Working with covariance: Using higher-order factors in structural equation modeling with trust constructs

Abstract

Clarifying the “conceptual morass” of the social science of trust is a critical endeavor, and Structural Equation Modeling (SEM) is an important tool for researchers seeking to investigate the relationships among and relative influence of the many trust constructs in this expanding literature. Problematically, however, the often conceptually overlapping nature of the constructs themselves can create covariance problems that are only exacerbated by SEM’s ability to partition shared and unshared variance among indicators. These challenges can, in some situations, entirely preclude researchers from using SEM to test theoretically important hypotheses. There are a number of potential strategies available to researchers to address these problems, notably including both item- and factor-level aggregation techniques. Importantly, however, these aggregation strategies often compromise many of the benefits that make SEM so attractive in the first place. We therefore recommend that researchers with strongly correlated latent constructs test a specific alternative model in which higher-order factors are used to predict the covariance among the latent factors. These models address the problems that arise from working with excessive covariance while preserving the conceptual and statistical distinctiveness of the lower-order factors and permitting researchers to test their independent influence on important outcomes. To aid in illustrating this approach, the chapter includes a real-world data example in which various alternative model specifications are tested, highlighting the utility of higher-order factor models for trust researchers.