example/matrixpls.plsc-example2.R

In mronkko/matrixpls: Matrix-Based Partial Least Squares Estimation

# Simulates the direct effects model from 
#
# Dijkstra, T. K., & Schermelleh-Engel, K. (2013). Consistent Partial Least Squares for Nonlinear
# Structural Equation Models. Psychometrika, 1–20. doi:10.1007/s11336-013-9370-0
#
# using PLSc, disattenuated summed scales, and disattenuated factor scores
#

####################################################################################################
#
# Simulation parameters
#
####################################################################################################

SAMPLE <- 100
REPLICATIONS <- 50
MULTICORE <- FALSE

MODEL <- "
Eta1 =~ .8*y1 + .8*y2 + .8*y3
Eta2 =~ .8*y4 + .8*y5 + .8*y6 
Eta3 =~ .8*y7 + .8*y8 + .8*y9 

y1~~.36*y1
y2~~.51*y2
y3~~.64*y3
y4~~.36*y4
y5~~.51*y5
y6~~.64*y6
y7~~.36*y7
y8~~.51*y8
y9~~.64*y9

Eta1~~1*Eta1
Eta2~~1*Eta2
Eta2~~-.3*Eta1

Eta3~.5*Eta1 + -.3*Eta2
Eta3~~.57*Eta3  # error tem is calculated as 1-(.5*(.5 + -.3*-.3) + -.3*(-.3 + -.3*.5))
"


####################################################################################################
#
# Run simulations
#
####################################################################################################

library(matrixpls)

sim.PLSf1 <- matrixpls.sim(nRep = REPLICATIONS, model = MODEL, n = SAMPLE, #General setup
                           disattenuate = TRUE,
                           fm = "minres", # Setup disattenuation with minres factor analysis
                           weightFun = weightFun.factor, # Use factor scores as proxies.
                           boot.R = FALSE, # We are not interested in bootstrap
                           multicore = MULTICORE, fitIndices = NULL, stopOnError = TRUE)


sim.PLSf2 <- matrixpls.sim(nRep = REPLICATIONS, model = MODEL, n = SAMPLE, #General setup
                           disattenuate = TRUE,
                           fm = "minres", # Setup disattenuation with minres factor analysis
                           weightFun = weightFun.fixed, # Use equal weights
                           parametersReflective = estimator.efaLoadings,
                           boot.R = FALSE, # We are not interested in bootstrap
                           multicore = MULTICORE, fitIndices = NULL, stopOnError = TRUE)

sim.PLSc <- matrixpls.sim(nRep = REPLICATIONS, model = MODEL, n = SAMPLE, #General setup
                          disattenuate = TRUE,
                          parametersReflective = estimator.plscLoadings,
                          boot.R = FALSE, # We are not interested in bootstrap
                          multicore = MULTICORE, fitIndices = NULL, stopOnError = TRUE)

####################################################################################################
#
# Report results
#
####################################################################################################


# SimSem summary methods have a bug that causes a crash with inadmissible results. (therefore 'try')

try(summary(sim.PLSf1))
try(summary(sim.PLSf2))
try(summary(sim.PLSc))

# Plot the sampling distribution of the parameters

par(mfrow=c(2,3))
for(i in 1:6){
  estimatorIndex <- (i-1) %% 3 +1
  obj <- switch(estimatorIndex, sim.PLSf1,sim.PLSf2,sim.PLSc)
  var <- colnames(obj@coef)[ceiling(i/3)]
  title <- paste(switch(estimatorIndex, "Factor scores", "Summed scales", "PLSc"),var,sep=": ")
  plot(density(obj@coef[,var], na.rm = TRUE), main = title)	
}