INTRODUCTION

Cluster sampling, as described Chapter 2, is a sampling technique in which all the units of a selected cluster are included in the sample. Because units in a cluster tend to be related to each other, the utility of a cluster sample reduces. Selection of all the units from a cluster, particularly if it is large in size curbs the ability to spread a sample and capture larger variation from the population under study. However, as we will see later, the effect of clustering tends to be inversely proportional to the cluster size. In other words, smaller the size of clusters, greater is the intra-cluster correlation. For example, consider clusters consisting of four or five consecutive houses in a village or residential flats in an urban dwelling unit. Because of the closeness between units in an area, it is expected that their social contacts and interpersonal communications will be higher among them and hence, apart from some socio-economic characteristics, many of their behaviours, attitudes, etc. are also likely to be similar giving rise to intra-cluster correlation among the individuals.

While in cluster sampling it is imperative to select all units within a cluster, it is often not viable to do so and as seen above, selecting all units from the cluster does not necessarily constitute an efficient sample. One alternative to taking all units from a selected cluster is to take subsamples from the selected clusters. That is, instead of giving all units of a selected cluster certainty of inclusion in a sample, one can select few units from each of them. Thus, in this case, selection of units is done in two stages – first selection of clusters from a list of clusters and then within each selected cluster selection of population units. In fact, this design, where the selection of samples is done in more than one stage, is widely used, particularly in large-scale population-based surveys. Generally, the administrative units existing in a country such as states, districts in a state, towns/villages in a district, etc. are used as sampling units at first stage of selection. These areal units have well-defined boundaries, each and every individual person or household can be uniquely identified as a resident of an area, and use of such units can also help in obtaining a sample that gives each resident a chance of being included in a sample.