Dyadic Data Analysis via Structural Equation Modeling with Latent Variables: A Tutorial with the dySEM package for R

Researchers wishing to analyze cross-sectional dyadic data must select a modeling framework to accomplish their statistical goals. Historically, most researchers have chosen dyadic multilevel modeling or dyadic path analysis of observed variables via structural equation modeling (SEM). Here, we advocate for—and provide open-source tools and open-access tutorials to advance—the use of dyadic structural equation modeling with latent variables. We begin with an overview of the subordination of latent variable modeling in dyadic data analysis, review its major value propositions, and discuss key research design considerations. We then introduce a new R package, dySEM, which helps users script a variety of dyadic SEMs with latent variables for model fitting with the lavaan package, and generate reproducible outputs including tables, path diagrams, and supplementary tests and indexes. We outline the typical workflow for deploying dySEM: (1) scraping indicator information, (2) scripting models, (3) fitting with lavaan, and (4) output generation. We also discuss our approach to quality control and inclusivity in dySEM’s development. Three reproducible tutorials follow, illustrating applications to (a) a Correlated Dyadic Factors Model (CDFM), (b) a Multiple Correlated Dyadic Factors Model (M-CDFM), and (c) a Latent Actor-Partner Interdependence Model (L-APIM). Each tutorial includes the formal model, a prototypical path diagram, key research questions or hypotheses, and reproducible code for scripting and outputting. We conclude with a discussion of future development targets for dySEM, and calls for more Monte Carlo guidance for dyadic analyses and support for open-source academic developers.