Create function with direct input into lavaan mediation.model in R

In order to run a mediation model using the lavaan package in R, you have to first create a mediation model using this code:

mediation.model <- ' 
             # mediator
             mediator ~ a*iv
             dv ~ b*mediator
           
             # direct effect
             dv ~  c*iv

           # indirect effect (a*b)
             ab := a*b

           # total effect
             total := c + (a*b)
         '

I am trying to write a function so that I can input specific variables and it'll automatically put them into the model and output the model summary. I have many mediators to test.

This is my current draft of the function, but it doesn't seem to work because the mediation.model part is between parentheses.

med <- function(iv, mediator, dv) {
  mediation.model <- ' 
             # mediator
             mediator ~ a*iv
             dv ~ b*mediator
           
             # direct effect
             dv ~  c*iv

           # indirect effect (a*b)
             ab := a*b

           # total effect
             total := c + (a*b)
         '
  fit <- sem(mediation.model, data = df, meanstructure = TRUE,
           se = "boot", bootstrap = 500)

  return(summary(fit, fit.measures=T, standardized=T, ci=TRUE))
}

How can I modify this so I can input iv (independent variable), mediator, and dv (dependent variable) and get the output of the entire model? I will eventually need to loop this over several mediators (probably using lapply). I am also open to the input of the function only being the "mediator" and putting in the iv and dv manually if necessary.

Solution

You can try using paste() to assemble your model syntax.

med <- function(iv, mediator, dv) {
  stopifnot(inherits(iv, "character"))
  stopifnot(inherits(dv, "character"))
  stopifnot(inherits(mediator, "character"))

  mediation.model <- paste0(' 
             # mediator
             mediator ~ a*', iv, '
             ', dv, ' ~ b*', mediator, '
           
             # direct effect
             ', dv, ' ~  c*', iv, '

           # indirect effect (a*b)
             ab := a*b

           # total effect
             total := c + (a*b)
         ')
  fit <- sem(mediation.model, data = df, meanstructure = TRUE,
           se = "boot", bootstrap = 500)

  return(summary(fit, fit.measures=T, standardized=T, ci=TRUE))
}

Note that lavaan's model.syntax need not be a single character string, but can be a character vector (e.g., each specified parameter can be 1 element in the vector).