🎯Stratified Sampling

NEW

A sampling method that partitions a population into strata and samples each subgroup to ensure proportional representation.

DESCRIPTION

Stratified sampling divides a population into homogeneous subgroups (strata) and samples from each, often proportionally to stratum size. It reduces variance compared to simple random sampling when strata are internally similar and differ between each other. In ML, it preserves class proportions for training, validation, and testing, mitigating class imbalance effects. Weighted estimators combine stratum results for unbiased population estimates. Properly defining strata and allocating sample sizes are critical choices.

ALTERNATIVE DEFINITIONS

Wikipedia

In statistics, stratified sampling is a method of sampling from a population which can be partitioned into subpopulations.

Classical statistical framing emphasizing partitioning into subpopulations.

Source

NIST/SEMATECH e-Handbook of Statistical Methods

Stratified sampling involves dividing the population into homogeneous subgroups called strata and selecting samples from each stratum.

Emphasizes homogeneity within strata and structured selection.

Source

RELATED TERMS

Systematic Sampling Weighting Oversampling Undersampling Class Imbalance Cross-Validation Sampling Bias Survey Design

COMPETING TERMS

Random Sampling Cluster Sampling

RELATIONS

• Broader: Sampling

KNOWLEDGE AREAS

Data Science Mathematics Testing & Validation

TARGET AUDIENCE

Data Scientist Data Analyst BI Specialist Data Steward

EXTERNAL SOURCES

• Wikipedia - Stratified sampling
• Penn State Statistics - University course material on sampling
• NIST - NIST handbook section on stratified sampling

TOP QUESTIONS

Why use Stratified Sampling in machine learning?

It preserves class or group proportions in train/test splits, improving estimate stability and mitigating class imbalance in performance evaluation.

Category: Relevancy

How do I choose strata for Stratified Sampling?

Select meaningful, homogeneous groups that explain variance, such as class labels or demographic segments relevant to the analysis.

Category: HowTo

What is proportional vs disproportional Stratified Sampling?

Proportional sampling reflects population proportions; disproportional oversamples rare strata to ensure adequate representation, requiring weighting in analysis.

Category: Concepts

What are common mistakes with Stratified Sampling?

Too many small strata cause high variance; leaking target information via strata derived from outcomes can bias results.

Category: Mistakes

Does Stratified Sampling always reduce variance?

Only if strata are internally homogeneous and differ across strata; otherwise benefits may be limited compared to simple random sampling.

Category: Misunderstanding

How is Stratified Sampling used in cross-validation?

Stratified K-fold maintains label proportions in each fold, providing more reliable performance estimates for imbalanced datasets.

Category: Application

How to analyze data from disproportional Stratified Sampling?

Apply sampling weights to recover unbiased population estimates and adjust standard errors using survey methods.

Category: Analysis

When is Cluster Sampling preferable?

When the population is geographically dispersed and sampling costs dominate; cluster sampling reduces logistics at the cost of higher variance.

Category: Alternatives

How many samples per stratum are needed?

Enough to achieve desired precision; Neyman allocation can optimize per-stratum sizes based on variance and cost.

Category: Requirements