Adaptive sampling (also known as response-adaptive designs) is a technique that allows you to change your entry requirements as the experiment advances, based on preliminary findings.
Conventional sampling and research methodologies in medical research might lead to ethical difficulties and clinical studies that aren’t in the best interests of the patients concerned. Traditional basic random selection with a control and an experimental group, for example, might result in half of the patients getting a placebo, “treatment as usual,” or potentially no therapy at all. If a new medicine appears to have the potential to save lives, enrolling more patients in the trial group may be the preferable option.
- Adaptive sampling isn’t just valuable in medicine: it’s also important in computer programming, industrial research and applications, and a variety of other sectors. Response-adaptive designs are not only ethically good, but they are also less expensive.
Adaptive sampling examples
To understand it easier, let’s look at practical examples.
A study endeavor in California looking for gold in streams and rivers is a good example. With no prior knowledge of where gold is most likely to be discovered, the researchers might start by studying river sand in a series of randomly selected river sites across California. However, if a trace quantity of gold was discovered in a river that traveled through a region from a particular mountain range, this would make sense to alter the probability density and run a greater proportion of tests in that location. This is the most basic kind of adaptive sampling.
Similar to the first example, when doing a rare plant survey, a botanist may be tempted to sample more extensively in an environment where the first plant is found to check if others are found in a clump. Adaptive sampling designs are used to take advantage of population spatial patterns in order to produce more exact estimates of population abundance. In many cases, adaptive sampling is far more efficient than traditional random sampling for a given amount of work.
Many randomly picked quadrats will include no animals or plants when creatures are scarce and strongly grouped in their geographical distribution. In certain circumstances, choosing groups in a non-random manner may be beneficial. Adaptive cluster sampling starts with a sample of quadrats chosen using simple random sampling with replacement or without replacement in the usual way. Additional quadrats in the region of the initial transect line are added to the sample when one of the chosen quadrats includes the organism of interest. Adaptive cluster sampling is best suited to heavily clumped populations.
Before we can apply adaptive cluster sampling, we must first define the sample universe:
- the criterion for quadrat selection: If a quadrat has at least y creatures (usually y = 1), it is chosen.
- All quadrats that share one side with quadrat x are said to be in their neighborhood.
- Edge quadrats are quadrats that do not meet the selection requirement but are next to quadrats that do (i.e. empty quadrats).
- A network is a collection of quadrats in which any one of them if chosen at random, would result in all of them included in the sample.
In traditional sampling approaches, you select your entire sample before looking at the data. Contrary to that, adaptive sampling allows you to pause sampling in the middle and analyze/observe what you’ve collected so far.
Essentially, you’re identifying the data points that provided you with the greatest information. Then you’ll use that information to make future data choices. This should provide you with the best adaptive sampling possible. In most real-world research situations, we don’t come up with the best solution. However, we strive towards a realistic demand for efficient sampling that yields outcomes that are close to ideal.