Adaptive Sampling

Background

The purpose of sampling is to learn about one or more characteristics of a population of interest by investigating a subset, which is referred to as a sample, of that population. Typical population quantities of interest include the population mean, total, and proportion. For example, a population quantity of interest might be the total number of people living in New York City, their average income, and so on. From the sample collected, estimates of the population quantities of interest are obtained. The manner in which the sample is taken is called a sampling design, and for a sampling design various estimators exist. There is a multitude of sampling designs and associated estimators. Many factors may be considered in determining the sampling design and estimator used. The main objective is to use a sampling design and estimator that yield the most precise and accurate estimates utilizing the resources available. In conventional sampling designs and estimators, the sample is taken without regard to the unit values observed. That is, the observations obtained during sampling are not used in any manner to alter or improve upon future sample selections.

In adaptive sampling, on the other hand, the sampling selections depend on the observations obtained during the survey. In this sense, adaptive sampling designs are adaptive in that, while sampling, the remaining units to be sampled may change according to previously observed units. For design-based sampling, adaptive sampling could be a more efficient design to improve the inference and also increase the sampling yield. For model-based sampling, it has been shown that the optimal sampling strategy should be an adaptive one in general under a given population model.

Adaptive sampling designs have been used in various disciplines, including the ecological, epidemiolog-ical, environmental, geographical, and social sciences.

Adaptive Cluster Sampling

Adaptive cluster sampling (ACS) is a subclass of adaptive sampling designs. There has been considerable research within the field of adaptive sampling, utilizing ACS designs and their associated estimators. There are variations of ACS, such as stratified ACS, systematic ACS, ACS without replacement of clusters, and so on. The ACS designs are often more efficient than their conventional counterparts on clustered, or patched, populations. Typically this type of sampling design—ACS—is not only more efficient but also more useful for obtaining observations of interest for rare, hard—to—find, or elusive clustered populations. For example, there are various species of animals known to travel in groups and that are rare, such as whales. Through ACS, more whales may be captured in the sample than through conventional sampling techniques using a comparable final sample size of geographic locations. For surveys focused on elusive or hidden populations, such as individuals who are intravenous drug users, or HIV-positive individuals, ACS can aid greatly in increasing the number of individuals in the survey who meet the desired characteristics.

...