Bootstrap Hypothesis Testing
You'll learn how to use bootstrap resampling to test hypotheses about means and other statistics without relying on parametric assumptions.
The Core Idea
Bootstrap hypothesis testing lets you test claims about your data by repeatedly resampling and simulating what would happen if the null hypothesis were true. Unlike traditional t-tests that assume normality, bootstrap methods work directly with your data's actual distribution.
Here's the key difference: instead of using a theoretical sampling distribution (like the t-distribution), you create an empirical sampling distribution through resampling.
How Bootstrap Hypothesis Testing Works
- State your hypotheses (H₀ and H₁) about a parameter (like a mean difference)
- Transform your data to reflect the null hypothesis being true (e.g., center both groups at the same mean, or shift data to match H₀)
- Resample from the null-centered data thousands of times, calculating your test statistic each time
- Build a bootstrap distribution of the test statistic under H₀
- Compare your observed statistic to this distribution to get a p-value
Think of it like this: you're asking "If the null hypothesis were true, how unusual is what I actually observed?" You answer by simulating that null world many times.
When to Use Bootstrap Hypothesis Testing
Bootstrap hypothesis tests shine when:
- Your sample size is small or moderate
- Data distributions are skewed or non-normal
- You're testing unusual statistics (medians, trimmed means, ratios)
- You want robust results without parametric assumptions
The p-value is simply the proportion of bootstrap statistics as extreme or more extreme than your observed value.
Key Takeaway: Bootstrap hypothesis testing creates an empirical null distribution through resampling, letting you test claims about any statistic without assuming normality or other parametric conditions.