摘要：Many empirical studies of household behavior in developing countries rely on probability sample surveys. Provided the sampling is random, such surveys reduce costs of gathering data while allowing valid inferences of the characteristics of the underlying population to be made. A prerequisite for the drawing of a random sample is a sampling frame, a list of the units in the population from which the units that will be enumerated are selected. In practice, this is often an actual list, a set of index cards, a map, or data stored in a computer. But unlike sampling issues such as the choice of sample size or the mechanism for randomly selecting units, the construction of the sample frame rarely receives much attention. This is unfortunate. For example, a sample frame that excludes the poorest households in a locality will lead to biased inferences regarding the incidence and severity of poverty in that community, irrespective of the quality of the data collection or the sophistication of the subsequent statistical analysis.
