lavPredictY_cv: Determine an optimal lambda penalty value through cross-validation

Description

This function can be used to determine an optimal lambda value for the lavPredictY function, based on cross-validation.

Usage

lavPredictY_cv(object, data = NULL,
              xnames = lav_object_vnames(object, "ov.x"),
              ynames = lav_object_vnames(object, "ov.y"),
              n_folds = 10L,
              lambda_seq = seq(0, 1, 0.1),
              ...)

Arguments

object: An object of class lavaan.
data: A data.frame, containing the same variables as the data.frame that was used when fitting the model in object.
xnames: The names of the observed variables that should be treated as the x-variables. Can also be a list to allow for a separate set of variable names per group (or block).
ynames: The names of the observed variables that should be treated as the y-variables. It is for these variables that the function will predict the (model-based) values for each observation. Can also be a list to allow for a separate set of variable names per group (or block).
n_folds: Integer. The number of folds to be used during cross-validation.
lambda_seq: An R seq() containing the range of lambda penalty values to be tested during cross-validation.
...: To support old argument names.

Details

This function determines an optimal lambda value for use with lavPredictY, in order to improve prediction accuracy.

References

de Rooij, M., Karch, J.D., Fokkema, M., Bakk, Z., Pratiwi, B.C, and Kelderman, H. (2022) SEM-Based Out-of-Sample Predictions, Structural Equation Modeling: A Multidisciplinary Journal. tools:::Rd_expr_doi("10.1080/10705511.2022.2061494")

Molina, M. D., Molina, L., & Zappaterra, M. W. (2024). Aspects of Higher Consciousness: A Psychometric Validation and Analysis of a New Model of Mystical Experience. tools:::Rd_expr_doi("10.31219/osf.io/cgb6e")

Examples

Run this code

if (FALSE) {
colnames(PoliticalDemocracy) <- c("z1", "z2", "z3", "z4",
                                  "y1", "y2", "y3", "y4",
                                  "x1", "x2", "x3")

model <- '
  # latent variable definitions
  ind60 =~ x1 + x2 + x3
  dem60 =~ z1 + z2 + z3 + z4
  dem65 =~ y1 + y2 + y3 + y4
  # regressions
  dem60 ~ ind60
  dem65 ~ ind60 + dem60
  # residual correlations
  z1 ~~ y1
  z2 ~~ z4 + y2
  z3 ~~ y3
  z4 ~~ y4
  y2 ~~ y4
'
fit <- sem(model, data = PoliticalDemocracy, meanstructure = TRUE)

percent <- 0.5
nobs <- lavInspect(fit, "ntotal")
idx <- sort(sample(x = nobs, size = floor(percent*nobs)))

xnames = c("z1", "z2", "z3", "z4", "x1", "x2", "x3")
ynames = c("y1", "y2", "y3", "y4")

reg.results <- lavPredictY_cv(
    fit,
    PoliticalDemocracy[-idx, ],
    xnames = xnames,
    ynames = ynames,
    n_folds = 10L,
    lambda_seq = seq(from = .6, to = 2.5, by = .1)
)
lam <- reg.results$lambda.min

lavPredictY(fit, newdata = PoliticalDemocracy[idx,],
                 ynames  = ynames,
                 xnames  = xnames,
                 lambda  = lam)
}