model.frame method

[LiNk-NY]

It seems like when using lm-type of functions, the DataFrame does not behave like the
data.frame.
An example as given in therneau/survival#46:

library(splines)
library(S4Vectors)
data(longley)
test <- DataFrame(longley)
test$year2 <- ns(test$Year, 3)

model.frame(GNP ~ Unemployed + year2, data = test)
## Error: object 'year2' not found

# similarly...


lm(GNP ~ Unemployed + year2, test)
## Error: object 'year2' not found

Regards,
Marcel

[lawremi]

How about submitting a PR adding an S3 method for model.frame() on DataFrame? Note that model.frame() can operate on environments, and one can coerce a DataFrame to an environment with as.env(). Thanks!

[LiNk-NY]

Hi Michael, @lawremi

The model.frame method could look like this:

model.frame.DataFrame <- function(formula, ...)
{
    formula <- as.data.frame(formula)
    NextMethod()
}

It uses the model.frame.default method on a data.frame.

model.frame dispatches on the formula class. I'm not sure how to make it work with DataFrame
other than modifying model.frame.default. Coercing to environment beforehand when
model.frame.default works though:

model.frame(GNP ~ Unemployed + year2, data = as.env(test))

[lawremi]

Yea, that's what I meant by using as.env(). So you have a way forward, right?

[LiNk-NY]

I'm not sure how to implement this method.

model.frame.DataFrame only works when the formula argument is a DataFrame.
It doesn't work when you have a model.frame(y ~ x, data = DF) type of construction.

What am I missing here?

[lawremi]

I see. Passing a formula dispatches to model.frame.default(), so we could introduce a model.frame.formula() that delegates to an S4 generic, which dispatches to a method that calls as.env(). If they pass something other than a formula, this breaks.

Now that I look into it, since 2010 (after we made as.env()), base::as.environment() is generic, dispatching internally. It lacks the extra arguments of as.env(), but a case could be made for model.frame.default() to call as.environment(data). It already works for data.frame. Up for submitting a patch to R?