMinR. Review of Model Specification, Composites, Modes (A/B), and Weighting Schemes

A PLS-SEM workflow hinges on two linked pieces: how constructs are turned into measurable “composites” (the measurement model) and how those constructs are connected by hypothesized paths (the structural model). In this seminar-style walkthrough, the model being tested links Vision development and rewards to collaborative culture, with each construct measured by multiple survey items. The practical takeaway is that once the measurement and structural models are specified, the PLS algorithm generates construct scores that then feed regression-based estimation for both indicator relationships (weights/loadings) and path relationships (path coefficients). That chain matters because it determines what gets estimated, how it gets estimated, and which assumptions (reflective vs. formative) control the math.

The process starts in R by loading the dataset from a CSV file (including a header row and comma separators) into a data object. Next comes the measurement model specification, where constructs such as Vision development, rewards, and collaborative culture are defined as composites built from item indicators (e.g., V1–V4 for Vision development, RW1–RW? for rewards, and CC1–CC? for collaborative culture). The transcript emphasizes the conceptual role of composites in PLS-SEM: indicators are linearly combined to form composite variables that act as valid proxies for the underlying latent constructs.

With the measurement model set, the structural model is specified as directional paths from independent variables to dependent variables. In the example, Vision development and rewards are treated as independent variables predicting collaborative culture as the dependent variable. The structural model is encoded by listing the IVs as a vector and the DV as a separate target; additional DVs can be handled by adding more paths.

Estimation then follows the PLS-SEM algorithm logic used in SmartPLS: after running the PLS algorithm, construct scores are produced and used as inputs to regression models within the path model. The results include measurement-model estimates (indicator weights and loadings) and structural-model estimates (path coefficients). A key fork in the road is whether constructs are reflective or formative. Reflective constructs assume an arrow from the construct to the indicators, treat items as interchangeable, and typically use Mode A estimation (often described as based on by-variate regression/correlation logic). Formative constructs assume the arrow runs from indicators to the construct, require Mode B estimation, and are validated differently; the transcript notes that reflective is the default assumption in PLS-SEM, while formative requires explicit Mode B settings.

Finally, the weighting scheme affects estimation details. While alternative weighting schemes can yield slightly different results, the transcript highlights that the path weighting scheme is the most popular and recommended because it tends to produce the highest R² for the endogenous latent variable and works broadly across PLS path model specifications. The session closes by pointing ahead to how to implement the PLS estimate function in R after this review of model specification, composites, modes (A/B), and weighting schemes.