Examination of Nonlinear Longitudinal Processes with Latent Variables, Latent Processes and Latent Classes: The R package NonLinearCurve
share
We introduce R package NonLinearCurve that provides a series of functions to evaluate longitudinal processes with individual measurement occasions in the structural equation modeling (SEM) framework. It aims to provide computational tools for nonlinear longitudinal models, especially intrinsically nonlinear longitudinal models, in the scenarios of (1) univariate longitudinal process captured by a series of latent variables, without or with covariates, including time-invariant covariates (TICs) and time-varying covariates (TVCs), (2) multivariate longitudinal processes to assess correlation or causation between longitudinal variables, and (3) mixture version of the models in scenario 1 or 2 with an assumption that trajectories are from heterogeneous latent classes. By interfacing to R package OpenMx, NonLinearCurve allows for the flexible specification of structural equation models and generates maximum likelihood estimators based on the full information maximum likelihood technique. The package provides an algorithm to have a set of initial values from the raw data, aiming to facilitate computation and improve the likelihood of model convergence. The package also provides functions for goodness-of-fit analyses, clustering analyses, plots, and predicted trajectories. This paper constitutes a companion paper to the package with introductions of each scenario of models, the estimation technique, some implementation details, output interpretation, and giving examples through a dataset on intelligence development.
READ FULL TEXT
