 |
|
|
|
|
|
Simple random or systematic random sampling does not have these disadvantages. Either method generally is safer to use when the proportions of the strata of some population are not known. Stratified sampling designs are frequently used when the investigator wishes to compare various groups on some set of variables. Hassan and others (1988) selected separate samples of males and females from each of two levels of schooling. Gender represented one strata; level of schooling another. Mageed, Sulima and Kawther (2000) also used a stratified sampling design. They selected separate samples from two contrasting socio-economic areas of Omdurman for their study of male attitudes toward female genital mutilation. Muneer's (1989) sample was based on whether or not farmers were members of an agricultural cooperative. Khamis and Alsuni (1998) also based their study on a stratified sample. They selected separate samples within each of three urban and three rural areas for measuring the prevalence of tuberculosis. In each of these investigations, data analysis consisted of comparing the results from the separate samples used in each study. Cluster sampling Each of the three sampling methods described so far requires use of a sampling frame. These methods also work best when the population is in the low thousands or smaller, and when the population is reasonably accessible or concentrated in small areas. But what can be done when there is no sampling frame or when the population is very large and scattered over a wide area? Then, the methods described previously are impractical. To get around this problem, researchers have developed another sampling method, called the cluster, multistage or area sampling. The idea behind this method is quite simple. Instead of selecting the ultimate sampling elements (usually households or individuals) right off, a several-step process is used. First, the investigator defines a set of large clusters that together cover the target population. Clusters often are geographic areas, such as villages, regions of a rural area, sections of a town or city, or physical units such as schools or hospitals. Each cluster defines some part of the target population. Depending on the size of the ultimate sample wanted, some number of these clusters is selected at random. Each of these first order clusters is subdivided into small clusters and some number of these secondary clusters is randomly selected. Generally, only first and second order clusters are used. In rare circumstances, a third set of clusters might be needed. These choices depend on how large a sample is wanted and how widely the population is spread over an area. With a larger sample and greater population dispersion, more clusters might be required. Using a simple random or systematic method to randomly select sampling elements from the last set of clusters completes the process. Selecting A map of the area to be sampled is usually necessary for designing a cluster sample. If available, aerial maps provide a good basis for creating clusters. Lacking any map, investigators often draw their own maps, based on driving and walking around the areas where clusters are to be defined. If the purpose were to design a sample of households in a certain section of a city, the area could be inspected and houses noted on a map, which could then be used in selecting clusters. This approach, however, will work only for relatively small areas. For large areas, such as provinces or large cities, maps are essential. Cluster sampling requires making decisions about the number and sizes of clusters to use at each stage of sampling. A good way to start is to decide on the size of the final sample. Suppose that you think you have enough time to interview 150 persons in a city of approximately 15,000 persons. To get this sample, you will need to make three decisions:
|