Controlled sampling for comparing two populations - An example
When the objective of a study is to compare with respect to one variable two populations that are inherently different along related dimensions, a standard probability sample is not always appropriate since there is no control over extraneous variables that may be related to the key indicators of interest. This paper discusses a sample designed to study the effect of ethnicity on drinking behavior by comparing Irish and Puerto Ricans in New York City. A random sample of both populations would have almost certainly produced an Irish sample with a higher average socioeconomic level than the Puerto Rican sample. Furthermore, since the total sample size was only 1,000, limiting the analysis to the comparable subpopulations would have resulted in unacceptably small effective sample sizes. Controlled sampling was used to favor selection of areas which according to census data contained members of each ethnic group of similar socioeconomic background I . The paper concentrates on the preparation of the sample frame which is the innovative aspect of this design.