Inference and Non-Random Sampling

点击翻译按钮获取中文摘要

A common strategy in econometric analyses is to divide the variables of interest into two types. The first type is that of a dependent variables whose distribution in the population is to be explained in terms of variables of the second type, the independent or explanatory variables. If the data to be analyzed were randomly selected from the population of interest, then methods exist for estimating the parameters of the conditional distribution of the dependent variable given the explanatory variable. These estimators are consistent, efficient and have other desirable properties. But some of the most important properties do not hold if, as is frequently the case in applied economics, the data are not random. This project develops estimation procedures and specification tests that are both intuitive and computationally feasible for nonrandom or partially random sampling procedures that are frequently encountered in economic problems. These techniques are used to analyze partially random data sets on Dutch labor market participation. This project addresses the question of inference if the researcher has information on the parameters other than a random dataset. Examples of sampling schemes studied by this project include: 1) samples that are endogenously stratified; 2) additional information on moments of the distribution of the dependent variables from aggregate sources; and 3) datasets containing different sets of variables. The application of these techniques to actual problems is used to show that they are feasible and that they can lead to major efficient gains.