Issue/ feature request: step.lmrobdetMM

[tobiasschoch]

Dear Matias,

If step.lmrobdetMM() is called in a function, it fails. For instance, consider this function:

test <- function(formula, dat) {
    fm <- lmrobdetMM(formula, dat)
    step.lmrobdetMM(fm, formula, "backward")
}

The call of test(formula = stack.loss ~ Air.Flow + Water.Temp + Acid.Conc., dat = stackloss) terminates with Error in eval(mf, parent.frame()) : object 'dat' not found.

The issue is due to the call of update() in Line 135 of function drop1.lmrobdetMM(); see here R/RFPE.R, that is

curobj <- update(object, curfrm)

If we replace this line by

curobj <- update(object, curfrm, data = object$model)

the issue is resolved. However, this change assumes that lmrobdetMM() has been called with argument model = TRUE, otherwise it fails. The current implementation does not impose this assumption. So, in order for the proposed change (feature request) to work correctly, we need (at least) a guard in step.lmrobdetMM() which checks that object$model is not NULL and otherwise terminates with an error. Another solution is something along the lines

curobj <- if(is.null(object$model))
   update(object, curfrm)
else
   update(object, curfrm, data = object$model)

However, I have not studied or tested this if-statement ... Beyond that, I have not looked into the problem in any greater detail.

Kind regards, Tobias

[msalibian]

Thank you @tobiasschoch I will look at this shortly.