COLLEGE STATISTICS • PROBABILITY AND SIMULATION

Bootstrap Resampling

Estimating the sampling distribution of nearly any statistic by resampling from the data you already have.

Historical Context & Motivation

Classical statistical inference relies heavily on distributional assumptions—normality, known functional forms, and large-sample asymptotics—to derive the sampling distributions of estimators. For statistics like the sample mean, the Central Limit Theorem provides an elegant theoretical foundation, but many practical problems involve statistics for which no clean closed-form sampling distribution exists. Consider, for example, the sampling distribution of a trimmed mean, a ratio of medians, or a complex regression coefficient in a nonlinear model. Before the late 1970s, statisticians faced a difficult choice: impose strong parametric assumptions that might not hold, or derive problem-specific asymptotic results that could be technically demanding and unreliable in finite samples.

The bootstrap was introduced by Bradley Efron in 1979 as a computationally intensive, assumption-lean method for approximating sampling distributions. The core insight is deceptively simple: treat the observed sample as a stand-in for the population, draw repeated samples with replacement from it, and compute the statistic of interest on each resample. The distribution of those recomputed statistics approximates the true sampling distribution. The name itself alludes to the impossible feat of pulling oneself up by one's own bootstraps—extracting information about population-level variability from a single sample.

1936
Jackknife Precursor
Maurice Quenouille introduces the jackknife technique for bias reduction, foreshadowing resampling-based inference by systematically leaving out one observation at a time.
1979
Efron's Bootstrap
Bradley Efron publishes "Bootstrap Methods: Another Look at the Jackknife" in The Annals of Statistics, formally introducing the nonparametric bootstrap and demonstrating its broad applicability.
1986
BCa Confidence Intervals
Efron introduces the bias-corrected and accelerated (BCₐ) bootstrap confidence interval, addressing skewness and bias in bootstrap distributions to yield more accurate coverage.
1993
Efron & Tibshirani Textbook
The publication of An Introduction to the Bootstrap provides a comprehensive, accessible treatment that accelerates adoption across applied disciplines.
2000s–present
Modern Computation Era
Cheap computing power and software packages (R's boot, Python's scikit-learn) make bootstrap methods routine in fields ranging from genomics to economics and machine learning.

The central question the bootstrap addresses is both fundamental and practical: How can we quantify the uncertainty of a statistic when analytical formulas are unavailable or unreliable? By leveraging computational power rather than mathematical tractability, the bootstrap democratized inference and made rigorous uncertainty quantification accessible for virtually any estimator.

Core Principles & Definitions

The bootstrap rests on a small number of conceptually transparent ideas. Understanding these principles clarifies why the method works, when it can be trusted, and where its limits lie. The foundational logic connects the relationship between a population and a sample to the analogous relationship between a sample and its resamples.

1

The Plug-In Principle

Replace the unknown population distribution F with the empirical distribution function F̂, which places probability 1/n on each observed data point. Any population quantity can then be estimated by computing the analogous quantity from F̂.
2

Resampling With Replacement

Each bootstrap sample is formed by drawing n observations from the original sample with replacement. Some observations appear multiple times; others are omitted entirely. This mimics the variability that would arise from drawing new samples from the population.
3

Bootstrap Statistic

Compute the statistic of interest θ̂* on each bootstrap sample. The collection of B such values {θ̂*₁, θ̂*₂, …, θ̂*_B} forms the bootstrap distribution, which approximates the sampling distribution of θ̂.
4

Monte Carlo Approximation

Because the total number of possible resamples (nⁿ) is astronomically large, we use a finite number of replications B (commonly 1,000–10,000) and treat the resulting empirical distribution as a Monte Carlo estimate of the ideal bootstrap distribution.
5

Bootstrap Standard Error & CI

The standard deviation of the bootstrap distribution estimates the standard error of the statistic. Percentiles of the bootstrap distribution (e.g., the 2.5th and 97.5th) yield nonparametric confidence intervals without distributional assumptions.
KEY TAKEAWAY
Imagine you want to know the variability of fish sizes in a lake, but you can only make one fishing trip and catch 50 fish. The bootstrap says: treat those 50 fish as if they were the lake. Repeatedly scoop out 50 fish from your catch (with replacement, so you might grab the same fish twice), measure the average each time, and study how those averages vary. The spread of those re-averages tells you roughly how uncertain your single-trip average is—no need for a formula about fish distributions.

Visual Explanation of the Bootstrap Process

The following diagram illustrates the complete bootstrap workflow. On the left, the original sample of n observations is shown. Arrows indicate the process of drawing bootstrap samples with replacement, each of size n. From each resample, the statistic of interest is computed, and the resulting collection of bootstrap statistics is displayed as a histogram on the right—the bootstrap distribution.

The left panel shows the original sample. Each arrow represents a resample drawn with replacement, producing a bootstrap sample of the same size n. The statistic θ̂* is computed on every resample. The histogram on the right—the bootstrap distribution—approximates the sampling distribution. The dashed red line marks the original point estimate; the standard deviation of this distribution is the bootstrap standard error.

Notice that some observations in each bootstrap sample are repeated (e.g., x₁ appears twice in Bootstrap Sample 1) while others are absent entirely. On average, about 63.2% of the original observations appear at least once in any given bootstrap sample—a result that follows from the probability 1 − (1 − 1/n)ⁿ ≈ 1 − e⁻¹. This controlled randomness is precisely what generates the variability needed to approximate the sampling distribution.

Mathematical Framework

The bootstrap can be formalized through the lens of the empirical distribution function and the plug-in principle. Let X₁, X₂, …, Xₙ be an i.i.d. sample from an unknown distribution F, and let θ = T(F) be a parameter of interest that is a functional of the distribution. The empirical distribution F̂ₙ assigns mass 1/n to each observed value. The bootstrap estimate of θ is θ̂ = T(F̂ₙ), and the goal is to estimate the sampling distribution of θ̂ around θ.

EMPIRICAL DISTRIBUTION FUNCTION
F̂ₙ(x) = (1/n) × Σᵢ₌₁ⁿ 𝟙(Xᵢ ≤ x)
F̂ₙ(x) is the proportion of sample values at or below x. The indicator function 𝟙(Xᵢ ≤ x) equals 1 when Xᵢ ≤ x and 0 otherwise. By the Glivenko–Cantelli theorem, F̂ₙ converges uniformly to F as n → ∞.
BOOTSTRAP STANDARD ERROR
SE_boot(θ̂) = √[ (1/(B−1)) × Σᵦ₌₁ᴮ (θ̂*ᵦ − θ̄*)² ]
Here θ̂*ᵦ is the statistic computed from the b-th bootstrap sample, θ̄* = (1/B) × Σᵦ₌₁ᴮ θ̂*ᵦ is the mean of all bootstrap statistics, and B is the total number of bootstrap replications. This is simply the sample standard deviation of the B bootstrap estimates.
PERCENTILE CONFIDENCE INTERVAL
CI₁₋α = [ θ̂*_(α/2) , θ̂*_(1−α/2) ]
The 100(1 − α)% bootstrap percentile confidence interval is formed by taking the α/2 and 1 − α/2 quantiles of the bootstrap distribution. For a 95% CI, these are the 2.5th and 97.5th percentiles of the B bootstrap statistics.
BOOTSTRAP BIAS ESTIMATE
Bias_boot = θ̄* − θ̂
The bootstrap estimate of bias is the difference between the mean of the bootstrap distribution (θ̄*) and the original sample estimate (θ̂). A bias-corrected estimate is θ̂_corrected = 2θ̂ − θ̄* = θ̂ − Bias_boot.

The theoretical justification for the bootstrap relies on the fact that the distribution of θ̂ − θ under F is well-approximated by the distribution of θ̂* − θ̂ under F̂ₙ. More precisely, under regularity conditions (smoothness of T, finite variance), the bootstrap is first-order asymptotically consistent, meaning that the bootstrap distribution converges (in probability) to the true sampling distribution as n → ∞. For pivotal statistics (those whose distribution does not depend on unknown parameters), the bootstrap can achieve second-order accuracy, converging faster than the normal approximation.

Bootstrap Variants & Confidence Interval Methods

The basic nonparametric bootstrap described above is the most common variant, but several important extensions and alternative confidence interval constructions have been developed to handle different data structures and improve coverage accuracy. Understanding these variants is essential for selecting the right approach in practice.

The tree at top distinguishes the nonparametric bootstrap (resample raw data) from the parametric bootstrap (simulate from a fitted model). Within the nonparametric family, the pairs and residual approaches handle regression contexts. The four confidence interval methods are ordered by increasing theoretical sophistication and coverage accuracy.

The nonparametric bootstrap resamples directly from the observed data and makes no assumption about the underlying distribution, making it the default choice for most applications. The parametric bootstrap instead fits a parametric model (e.g., normal, exponential) to the data and generates bootstrap samples by simulating from the fitted distribution. It can be more efficient when the parametric model is correct but introduces model misspecification risk. In regression settings, the pairs bootstrap resamples entire (xᵢ, yᵢ) pairs and is robust to heteroscedasticity, while the residual bootstrap resamples residuals and adds them back to fitted values, which is more efficient when the model is well-specified and errors are homoscedastic.

📊 Choosing B: How Many Resamples?
For standard error estimation, B = 200–500 is often adequate. For confidence intervals—especially percentile-based ones where tail quantiles matter—B = 2,000–10,000 is recommended. The Monte Carlo error of the bootstrap SE estimate decreases as 1/√(2B), so doubling B reduces Monte Carlo noise by about 30%. Computing is cheap; err on the side of more resamples.

Worked Example: Bootstrap CI for a Median

A researcher measures the reaction times (in milliseconds) of n = 12 participants: {245, 312, 278, 295, 340, 267, 310, 289, 305, 330, 258, 375}. She wants a 95% confidence interval for the population median reaction time. No standard formula for the SE of a median exists in closed form for small n without normality assumptions, making this an ideal setting for the bootstrap.

Bootstrap 95% CI for the Median Reaction Time
1
Step 1 — Compute the Original StatisticSort the data: {245, 258, 267, 278, 289, 295, 305, 310, 312, 330, 340, 375}. With n = 12 (even), the sample median is the average of the 6th and 7th values: (295 + 305) / 2.
θ̂ = median = 300.0 ms
2
Step 2 — Generate Bootstrap SamplesSet B = 5,000 (a practical choice for CI construction). For each b = 1, 2, …, 5000, draw a sample of size 12 with replacement from the original data. For illustration, one such resample might be: {289, 312, 245, 312, 305, 278, 340, 289, 267, 295, 375, 289}.
3
Step 3 — Compute the Bootstrap StatisticFor the illustrative resample above, sorting gives {245, 267, 278, 289, 289, 289, 295, 305, 312, 312, 340, 375}. The median is (289 + 295) / 2 = 292.0. Repeat this for all 5,000 resamples to obtain θ̂*₁, θ̂*₂, …, θ̂*₅₀₀₀.
Example: θ̂*₁ = 292.0 ms
4
Step 4 — Compute the Bootstrap Standard ErrorCalculate the standard deviation of the 5,000 bootstrap medians. Suppose this yields SD(θ̂*) = 18.7 ms. This is the bootstrap estimate of the standard error of the median.
SE_boot ≈ 18.7 ms
5
Step 5 — Construct the Percentile CISort the 5,000 bootstrap medians. The 2.5th percentile is the value at position ⌈0.025 × 5000⌉ = 125, and the 97.5th percentile is at position ⌈0.975 × 5000⌉ = 4875. Suppose these values are 267.0 and 330.0 respectively.
95% Bootstrap Percentile CI: [267.0, 330.0] ms
6
Step 6 — InterpretWe are 95% confident that the population median reaction time lies between 267 and 330 milliseconds. Note that the interval is not perfectly symmetric about the sample median of 300—the bootstrap naturally captures any skewness in the sampling distribution of the median.

Strengths, Limitations, and Practical Guidance

The bootstrap is remarkably versatile, but it is not a universal remedy. Understanding where it excels and where it can fail is critical for responsible application. The table below summarizes the main strengths and limitations, followed by practical guidelines for avoiding common pitfalls.

Strengths and limitations of bootstrap resampling
StrengthsLimitations
Distribution-free. No need to assume normality or any particular parametric form.Requires representative sample. If the original sample poorly represents the population (e.g., extreme outliers missing), the bootstrap inherits that bias.
General-purpose. Works for means, medians, correlations, regression coefficients, ratios, custom statistics.Fails for extreme quantiles. Statistics that depend on the tails (e.g., max, min, 99th percentile) are poorly estimated because the sample cannot represent unseen extremes.
Captures skewness. Unlike symmetric normal-based CIs, bootstrap intervals can reflect asymmetry in the sampling distribution.Small samples. With very small n (< 10–15), the empirical distribution is a coarse approximation of F, and coverage can be poor.
Easy to implement. A simple for-loop plus a random number generator suffices; no complex derivations needed.Dependent data. Naïve resampling destroys temporal or spatial correlation. Specialized methods (block bootstrap, circular bootstrap) are needed for time series.
Automatic. Avoids error-prone delta-method calculations for complex functions of estimators.Computationally intensive. For expensive-to-compute statistics or massive datasets, even B = 1,000 may be slow (though parallelization helps).
KEY TAKEAWAY
Think of the bootstrap like a flight simulator for statistical estimation. A simulator cannot recreate conditions the pilot has never encountered (analogous to population features absent from your sample), but it can faithfully reproduce variability within its programmed scenarios. The bootstrap's accuracy depends on how well your sample captures the essential features of the population—just as a simulator's realism depends on the fidelity of its model.

Connections to Advanced Theory

The bootstrap sits within a broader ecosystem of resampling and simulation-based inference methods. Understanding how it relates to classical asymptotics, permutation tests, and modern machine learning techniques provides deeper insight into its role in statistical theory and practice.

Comparison of bootstrap, normal approximation, and permutation test
FeatureBootstrapNormal Approximation (CLT)Permutation Test
GoalEstimate sampling distribution of any statisticApproximate sampling distribution via normal theoryTest null hypothesis of exchangeability
Assumptionsi.i.d. sample; F̂ₙ ≈ FLarge n; finite variance; smooth statisticExchangeability under H₀
ResamplingWith replacementNone (analytical)Without replacement (permute labels)
Confidence intervalsYes (multiple methods)Yes (θ̂ ± z*SE)Not directly; p-values only
Captures skewnessYesNo (symmetric by construction)Yes

In machine learning, the bootstrap is the engine behind bagging (bootstrap aggregating), introduced by Leo Breiman in 1996. Bagging trains multiple models on different bootstrap samples and averages their predictions, reducing variance. Random forests extend this idea by adding random feature selection at each split. The out-of-bag (OOB) error—which exploits the roughly 36.8% of observations left out of each bootstrap sample—provides an internal estimate of prediction error without a separate validation set. Additionally, in Bayesian statistics, the Bayesian bootstrap (Rubin, 1981) replaces uniform 1/n weights with random Dirichlet weights, connecting resampling to posterior inference under a nonparametric prior.

Practice Problems

PROBLEM 1CONCEPTUAL
Explain why bootstrap samples are drawn with replacement rather than without replacement. What would happen to the bootstrap distribution if we sampled without replacement instead?
PROBLEM 2BASIC CALCULATION
A bootstrap procedure with B = 4,000 replicates yields a mean of the bootstrap distribution θ̄* = 12.3 and a standard deviation of the bootstrap statistics SD(θ̂*) = 2.1. The original sample statistic is θ̂ = 12.0. Calculate (a) the bootstrap standard error, (b) the bootstrap estimate of bias, and (c) the bias-corrected estimate.
PROBLEM 3INTERMEDIATE
You have a sample of size n = 20. After running B = 10,000 bootstrap replicates for the sample median, the sorted bootstrap medians have the following selected values: θ̂*₍₂₅₀₎ = 41.2, θ̂*₍₅₀₀₎ = 43.8, θ̂*₍₉₅₀₀₎ = 58.1, θ̂*₍₉₇₅₀₎ = 60.4. Construct (a) a 95% bootstrap percentile CI and (b) a 90% bootstrap percentile CI.
PROBLEM 4APPLIED
An economist wants to estimate the ratio of median household income in City A to that of City B. She has independent random samples of 50 households from each city. Describe how to construct a bootstrap 95% confidence interval for this ratio. Address which type of resampling is appropriate and how many replicates you would recommend.
PROBLEM 5CRITICAL THINKING
Consider using the bootstrap to estimate the standard error of the sample maximum from a sample of size n = 30 drawn from an unknown continuous distribution. Explain why the bootstrap may perform poorly in this setting, and propose an alternative approach that might yield more reliable inference.

Summary

Bootstrap resampling is a simulation-based method for estimating the sampling distribution of virtually any statistic. The method applies the plug-in principle: the unknown population distribution F is replaced by the empirical distribution function F̂ₙ, and repeated resampling with replacement generates B bootstrap samples from which the statistic is recomputed. The standard deviation of these B values provides the bootstrap standard error, and quantiles of the bootstrap distribution yield nonparametric confidence intervals without distributional assumptions.

Key variants include the nonparametric bootstrap (resample raw data) and the parametric bootstrap (simulate from a fitted model). CI methods range from the simple percentile method to the more accurate BCₐ interval. The bootstrap excels when analytical formulas are unavailable, naturally captures skewness and bias, and connects to modern methods like bagging and random forests in machine learning. Its main limitations arise with very small samples, dependent data, and statistics that depend on extreme tail behavior.

Varsity Tutors • College Statistics • Bootstrap Resampling