Problem with character vectors and linear regression in R

Question

Quick question.

I want to perform a linear regression that looks like this:

lm(y ~ x1 + x2 + x3 + x4 +x5, mydata)

This works fine if I manually write out this code.

However, the independent variables that I want to use are stored as a character, like this:

> vars
[1] "x1 + x2 + x3 + x4 +x5"

I tried typing this:

lm(y ~ vars, mydata)
Error in model.frame.default...

But it gives an error!

So then I tried this:

lm(y ~ noquote(vars), mydata)
Error in model.frame.default...

And then this

lm(y ~ print(vars, quote = FALSE), mydata)
Error in model.frame.default...

Anyone have a clue how I can get around this problem? The character string in "vars" is being provided to me by a program upstream, so I can't work around it at that level.

Thanks!

Answer 1

You can build a formula from character vectors using standard R functions and as.formula(). The trick is to note that you need to have a full formula (containing at least a ~) for R to create a formula object for you. Here is an example

## predictors
vars <- "x1 + x2 + x3 + x4 + x5"

## dummy data for example
dat <- data.frame(matrix(rnorm(120), ncol = 6))
names(dat) <- c("y", paste("x", 1:5, sep = ""))

## create a formula - here we need to paste on the response part
##  y ~
form <- as.formula(paste("y ~", vars))

## Fit the model using `form`
mod <- lm(form, data = dat)

If you print form you'll see that R has created a special object that no longer prints as a character string would:

> form
y ~ x1 + x2 + x3 + x4 + x5