Sampling, Random and Nonrandom

In its most general form, a sample is simply a subset of a larger population or group. In scientific applications, the purpose of sampling is to be able to draw conclusions about a larger population. There are many reasons to sample rather than attempt to collect data on the full population. They include the following:

• Cost: It is often cost prohibitive to collect data on the entire population.
• Speed: To collect data on the entire population would take an excessive amount of time.
• Feasibility: Often the complete set of population elements is unknown. Most populations do not have centralized lists of their members. Even when such lists exist, it is usually impossible to collect data on every single member.

This entry discusses the major types and problems of sampling procedures.

The sampling procedure can be described using the following terminology.

Population Unit. The population unit is the unit at which the data are collected. In survey research, population units are typically people, although they can also be at other levels, such as households. Other types of research designs may look at larger units of analysis, such as nations, cities, organizations, or institutions.

Target Population. The target population is the group of elements that the sampling procedure aims to represent. For example, a public opinion survey may attempt to represent all adult Americans or may limit the population to individuals with certain qualifying characteristics, such as parents of school-age children, voters, people with heart disease, or residents of South Carolina.

Sampled Population. In contrast to the target population, the sampled population refers to the portion of the population from which the sample is actually drawn. In an ideal world, the sampled population and the target population are nearly the same thing, but for a number of reasons drawing a sample from the target population is usually not feasible. For example, survey data collected using random digit dialing is unable to include respondents who do not possess telephones. Low-incidence target populations exacerbate these issues. For instance, if a researcher wants to do a representative study of people who run for local office, using a random digit dialing technique would be enormously expensive, as many contacts would need to be made to get a single eligible respondent. For this reason, studies of low-incidence populations are often conducted on prescreened individuals (e.g., those from membership lists or clinics). In all cases, it is important to consider any systematic differences between the target population and the sampled population.

Sampling Unit and Sampling Frame. To select the sample from the population, the population must [Page 2348]first be divided into sampling units, or elements that are subject to being selected. The collection of the list of sampling units is referred to as the sampling frame. In a simple random sample design, the sampling frame may consist of a list of individuals who belong to an organization, with the sampling units being each individual. Cluster designs may have a more complex relationship between the sampling frame and the sampling unit, as described in the next section.