Theoretical and practical issues in estimation of nonlinear panel data models

点击翻译按钮获取中文摘要

Panel data sets play an important role in empirical work in almost all areas of economics. Formally, panel data sets are data sets where there is more than one observation for each unit of observation. Typically, this is because the data set consists of observations for the same individual over a period of time, but the methods for dealing with panel data are equally applicable to situations where the unit of observation is firms, municipalities, etc. This project develops new econometric methods for dealing with nonlinear panel data models, including duration models. The project also develops software to implement these methods. The advantage of panel data is that by following an individual over time, it is possible to investigate an individual's dynamic behavior while controlling for that individual's characteristics. It is easiest to illustrate the importance of this by considering the following simple example. Suppose, as is often the case, that one observed that individuals, who use a certain social program in one period, are also more likely to use it the next period. This ``persistence'' could be observed because some characteristic of the individual makes him or her more likely to use the program. If that characteristic does not change much over time, then the same people will tend to use the program in different time periods. On the other hand, the persistence is also consistent with a situation in which using the program causes the individual to be more likely to use the program in the future (for example, because the individual learns about the program by using it). Being able to distinguish between these explanations of the persistence is important because under the second explanation, a policy that changes the availability of the program in a single period will also have an indirect effect on participation in the program in future periods. This project develops econometric tools for dealing with panel data in situations like this one, although the tools are more generic and hence applicable in many other empirical situations as well.