Panel data sets, also called longitudinal data sets, are sets of data where the same units (for instance individuals, firms, or countries) are observed more than one time. Models that exploit the specific structure of these data sets are called panel data models. One of the main advantage of using these models is the possibility of appropriately including unobserved variables characterizing individual heterogeneity and heterogeneity of individual decisions. In the last sixty years, panel data and methods of econometric analysis appropriate to such data have become increasingly important in the discipline. Unfortunately, almost all related literature focuses on models assuming that data are free of outlying or aberrant observations. This is often not the case in reality. The majority of the regression methods used in linear panel data models are very sensitive to data contamination and outliers. This doctoral thesis focuses on the estimation of linear panel data models with and without outliers. It consists of two parts. In the first part, some new estimation methods are proposed for static (Chapter 2) and dynamic (Chapter 3) models when data sets are assumed to be contaminated by outlying or aberrant observations. The second part (Chapter 4) is a contribution to the theory of estimation of dynamic models when data are assumed not to be contaminated.
|Qualification||Doctor of Philosophy|
|Award date||10 Apr 2013|
|Place of Publication||Tilburg|
|Publication status||Published - 2013|