Remove variable wrapped in function from model formula in R

Question

I have a model with transformed variables, e.g.:

 data = data.frame(y = runif(100,0,10), x1 = runif(100,0,10), x2 =   runif(100, 0, 10))
 mod = lm(y ~ scale(x1) + scale(x2), data)

I would like to remove one entire variable from the formula, like so:

mod = lm(y ~ scale(x1), # x2 is gone! 
 data)

But I would like to do this using a user-supplied character string of the variable to be removed (in other words, I'm wrapping this in a function and its not feasible to edit the formula by hand, as I have here).

If the variable was untransformed, this would be simple using gsub:

remove.var = "x2"
update(mod, formula. = as.formula(gsub(remove.var, "", format(formula(mod)))))

but as such, it returns the wholly predictable error:

 > Error in as.matrix(x) : argument "x" is missing, with no default

because scale() is still in the formula!

Is there a way to do this with regexpr, or some way that I am not seeing that is totally obvious? I would like it to be scalable to other types of transformations, e.g.: log, log10,etc.

As another layer of complexity, suppose that the variable to be removed also appeared in an interaction:

 mod = lm(y ~ scale(x1) * scale(x2), data)

In this case, one would have to remove the interaction * as well (errant +s, I have found, are ok).

Any help is much appreciated. Thanks!

Answer 1

A terms-object is a formula with additional attributes:

update(mod, formula=drop.terms(mod$terms, 2, keep.response=TRUE)  )

Call:
lm(formula = y ~ scale(x1), data = data)

Coefficients:
(Intercept)    scale(x1)  
     5.0121       0.1236  

If you need to calculate that position from a string argument, then you can grep the term.labels attribute:

> grep( "x2", attr( mod$terms, "term.labels") )
[1] 2

Notice that this also succeeds with the interaction formula:

update(mod, formula=drop.terms(mod$terms, grep( "x2", attr( mod$terms, "term.labels") ), keep.response=TRUE) )
#----------

Call:
lm(formula = y ~ scale(x1), data = data)

Coefficients:
(Intercept)    scale(x1)  
     5.0121       0.1236