Home Statistical Dictionary Adaptive Sampling
Adaptive Sampling
A sampling design in which the selection of later units depends on the observed values of units already sampled, commonly used to concentrate sampling effort on rare, clustered, or hard-to-find populations discovered mid-survey.
In Plain English
Most sampling designs are fixed in advance, you decide the whole plan before collecting any data. Adaptive sampling breaks from that: what gets sampled next depends on what you've already found. A classic example is surveying a lake for a rare, patchily distributed species, if a survey plot turns up a high count, adaptive sampling adds extra nearby plots to trace out the full extent of that cluster, rather than sticking rigidly to a predetermined grid that might miss most of it. This makes adaptive sampling especially good at finding and characterizing rare or clustered phenomena that a fixed design would likely undersample.
Definition
Adaptive sampling is any sampling design in which the procedure for selecting subsequent sampling units is allowed to depend on the values observed in units already sampled, in contrast to conventional designs, which fix the entire selection procedure in advance regardless of the data collected. A common form, adaptive cluster sampling, begins with an initial probability sample and then adds neighboring units whenever a sampled unit's observed value exceeds some threshold of interest, continuing to add neighbors of newly qualifying units until the cluster of high values is fully traced out; because the final sample size and composition are themselves random and depend on the data, valid estimation requires specialized estimators (built on the Horvitz-Thompson framework, adjusted for the design's data-dependent selection) rather than the standard formulas used for fixed designs, but the payoff is a substantial gain in efficiency when the population of interest is rare and spatially or socially clustered, since a fixed design of the same total sample size would likely miss most of the clustered population entirely.
Properties
- Adaptive sampling's efficiency advantage is greatest precisely when a fixed design would be at its weakest, populations that are rare in overall prevalence but strongly clustered spatially or socially, in such cases a conventional fixed sample of reasonable size is likely to entirely miss the clusters, while an adaptive design is specifically built to find and thoroughly trace them out once discovered.
- Because the sample size and composition under adaptive sampling depend on the data collected, standard variance formulas built for fixed-size designs no longer apply directly, valid inference requires estimators specifically derived for the adaptive design, most commonly extensions of the Horvitz-Thompson estimator that account for the data-dependent selection probabilities.
- Adaptive sampling originated in and remains most associated with ecological and environmental surveys (locating rare or patchily distributed plant and animal populations), but the same logic has been extended to epidemiology and social network studies for finding and characterizing rare or hidden subpopulations more efficiently than a fixed design would allow.
At a Glance