Cluster Sampling

A variety of sampling strategies are available in cases when setting or context create restrictions. For example, stratified sampling is used when the population's characteristics such as ethnicity or gender are related to the outcome or dependent variables being studied. Simple random sampling, in contrast, is used when there is no regard for strata or defining characteristics of the population from which the sample is drawn. The assumption is that the differences in these characteristics are normally distributed across all potential participants.

Cluster sampling is the selection of units of natural groupings rather than individuals. For example, in marketing research, the question at hand might be how adolescents react to a particular brand of chewing gum. The researcher may access such a population through traditional channels such as the high school but may also visit places where these potential participants tend to spend time together, such as shopping malls and movie theaters. Rather than counting each one of the adolescents’ responses to a survey as one data point, the researcher would count the entire group's average as the data point. The assumption is that the group data point is a small representative of the population of all adolescents. The fact that the collection of individuals in the unit serves as the data point, rather than each individual serving as a data point, differentiates this sampling technique from most others.

An advantage of cluster sampling is that it is a great time saver and relatively efficient in that travel time and other expenses are saved. The primary disadvantage is that one can lose the heterogeneity that exists within groups by taking all in the group as a single unit; in other words, the strategy may introduce sampling error. Cluster sampling is also known as geographical sampling because areas such as neighborhoods become the unit of analysis.

An example of cluster sampling can be seen in a study by Michael Burton from the University of California and his colleagues, who used both stratified and cluster sampling to draw a sample from the United States Census Archives for California in 1880. These researchers emphasized that with little effort and expended resources, they obtained very useful knowledge about California in 1880 pertaining to marriage patterns, migration patterns, occupational status, and categories of ethnicity. Another example is a study by Lawrence T. Lam and L. Yang of duration of sleep and attention deficit/hyperactivity disorder among adolescents in China. The researchers used a variant of simple cluster sampling, two-stage random cluster sampling design, to assess duration of sleep.

While cluster sampling may not be the first choice given all options, it can be a highly targeted sampling method when resources are limited and sampling error is not a significant concern.