(Division of Research Techniques, London School of Economics and Political Science)
Random Sampling and Quota Sampling In the sampling of human populations the selection of the final sample units is usually carried out by one of two methods: random sampling or quota sampling: Of the former there are many variants, ranging from pure random selection (by random numbers or the lottery method) to more or less systematic selection from some complete record of the population. Area sampling, predominantly used in the United States, is a form of random sampling. Whatever the details of the sampling designs or frames employed, one characteristic is-or should be-common to all random sampling. The selection of the sample units is carried out by some impersonal (strictly determined) method and is uninfluenced by human choice. That is to say, the interviewers are not allowed any freedom in deciding which members of the population shall be included in the sample. This independence of selection from human judgment is a prerequisite for ensuring that every member of the population shall have a known chance of being included in the sample-which is what we mean by randomness. Quota sampling differs from random sampling in several minor ways, but the fundalnentdl difference is that, once the general breakdown of the sample is decided (e.g., how many men and women, how many people in each age-group it is to include) and the quota assignments allocated to eacb interviewer, the choice of the actual sample units to fit into this framework is left to the interviewer. Much discussion has centred around the merits of the two techniques. Some experts believe the quota method to be so unreliable and prone to bias as to be almost worthless; others think that, although not as accurate as random sampling, quota sampling can be used safely on some types of inquiry; while some believe that, if careful instructions are given and if sufficient constraints are imposed on the freedom of the interviewer, quota sampling can be made highly reliable, and that the heavy extra cost of random sampling does not result in a sufficient increase in accuracy to be worth while. In general, academic statisticians have criticized the method for its theoretical weakness, while market research workers have defended it for its cheapness and ease of practical application. This issue is not by any means the most important problem in survey methodology, nor even perhaps in the more narrow field of sampling. Yet the fact is that the controversy has continued -most fiercely in the United States-for many years, largely unaided by any experimental evidence, and that the argument is still conducted mainly on the basis of prejudice and untested assumptions. There is no clear-cut answer to the problem, in the sense that either method could be shown to be preferable in all circumstances. There are some types of survey on which nobody would suggest using a quota sample, while there are others on which random sampling may be impracticable. Still, there remains a large field on which either method could bs used, and research is required to throw factual light on the merits of both techniques. It therefore seemed worth while to embark on a programme of reasearch into quota sampling, in the hope that the results will be of value and interest both to statisticians generally and to market research practitioners. The present report represents the first, preliminary, stage in the research. It consists of three parts. The remainder of the present part is devoted to an examination of the usual cases for and against quota sampling, followed by a statement of the general aims of this research programme. The second and major part is a description of current quota sampling practice in this country.