ABSTRACT
In the last five decades there has been a significant growth of research in econometric methods and their application in various areas of economics. Indeed, in the last two decades, the tremendous growth in econometrics has dichotomized the subject into cross-sectional (micro) econometrics and time-series (macro) econometrics. Whereas the new cross-sectional methodology was partly due to the nature of the data and the empirical issues in microbased labor economics and industrial organization, the new time-series methodology was an outcome of the challenging empirical issues a~d data problems in macroeconomics and finance. Despite these developments, econometric inference methods (especially in cross-sectional econometrics) have been confined to the assumptions that data is generated as a simple random sample with replacement or that it is coming from an infinite population (Johnston 1991, Greene 1993). These assumptions are certainly not valid in the case of survey data used in development and labor economics. Surveys usually have a well-defined frame consisting of a finite population of individuals, households, or villages. Sample data for analysis is generated from this finite population using a sampling design different from random sampling with replacement (RSWR). Sampling schemes such as systematic sampling, stratified random sampling, and cluster sampling may be used alone or in combination. These have been the subject of four decades of extensive work in statistics literature (Kish 1965, Cochran 1953, Sukhatme 1984, Levy and Lemeshow 1991, Thompson 1992).
