lessTemplates

The objective of lessTemplates is to provide a simplified implementation for some common models which can be used in lessSEM. Currently included are:

1. cross-lagged panel models (with random intercepts).
2. continuous time structural equation models.
3. SEM with definition variables.

Additionally, lessSEM also provides some transformations for these models which allow for regularizing specific model structures (e.g., to test measurement invariance in cross-lagged panel models).

Installation

You can install the development version of lessTemplates from GitHub with:

# install.packages("devtools")
devtools::install_github("jhorzek/lessTemplates")

Main functions

The package includes vignettes to explain the main functions. See…

1. … vignette(topic = "Cross-Lagged-Panel-Model", package = "lessTemplates") for the cross-lagged panel model
2. … vignette(topic = "Continuous-Time-SEM", package = "lessTemplates") for the continuous time structural equation model implementation.
3. … ?lessTemplates::SEMWithDefinitionVariables for definition variables (there is currently no vignette for definition variables)

Example

# The following simulation and analysis of a random intercept cross-lagged panel model
# is based on the syntax from Jeroen D. Mulder & Ellen L. Hamaker (2021) Three Extensions of the Random
# Intercept Cross-Lagged Panel Model, Structural Equation Modeling: A Multidisciplinary Journal,
# 28:4, 638-648, DOI: 10.1080/10705511.2020.1784738
#
# See https://jeroendmulder.github.io/RI-CLPM/lavaan.html
set.seed(123)

library(lessTemplates)
library(lavaan)
library(lessSEM)

# Simulate Data
data <- simulateRICLPM(seed = 123)

# Set up model
model <- "
# autoregressive and cross-lagged parameters:
eta1_(u) ~ a11_(u)*eta1_(u-1) + a12_(u)*eta2_(u-1)
eta2_(u) ~ a21_(u)*eta1_(u-1) + a22_(u)*eta2_(u-1)

# covariances
eta1_(u) ~~ 0*eta2_(u) + v11*eta1_(u)
eta2_(u) ~~ v22*eta2_(u)

# Add observations:
eta1_(u) =~ 1*y1_(u)
eta2_(u) =~ 1*y2_(u)

y1_(u) ~~ 0*y1_(u)
y2_(u) ~~ 0*y2_(u)

# random intercepts
RI_eta1 =~ 1*y1_(u)
RI_eta2 =~ 1*y2_(u)

RI_eta1 ~~ vri11*RI_eta1 + vri12*RI_eta2
RI_eta2 ~~ vri22*RI_eta2
"

# create the lavaan syntax using lessTemplates:
m <- lessTemplates::CLPM(model = model,
                         data = data,
                         addManifestVar = "no")
# fit the model:
fit <- sem(model = m$model,
           data = m$data,
           meanstructure = TRUE,
           missing = "ml")
# get the parameter estimates
coef(fit)

# In the simulation, we assumed that the autoregressive and cross-lagged parameters
# all stay the same over time (e.g, a11_u1 = a11_u2 = a11_u3 = a11_u4 = a11_u5).
# In practice, however, we won't know that. In the following, we will test this
# automatically
transform <- lessTemplates::transformCLPM(CLPM = m,
                                          parameters = c("a11_(u)", "a12_(u)", "a21_(u)", "a22_(u)"),
                                          # check measurement invariance of these parameters:
                                          transformation = "measurementInvariance")
# fit using lasso regularization:
lf <- lasso(lavaanModel = fit,
            regularized = transform$regularized,
            nLambdas = 50,
            modifyModel = modifyModel(transformations = transform$transformation))
# let's plot the regularized parameters:
plot(lf)
# In the best model, all regularized parameters have been set to zero:
coef(lf, criterion = "BIC")
# We can conclude that the lasso regularization suggests invariance of the
# autoregressive and cross-lagged parameters