Using R's lm on a dataframe with a list of predictors

Question

I have a dataframe with let's say N+2 columns. The first is just dates (mainly used for plotting later on), the second is a variable whose response to the remaining N columns I would like to compute. I'm thinking there must be something like

df = data.frame(y = 1:10, x1 = runif(10), x2 = rnorm(10))
fit = lm(y~df[,2:3],data=df)

This doesn't work. I've also tried and failed with

fit = lm(y~sapply(colnames(df)[2:3],as.name),data=df)

Any thoughts?

Answer 1

Using the formula notation y ~ . specifies that you want to regress y on all of the other variables in the dataset.

df = data.frame(y = 1:10, x1 = runif(10), x2 = rnorm(10))
# fits a model using x1 and x2
fit <- lm(y ~ ., data = df) 
# Removes the column containing x1 so regression on x2 only
fit <- lm(y ~ ., data = df[, -2]) 

Answer 2

There is an alternative to Dason's answer, for when you want to specify the columns, to exclude, by name. It is to use subset(), and specify the select argument:

df = data.frame(y = 1:10, x1 = runif(10), x2 = rnorm(10))
fit = lm(y ~ ., data = subset(df, select=-x1))

Trying to use data[,-c("x1")] fails with "invalid argument to unary operator".

It can extend to excluding multiple columns: subset(df, select = -c(x1,x2))

And you can still use numeric columns:

df = data.frame(y = 1:10, x1 = runif(10), x2 = rnorm(10))
fit = lm(y ~ ., data = subset(df, select = -2))

(That is equivalent to subset(df, select=-x1) because x1 is the 2nd column.)

Naturally you can also use this to specify the columns to include.

df = data.frame(y = 1:10, x1 = runif(10), x2 = rnorm(10))
fit = lm(y ~ ., data = subset(df, select=c(y,x2)) )

(Yes, that is equivalent to lm(y ~ x2, df) but is distinct if you were then going to be using step(), for instance.)

Answer 3

I am fairly new to R, but I found another way to do this for named columns in a data frame. Say you want to run regression using all columns except for column x2, then you'll write:

df = data.frame(y = 1:10, x1 = runif(10), x2 = rnorm(10))
# Removes the column containing x2 so regression on x1 only
model <- lm(Y ~ . - x2, data = df)
# to remove more columns (assuming there were more columns in the data frame)
model <- lm(Y ~ . - x2 - x3 - x4, data = df)

The rest of the answers are pretty old, so maybe it's a new feature, but it's pretty neat!