Adaptive Stratified Sampling
Also known as: ASS, adaptive stratified design, stratified adaptive sampling, adaptive allocation stratified sampling
Adaptive stratified sampling divides the population into strata and then applies an adaptive rule within each stratum: whenever an initially selected unit satisfies a pre-specified condition (e.g., a rare species is found, a variable exceeds a threshold), neighboring or related units are added to the sample. This combines the variance-reduction power of stratification with the ability to concentrate sampling effort where the phenomenon of interest is actually present.
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 adaptive stratified sampling when the target variable is rare, spatially clustered, or unevenly distributed across a population that can be meaningfully stratified. It is particularly suited to ecological surveys (rare species, invasive organisms), epidemiological surveillance (disease clusters), and environmental monitoring (pollutant hotspots). The method requires that strata can be defined in advance and that a clear adaptive condition can be stated before data collection. Do NOT use it when the phenomenon is uniformly distributed (adaptive additions provide no efficiency gain), when the population structure prevents defining meaningful strata, when real-time field decisions to add units are logistically impossible, or when a simple, easily auditable design is required by regulatory standards. For small or well-enumerated populations where a census is feasible, stratification alone is sufficient.
Strengths & limitations
- Concentrates sampling effort on rare or clustered phenomena, substantially improving precision relative to fixed designs of the same size.
- Stratification controls overall geographic or demographic coverage, preventing the adaptive mechanism from leaving entire strata underrepresented.
- Unbiased estimators exist (modified Horvitz-Thompson) that properly account for adaptive inclusion probabilities.
- Flexible: the adaptive condition and network definition can be tailored to the specific target variable and field context.
- Widely validated in ecological, environmental, and epidemiological applications with published estimator theory.
- Final sample size is random and unknown before data collection, complicating logistical planning and budget management.
- Estimation is more complex than for fixed designs; software support for modified Horvitz-Thompson estimators is not as widespread as for standard stratified estimators.
- Requires a well-defined adaptive condition and network structure before fieldwork; poorly chosen conditions can negate efficiency gains.
- If the phenomenon is not clustered, the adaptive additions rarely trigger and the design reduces to ordinary stratified sampling with no efficiency advantage.
- Variance estimation can be unstable when observed networks are very large or when many networks overlap across strata boundaries.
Frequently asked
How does adaptive stratified sampling differ from ordinary stratified sampling?
In ordinary stratified sampling the sample size within each stratum is fixed in advance and every unit has a known, pre-specified inclusion probability. In adaptive stratified sampling the initial sample is fixed, but additional units are added within each stratum whenever sampled units satisfy a condition. This makes the final sample size random and creates data-dependent inclusion probabilities that require a modified estimator.
Is the adaptive stratified estimator still unbiased?
Yes, provided the correct modified Horvitz-Thompson or Hansen-Hurwitz estimator is used, which accounts for the unequal and adaptive inclusion probabilities. Using a standard stratified mean or ratio estimator on adaptively collected data produces biased estimates.
What is a 'network' in adaptive sampling?
A network is the set of units that would all be added to the sample if any one of them satisfied the adaptive condition and was selected. Networks are defined by the neighborhood rule — for example, the four cells adjacent to a selected grid cell, or all contacts of a selected respondent. Units that do not satisfy the condition form single-unit networks called edge units.
How do I choose the adaptive threshold?
The threshold should be set so that the condition is triggered infrequently enough to control sample size growth but frequently enough to provide meaningful efficiency gains. Pilot data, simulation studies, or expert knowledge of the expected prevalence and clustering intensity are the standard tools for calibrating the threshold before fieldwork.
Can I use standard survey software (e.g., R survey package) for analysis?
Standard survey software assumes fixed, pre-specified inclusion probabilities and will give incorrect variance estimates for adaptive designs. Specialized functions implementing the modified Horvitz-Thompson estimator for adaptive samples are available in the R package 'ACS' (Adaptive Cluster Sampling) and in Thompson's reference code. Always verify that your software correctly handles adaptive inclusion probabilities.
Sources
- Thompson, S. K. (1990). Adaptive cluster sampling. Journal of the American Statistical Association, 85(412), 1050–1059. DOI: 10.2307/2289601 ↗
- Thompson, S. K. (2002). Sampling (2nd ed.). Wiley-Interscience. ISBN: 978-0471360100
How to cite this page
ScholarGate. (2026, June 3). Adaptive Stratified Sampling. ScholarGate. https://scholargate.app/en/survey-methodology/adaptive-stratified-sampling
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.
- Adaptive Cluster SamplingSurvey Methodology↔ compare
- Disproportional Stratified SamplingSurvey Methodology↔ compare
- Multistage SamplingSurvey Methodology↔ compare
- Proportional Stratified SamplingSurvey Methodology↔ compare
- Stratified SamplingSurvey Methodology↔ compare
- Systematic SamplingSurvey Methodology↔ compare