A stratified sample is composed of the same potential data points as a purely random sample. The main difference between a stratified sample and a pure, random sample is that the stratified sample has a unique structure. This structure allows the sample to account for high levels of variation within the group sampled. A researcher typically gives structure to a stratified sample by dividing the data into groups that are similar on a specific parameter. There is no requirement for these groups to be of equal size, although some researchers prefer to sample in a way that leads to equal groups, thereby allowing easier data analysis.
The purpose of stratified sampling is twofold. First, stratified samples help reduce error that arises simply because the sample is random. This sampling error can distort the results of the study, sometimes leading to conclusions that are later verified to be incorrect. Second, stratified samples are representative. When a researcher wishes to sample a population that is diverse or contains multiple groups, the researcher may wish to assure that the sample can dependably represent all subgroups. Stratified samples allow the researcher to do just this, as they give researchers the freedom to determine how much of the entire sample should represent each group.
While the advantages of stratified samples are clear, the disadvantages may only become apparent as a researcher begins to structure such a sample. In situations where the population is unknown or has not been previously studied, researchers should find it difficult to decide how to stratify a sample. This is due to the lack of knowledge regarding the diversity inside the sample as well as the possible subgroups of the sample. In addition, deciding the variable on which to stratify the sample can be a subjective process and expose the sample to criticism.
Perhaps a researcher is interested in the proportion of marriages that have experienced affairs. Because the researcher is interested in human nature and not cultural issues, his population is the entire world. He can then stratify the world by nation, thereby yielding a stratified sample in which a certain number of couples from each nation are sampled. This will reduce the overall sampling error and increase the representativeness of the study.