Sampling From Structured Populations: Some Issues and Answers

Sample surveys have played an increasingly important role in the Bell System in recent years as a means of providing an objective basis for decision making. To an extent, this has been due to the growing awareness among users of the survey results that, in most surveys, sampling is not the only source of error and often not the primary source. Even if a presumably complete census were taken instead of a sample, serious errors might exist in the results arising from various causes such as measurement or response errors. * Brown University. 1235 The growth in numbers, in recent years, has also been accompanied by a widening of the range (both in type and complexity) of the surveys. For many of these surveys, a simple and readily available sampling design can easily be adapted to the needs of the prevailing situation. More often, however, the problem at hand is sufficiently complex and nonstandard so that various parts of existing sampling theory have to be modified and pieced together to arrive at a reasonable solution. Nevertheless, some sampling issues are common to a number of Bell System surveys. Most of these surveys involve sampling from populations that are highly structured, and any cost-efficient sampling design must take this structure into account. In this paper, we review some sampling issues that arose in two surveys currently under implementation. Both surveys possess some common features as well as features unique to themselves. Since these features are common to a large number of other surveys, an exposition of both the theoretical and practical considerations involved may prove beneficial to other survey practitioners.