Provide a function for converting lavaan parameter tables or model strings to simsem templates

[mronkko]

A user might know lavaan syntax well and prefer that over specifying simsem templates. However, some features are available only when simsem templates are used. Therefore it would be useful to have a function to covert a lavaan model string to simsem template.

Here is a code snippet demonstrating how this could be implemented.

llibrary(simsem)
library(lavaan)


MODEL <- "
A =~ .8*a1 + .7*a2 + .6*a3
a1~~(1-.8^2)*a1
a2~~(1-.7^2)*a2
a3~~(1-.6^2)*a3
A~~1*A
B =~ .8*b1 + .7*b2 + .6*b3
b1~~(1-.8^2)*b1
b2~~(1-.7^2)*b2
b3~~(1-.6^2)*b3
B ~ .3*A
B~~(1-.3^2)*B
"

parTable <- lavaanify(MODEL)
allVars <- unique(c(parTable$rhs, parTable$lhs))
latentVars <- parTable$lhs[parTable$op == "=~"]
observedVars <- setdiff(allVars,latentVars)

cvMat <- diag(length(observedVars))
colnames(cvMat) <- rownames(cvMat) <- observedVars
fit <- lavaan::lavaan(MODEL, sample.cov = cvMat, sample.nobs = 100)

simSemTemplate <- simsem::model.lavaan(fit)

# Generate some data

d1 <- sim(nRep = 1, simSemTemplate, 100, dataOnly = TRUE, sequential = TRUE,
          saveLatentVar = TRUE)

attributes(d1[[1]])

While I was implementing this, I noticed that the code produces the following error message

[1] "The regression matrix is not recursive. Simsem template does not support non-recursive matrix.

There are two issues: First, the model is recursive, so the recursiveness check produces a false positive. Second this could be more informative by giving the the matrix name (beta, I assume) as a part of the error message.

[psunthud]

My current codes do not provide any problems using your codes. I did not update the github repository for a while. I will make sure you have the latest one soon.

[mronkko]

I will test the error again with the latest code once I have it.

Your comment addresses the request only partly: "Provide a function for converting lavaan parameter tables or model strings to simsem templates"

I would like to have the following function to be part of SimSem:

simsemTemplateFromLavaanModel <- function(model){
   parTable <- lavaanify(model)
   allVars <- unique(c(parTable$rhs, parTable$lhs))
   latentVars <- parTable$lhs[parTable$op == "=~"]
   observedVars <- setdiff(allVars,latentVars)

   cvMat <- diag(length(observedVars))
   colnames(cvMat) <- rownames(cvMat) <- observedVars
   fit <- lavaan::lavaan(MODEL, sample.cov = cvMat, sample.nobs = 100)

   simsem::model.lavaan(fit)
}

[psunthud]

I will use this code by allowing model.lavaan to take syntax or parameter tables.