Cluster sampling is a specific type of sampling technique which is used on homogeneous populations, involving these target populations being divided into groups which are then randomly sampled. These groups are referred to as clusters. It is all the individuals within a randomly selected cluster which are sampled, not a certain number of individuals from multiple clusters.
Cluster sampling is rather similar to another sampling method called stratified sampling. However, the main difference is that in cluster sampling the divisions are made naturally, rather than being based upon particular attributes as is done with stratified sampling. The term naturally here implies that these divisions are geographically based.
For example: Pretend you have a population of 50 individuals and want a total sample size of 20. You want to interview individuals about the upcoming election and due to the length of each interview and limited resources you only have time to host 20 interviews. This area contains 5 streets and each street has 10 residents.
London Road: 10 individuals, University Avenue: 10 individuals, Main Street: 10 individuals, Gore Street: 10 individuals, Chester Avenue: 10 individuals.
The next step is to number each road from 1 to 5 and then randomly selected two numbers.
London Road: 1, University Avenue: 2, Main Street: 3, Gore Street: 4, Chester Avenue: 5
If 3 and 5 were randomly selected (for example, say you put all five numbers in a hat and randomly selected two) then those 20 individuals would be representative of the sample. This is a simplified example to illustrate how cluster sampling can be utilized.
References:
Image Credit: Elgin Community College. (2014). Other Effective Sampling Methods. Retrieved from: http://faculty.elgin.edu/dkernler/statistics/ch01/1-4.html