Econometric Methods for Nonlinear Panel Data Models

[709598]面板数据集在经济学几乎所有领域的实证工作中都发挥着重要作用。在面板数据中，每个横截面单元都有不止一个观测值，因此希望考虑到个体的特定影响。例如，在最典型的面板数据情况下，可以在多个不同的时间段观察一个人。该项目开发了新的计量经济学方法来处理非线性面板数据模型。考虑的模型类型包括感兴趣的变量是“离散”的模型。例如，一个人可能有兴趣调查一个人是否在给定的时间内购买了一种特定的产品，或者一个公司是否在出口。该项目还考虑了仅部分观察到感兴趣的变量（“删减”）的模型。例如，如果使用社会保障记录来研究收入，那么对于收入高于社会保障最高限额的个人来说，确切的收入是不知道的。本项目开发的方法将放宽一些为了应用现有方法而必须保持的假设。具体来说，对于有审查的模型，以前需要假设在所有时间段内未观测误差分布的分布是相同的。在许多应用程序中，这种假设过于严格。该项目还将放宽对解释变量所作的假设。在许多应用程序中，很自然地允许一个时间段内的解释变量与感兴趣的变量的值相关的可能性。例如，一个女人是否决定工作可能取决于她是否有小孩，而小孩又可能取决于这个女人以前是否工作过。要使用现有的方法，必须假设这样的“反馈”不存在。本项目开发的方法将允许这样的反馈。?？

9709598 Honore Panel data sets play an important role in empirical work in almost all areas oaf economics. In panel data, there is more than one observation for each cross sectional unit, and it is desirable to allow for an individual specific effect. For example, in the most typical panel data situation, an individual is observed in a number of different time periods. This project develops new econometric methods for dealing with non-linear panel data models. The types of models considered include models in which the variable of interest is "discrete." For example, one might be interested in investigating whether an individual buys a particular product in a given time period, or whether or not a firm is exporting. The project also considers models where the variable of interest is only partially observed ("censored"). For example, if one studies earnings using social security records, then the exact earnings are not known for individuals whose earnings are above the social security maximum. The methods developed in this project will relax some of the assumptions that must be maintained in order to apply existing methods. Specifically, for models with censoring, it was previously necessary to assume that the distribution of the unobserved error distribution is the same in all time periods. In many applications, this assumption is too restrictive. The project will also relax the assumptions that are made on the explanatory variables. In many applications, it is natural to allow for the possibility that explanatory variables in one time period are related to the value of the variable of interest. For example, a woman's decision to work may depend on whether or not she has a small child, which, in turn, may depend on whether the woman worked in previous time periods. To use existing methods, one must assume that such a "feedback" is not present. The methods developed by this project will allow for such a feedback. ??