Course contentsShow
← Data Science
Lesson 298 of 2,1458. Confidence IntervalsFree lesson

The Bootstrap Method: Resampling Your Data

Learn the bootstrap principle of resampling with replacement to estimate sampling distributions without assumptions.

The Bootstrap Method: Resampling Your Data

What you'll learn: How to estimate confidence intervals by repeatedly resampling your existing data, no distribution assumptions required.

The Core Idea

When you have limited data and don't know if it follows a normal distribution, the bootstrap method offers a clever solution. Instead of making assumptions about the population, you treat your sample as if it were the population and draw many new samples from it—with replacement.

Here's the key insight: by randomly selecting observations from your sample (allowing the same observation to appear multiple times in each resample), you create hundreds or thousands of "bootstrap samples." Each bootstrap sample has the same size as your original data. You calculate your statistic of interest (mean, median, etc.) for each bootstrap sample, building up an empirical sampling distribution.

Why "With Replacement" Matters

Imagine you have 100 survey responses in a bag. You draw one out, record it, put it back, shake the bag, and draw again. You repeat this 100 times to create one bootstrap sample. Some original responses appear multiple times; others don't appear at all. This randomness mimics the natural variability you'd see if you could repeatedly sample from the true population.

By creating 1,000+ bootstrap samples this way, you approximate what the sampling distribution would look like—without needing normality assumptions or complex formulas.

When to Use Bootstrap

The bootstrap shines when:

  • Your sample size is modest but not tiny (typically n > 20)
  • You can't assume normality
  • Standard formulas don't exist for your statistic
  • You want distribution-free confidence intervals

Key Takeaway: The bootstrap method resamples your data with replacement many times to build an empirical sampling distribution, letting you estimate confidence intervals without assuming your data follows any particular distribution.