[R] predict.lm[e] with formula passed as a variable

[Thaler, Thorn, LAUSANNE, Applied Mathematics]

Dear all,

In a function I paste a string and convert it to a formula which I pass
to lm[e]. The idea is to write a function which takes the name of the
response variable and the explanatory variable and the data frame as an
argument and calculates an lm[e]. (see example below)

This works fine, but if I want to make a prediction on this model, R
complains that the object holding the formula ("form") cannot be found.
How can I circumvent this problem? I think I've to provide somehow an
environment to "predict" holding the binding for the variable "form",
such that predict can resolve the variable, but I've no clue how to do
this.

Help is very much appreciated.

BR + thanks,

Thorn

----------------8<----------------
df <- data.frame(x=factor(rep(1:2, each=10)), y=c(rnorm(10), rnorm(10,
10)), z=rep(1:10,2))

test <- function(df, resp, x, rf, LM = FALSE) {
  form <- paste(resp, x, sep = " ~ ")
  form <- as.formula(form)
  if (LM) {
    mod <- lm(form, data=df)
  } else {
    rand <- as.formula(paste("~1", rf, sep = " | "))
    mod <- lme(form, data = df, random = rand)
  }
  x.new <- data.frame(levels(df[[x]]))
  names(x.new) <- x
  if (LM)
    predict(mod, x.new)
  else
    predict(mod, x.new, level=0)
}

> test(df, "y", "x", "z")
Error in eval(expr, envir, enclos) : object 'form' not found
---------------->8----------------