Hierarchical Bootstrap Simulation
Also known as: cluster bootstrap, multilevel bootstrap, nested bootstrap resampling, hierarchical resampling
Hierarchical bootstrap simulation is a resampling technique designed for data with nested or clustered structure — students within schools, patients within hospitals, repeated measures within subjects. It preserves the natural grouping of the data by resampling at each level of the hierarchy in sequence, producing a sampling distribution that correctly reflects both between-group and within-group variability.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use hierarchical bootstrap when your data have a nested or clustered structure — schools, clinics, repeated measures, geographic regions — and you want valid inference without strong distributional assumptions. It is particularly valuable for small numbers of clusters (fewer than 30-50), where cluster-robust sandwich standard errors perform poorly. Use it also when the quantity of interest is a complex function of nested data (ICCs, cross-level interactions) with no obvious analytical variance formula. Do not use it as a substitute for thinking about the sampling design: if data were collected with simple random sampling and no clustering, ordinary bootstrap suffices and hierarchical bootstrap adds unnecessary complexity.
Strengths & limitations
- Preserves the nested structure of the data, giving valid standard errors and confidence intervals for clustered observations.
- Makes no distributional assumptions; validity rests only on the resampling approximation to the true data-generating process.
- Handles small numbers of clusters better than asymptotic cluster-robust standard errors.
- Applicable to any estimator — means, regression coefficients, ICCs, nonlinear functionals — without requiring an analytical variance formula.
- Naturally extends to three or more levels by adding resampling steps for each additional level.
- Computationally expensive: each bootstrap replicate requires fitting the model to a new dataset, and thousands of replicates are needed.
- With very few clusters (fewer than 5-10), the bootstrap distribution becomes too coarse to be reliable.
- Requires the analyst to correctly specify and implement the full hierarchical resampling scheme; errors in level assignment invalidate the results.
- Parallel resampling within duplicated clusters can inflate apparent variability if not handled carefully.
Frequently asked
How does hierarchical bootstrap differ from ordinary bootstrap?
Ordinary bootstrap resamples individual observations independently, which is only valid when observations are truly independent. Hierarchical bootstrap resamples at each level of the hierarchy — groups first, then units within groups — preserving the clustering that makes observations within a group correlated. Ignoring this structure with ordinary bootstrap typically produces standard errors that are too small and confidence intervals that are too narrow.
How many bootstrap replicates do I need?
For standard error estimation, 1,000 replicates is usually sufficient. For confidence interval endpoints — especially percentile or BCa intervals — 5,000 to 10,000 replicates are recommended to reduce Monte Carlo error in the tails. With computationally intensive models, 1,000 replicates with a percentile interval is a reasonable practical compromise.
What if I have very few clusters (e.g., 5-10)?
With fewer than about 10-15 clusters, even the hierarchical bootstrap may be unreliable because there are too few unique groups to resample. In this situation, the wild cluster bootstrap (which does not resample clusters but instead randomly flips the sign of cluster-level residuals) is often preferred. Some researchers also report results under both methods as a sensitivity check.
Can hierarchical bootstrap handle three or more levels?
Yes. Simply add a resampling step for each additional level, working from the highest level down. For example, with schools nested within districts nested within states, resample states, then schools within selected states, then students within selected schools. Each level of resampling propagates that level's uncertainty into the final bootstrap distribution.
Is hierarchical bootstrap a Bayesian method?
No — it is a frequentist resampling method. It produces bootstrap confidence intervals and standard errors, not posterior credible intervals. However, it is often used alongside Bayesian hierarchical models as a frequentist diagnostic or alternative, and both approaches aim to correctly propagate uncertainty across levels of a nested data structure.
Sources
- Davison, A. C. & Hinkley, D. V. (1997). Bootstrap Methods and their Application. Cambridge University Press. ISBN: 978-0521574716
- Cameron, A. C., Gelbach, J. B. & Miller, D. L. (2008). Bootstrap-based improvements for inference with clustered errors. Review of Economics and Statistics, 90(3), 414-427. DOI: 10.1162/rest.90.3.414 ↗
How to cite this page
ScholarGate. (2026, June 3). Hierarchical Bootstrap Simulation. ScholarGate. https://scholargate.app/en/bayesian/hierarchical-bootstrap-simulation
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Gibbs SamplingBayesian↔ compare
- Hierarchical Bayesian InferenceBayesian↔ compare
- Kalman FilterBayesian↔ compare
- Multilevel Bootstrap SimulationBayesian↔ compare
- Sequential Monte CarloBayesian↔ compare