There are a range of different sampling methods used when selecting a testing panel for research. This research can involve testing either a theory or a specific product, carrying out an opinion poll, or any other research which aims to cover a particular group in its entirety. This group is known as the population, though it can involve any type of group, not just the citizens of one country.
With a small population, such as staff who work in a particular office, it's usually possible to question or test everyone involved. This is known as a census study. With most populations, such as "everyone in China aged 65 or over," it's impossible to question or test everyone, so a sample group must be selected. The different ways of choosing these participants are known as sampling methods.
Sampling methods fit into one of two main categories: probability and non-probability. In a probability sampling method, everyone has a known likelihood of being selected, though this likelihood can vary from person to person. In a non-probability sampling method, some people have no chance of being selected as the participants are chosen from specified sections of the population. This can be more convenient, but comes at a price: unlike probability sampling, non-probability sampling makes it impossible to estimate how accurately the sample group represents the entire population.
The simplest form of probability sampling is to randomly select people from a list of the entire population. A variation of this method, systematic sampling, involves picking out people at fixed intervals along the list, for example every hundredth person. Both of these sampling methods have flawed as the resulting sample group may not represent the make-up of the population. For example, the sample group may have three children and seven adults, which is clearly not representative if the entire population is 20% children and 80% adults.
This can be solved by using stratified sampling, in which the population is broken down into particular groups sharing common factors and participants are selected randomly from these groups in the appropriate proportions. In the example above, the researchers would randomly select two people from a list of all children and eight people from a list of all adults. Naturally this can be extended to cover other types of group, such as by gender, to make a sample group which more accurately reflects the entire population.
The simplest forms of non-probability sampling is known as convenience sampling. The researchers simply choose the participants who are easiest to get hold of. Clearly there is a strong risk of this being very unrepresentative of the population. For example if researchers knock on doors during the day they will be less likely to get participants who are in full-time employment.
Quota sampling combines stratified sampling and convenience sampling and usually involves researchers setting out to find participants to fill quotas. In the example above, the researchers might knock on doors until they had spoken to a total of two children and eight adults. Although this method means the sample group is in the right proportions, the selection process makes it impossible to know how representative it is. In our example, the eight adults might all be unemployed, which would make them unrepresentative of the opinions of the whole population in a question about social security benefits. Because of this, quota sampling is classed as a type of nonprobability sampling.