Describing and Modeling Panel Data

Since the panel design observes the same units repeatedly over time, observations at time point t will not be independent of observations made before t. This serial correlation contradicts the assumption of independent observations, which forms the basis of many classical methods of statistical inference. While the independence assumption is easily defended in case of cross-section data, its failure in case of panel data has to be explicitly dealt with. The chapter illustrates these statistical dependencies with simple examples using continuous and categorical variables that have been measured repeatedly over time. Statistical techniques are introduced to describe not only these serial correlations, but also the trend and the change of these variables. These descriptive analyses are followed by a discussion of the sources of these statistical dependencies and how cross-sectional regression models have to be extended to deal with them. The chapter concludes with a discussion of some important specification errors when applying these models to panel data.