Stratified Bootstrap Method

A Stratified Bootstrap Method is a bootstrap resampling method that maintains stratum proportions while performing resampling with replacement to control variance estimation across stratification variables.

• AKA: Stratified Bootstrap Resampling, Stratified Bootstrap Sampling, Proportional Bootstrap Method.
• Context:
  • It can typically preserve Stratum Distributions during bootstrap iterations.
  • It can typically control Variance Components per prompt category in NLG evaluation.
  • It can often improve Confidence Interval Precision over simple bootstrap.
  • It can often handle Imbalanced Datasets through proportional sampling.
  • It can support Multi-Level Stratification with nested structures.
  • It can generate Bootstrap Distributions respecting population heterogeneity.
  • It can enable Robust Statistical Inference for stratified designs.
  • It can integrate with Mixed Effects Models accounting for stratum effects.
  • It can range from being a Single-Stratum Bootstrap Method to being a Multi-Stratum Bootstrap Method, depending on its stratification level.
  • It can range from being a Fixed-Size Bootstrap Method to being a Variable-Size Bootstrap Method, depending on its sample size strategy.
  • It can range from being a Parametric Stratified Bootstrap to being a Non-Parametric Stratified Bootstrap, depending on its distribution assumption.
  • It can range from being a Balanced Stratified Bootstrap to being an Unbalanced Stratified Bootstrap, depending on its stratum weighting.
  • ...
• Examples:
  • NLG Stratified Bootstrap Methods, such as:
    • Prompt-Stratified Bootstrap for prompt-level variance control.
    • Domain-Stratified Bootstrap for domain representation.
    • Difficulty-Stratified Bootstrap for difficulty balance.
  • Evaluation-Specific Stratified Bootstraps, such as:
    • Rater-Stratified Bootstrap for rater effect control.
    • System-Stratified Bootstrap for system comparison.
  • Statistical Applications, such as:
    • Stratified Bootstrap Confidence Interval for interval estimation.
    • Stratified Bootstrap Hypothesis Test for significance testing.
  • ...
• Counter-Examples:
  • Simple Bootstrap Method, which ignores stratification structure.
  • Jackknife Method, which uses leave-one-out rather than resampling.
  • Permutation Test, which shuffles rather than resamples with replacement.
• See: Bootstrap Resampling Method, Simple Bootstrap Method, Resampling Method, Statistical Inference, Human Parity Metric, Mixed Effects Evaluation Model, Confidence Interval Estimation, Monte Carlo Method, Evaluation Protocol.

---