Introduction

I have previously written about bootstrap modeling with {purrr} and {modelr} here. What if you would like to do some simple bootstrap modeling without importing a library? This itself is easy too!

Example

We will be using a very simple for loop to accomplish this. You will find an excellent post on this on Stats StackExchange from Francisco Jos Goerlich Gisbert

n    <- 2000
df   <- mtcars
pred <- numeric(0)

library(tictoc) # for timing

tic()
set.seed(123)
for (i in 1:n){
  boot    <- sample(nrow(df), n, replace = TRUE)
  fit     <- lm(mpg ~ wt, data = df[boot,])
  pred[i] <- predict(fit, newdata = df[boot,]) +
    sample(resid(fit), size = 1)
}
toc()

6.8 sec elapsed

So we can see that the process ran pretty quickly and the loop itself is not a very difficult one. Let’s explain a little.

So the boot object is a sampling of df which in this case is the mtcars data set. We took a sample with replacement from this data set. We took 2000 samples and did this 2000 times.

Next we made the fit object by fitting a simple linear model to the data where mpg is a function of wt. Once this is done, we made out predictions.

That’s it!