sampling error

The principal aim of any sampling procedure is to obtain a sample which, subject to limitations of size, will reproduce the characteristics of the population being studied, especially those of immediate interest, as closely as possible. In practice two types of error will arise from any sampling procedure: first, sampling bias may arise in the way the selection is carried out; and second, random sampling error may arise in the sample obtained, due to chance differences between the members of the population included, or excluded, from the sample. Total sampling error in the sample issued for interviewing consists of these two taken together. The key difference between the two is that random sampling error decreases as the sample size is increased, whereas sampling bias is not eliminated or reduced in that way: it is a constant characteristic unless steps are taken to improve the quality of sample selection. An important source of sampling bias is a sampling frame (a list of the members of the total population of interest from which a sample for study is to be drawn) which does not in fact cover all of the intended population. For example, there may be systematic differences between people who do or do not enter themselves on the electoral register or own a telephone, so that lists of such persons are not completely representative of the adult population. Another source of bias is random sampling that is not in practice completely random, because the lists and records used as the sampling frame are not put together randomly, but presented in some systematic manner not known to the researcher using them.

After interviewing is completed, non-response bias may be discovered in survey results. The combination of sampling error and survey non-response bias together determine the representativeness of the survey data produced by the study.