A Comparison of Approaches to Forming Composite Measures in Structural Equation Models

TLDR

Structural equation modeling frequently uses composite measures derived from individual items. This study empirically compares multiple composite formation methods to assess their impact on model fit. Using data from 1,177 public school teachers, the authors tested a union‑commitment model with various composite specifications and employed bootstrapping to create two additional sample sizes. Composites consistently improved overall model fit versus item‑level indicators, with higher lambda values and explained criterion variance indicating stronger measurement models, underscoring their value for researchers.

Abstract

A common practice in applications of structural equation modeling techniques is to create composite measures from individual items. The purpose of this article was to provide an empirical comparison of several composite formation methods on model fit. Data from 1, 177 public school teachers were used to test a model of union commitment in which alternative composite formation methods were used to specify the measurement components of the model. Bootstrapping procedures were used to generate data for two additional sample sizes. Results indicated that the use of composites, in general, resulted in improved overall model fit as compared to treating all items as individual indicators. Lambda values and explained criterion variance indicated that this improved model fit was due to the creation of strong measurement models. Implications of these results for researchers using composites are discussed.