On the Theory and Practice of Obtaining Unbiased and Efficient Samples in Social Surveys

Analysis of statistical issues in the design of large-scale social experiments where economic phenomena are important, such as experiments in health insurance, income maintenance, and electricity peak-load pricing. Drawing on RAND experience in designing and implementing the HEW Health Insurance Study, the authors address four problems that are not adequately treated in classical experimental design: (1) defining the sampling frame when repeated sampling is attempted and part of the population is transient, (2) choosing optimal survey samples, (3) allocating subjects to treatments, and (4) balancing covariates when subjects are assigned to treatments in the presence of field constraints. The purpose of the report is to pass on lessons learned in the experimental portion of the Health Insurance Study, and to encourage systematic research on theory and methods needed for design in complex field experiments.