SEMinR: Domain-Specific Language for Building, Estimating, and Visualizing Structural Equation Models in R

Abstract

SEMinR seeks to bring the latest state-of-the-art advances in SEM methods to the R ecosystem. This package also seeks to make describing and analyzing SEMs easier for practitioners. There have been several recent advances in the various branches of SEM that are often not reflected in existing R packages. For example, the PLS-PM approach requires adjustment in how models with interaction terms are estimated. PLS-PM methods have recently incorporated predictive methods such as plsPredict. Meanwhile, CB-SEM approach can avail ten Berge factor-score extraction that obtains construct scores with the same correlation patterns as the latent factors themselves. CB-SEM researchers should also consider VIF scores in their regression models. SEMinR incorporates these and other advancements. Estimating an SEM using CB-SEM and PLS-PM requires different packages for the two estimation methods, which often requires researchers to wholly redescribe their models in different syntax. SEMinR allows researchers to describe their model once in a common syntax, and estimate the model using different estimation methods. SEMinr includes its own implementation of PLS-PM estimation that is tested against leading commercial applications to ensure comparable results. For CB-SEM estimation, SEMinR delegates the estimation to the popular Lavaan package. Regardless of which estimation method one uses, the results are structured in a similar way for reporting and visualization. R packages for SEM often use a custom syntax that does not correspond to any programming language; nor does the syntax not reflect the terminology of SEM with which practitioners are familiar. SEMinR offers researchers a domain-specific language for modeling SEMs that uses function names that evoke major SEM components: constructs, relationships, paths, reflective, composite, etc. As SEMinR’s syntax is built using R functions, researchers can inject their own custom functions to extend the behavior of SEMinR. SEMinR is the first package that allows researchers applying PLS-PM to visualize their graphical models and measurement qualities. Visualization of CB-SEM models is delegated to the semplot package. Moreover, SEMinR allows researchers to visualize models either before or after estimation.