Research Concepts

Cluster Samples

One of the main challenges to random sampling is compiling a listing of every member of the population in order to construct your random sample. Often this list is not available, or is incomplete, or has other possible sources of bias.

A different approach is cluster sampling. The population is divided in clusters, often with a geographical basis, and the sample is then selected from the clusters. The initial "sampling unit" is the cluster rather than the individual members of the population.

An obvious example of a cluster is a voting precinct. Instead of randomly sampling from the population of all voters, polling samples are often drawn from a random sample of precincts. Since it is easier to obtain a listing of the approximately 10,000 voting precincts in the US than of all 130 million voters, this kind of cluster sampling is both less expensive and often more reliable than random sampling.

Modern samples generally employ multiphase techniques that combine randomized, stratified and cluster strategies. For example, precincts (clusters) might be selected based on a stratified random strategy by dividing them among rural, suburban and urban. Then within each cluster, stratified random samples of individual subjects might be constructed.