Issues in the Analysis of Complex Surveys

Data from a survey of a population can be used for two goals: (1) description of the population (i.e. estimation of such numerical characteristics as the total values of a given variable); (2) inference from the population to the process which generated it. Deming (1950, Chap 7) refers, on the one hand, to descriptive (or enumerative) studies, and, on the other, to analytic studies, which are aimed at “the causes of patterns and variations…”. According to Skinner, Holt, &Smith 1989 [henceforth SHS], p.1, “ … the analytic use takes us beyond the summary measures embodied in description and into the causal explanation of the processes that underlie the descriptive measures. ” Thus, analysis is concerned with uncovering the causal relations between things, and the relations between corresponding variables, not only in the population sampled, but in like populations. A derivative, but not entirely coincident, definition of the analysis of surveys, understands the population to be characterized by a statistical model, and the goal to be the estimation of the parameters of the model (SHS, p.5). In recent times, this has come to be the working definition, but it is helpful to keep the primary, causal, definition in mind, and consider issues in its light. 2. Issues regarding Analysis of Survey Dat