How can I make a predict plot varying two variables in R?

Question

I want to plot the results of a regression model, but allowing two variable to vary simultaneously. I guess I could do that using predict() function in R, but I am running a model that does not have such function developed yet.

Let's suppose we have a data set

x1 <- rnorm(100)
x2 <- rnorm(100)
y <- 1 + x1*5 + x2*3 + rnorm(100)

And we run a simple model and calculate yhat.

mod <- lm(y ~ x1 + x2)
yhat <- predict(mod)
dt <- as.data.frame(cbind(yhat,x1,x2))

How can I plot the expected value of y for different values of x1 and x2? I tried this, but it didn't work:

contour(dt$x1, dt$x2, dt$yhat)

Error in contour.default(dt$x1, dt$x2, dt$yhat) : 
  increasing 'x' and 'y' values expected

Answer 1

To use a plot like contour() your vertical z axis is going to be a matrix so x and y need to be the points defining a grid over which to draw z. So you want something more like:

# Changed data and model from yours to make it easier to check the axes are the right way 
# around later on:

x1 <- rnorm(100)
x2 <- rnorm(100,5,1)

y <- 1 + x1*5 + x2*(-3) + rnorm(100) 

mod <- lm(y ~ x1 + x2)

gridded <- data.frame(
    x1=seq(from=min(x1), to=max(x1), length.out=100),
    x2=seq(from=min(x2), to=max(x2), length.out=100))

yhat <- predict(mod, newdata=expand.grid(gridded))

image(gridded$x1, gridded$x2, matrix(yhat,nrow=100, byrow=FALSE))
contour(gridded$x1, gridded$x2, matrix(yhat,nrow=100, byrow=FALSE), add=TRUE)