This research aims to study the performance of a sample of linear static panel data estimators. Many panel data estimators are incorporated in econometric packages, readily available to researchers. They use different assumptions to address different aspects of dependencies in the errors, with implications for the accuracy with which both the coefficient and its variance-covariance matrix are estimated. Researchers find themselves confronted with difficulty when choosing specific estimators that best suit the data they are modelling, since no clear guidance exists to facilitate this choice. Reed and Ye (2011) attempted to provide such guidelines. After conducting Monte Carlo experiments on 11 linear static panel data estimators, they proposed a set of recommendations to be used by other researchers. In this Thesis, we first replicate their results after addressing a flaw identified in their experimental design. We improve the formulation of the original recommendations and show that the new recommendations are robust to the errors’ parameters. We further use bootstrap techniques to investigate the Parks (1967) estimator, which is most efficient for certain-sized data sets, but which also has poor test size performance. The bootstrap techniques effectively remove the size distortion related to this estimator. Lessons from these two chapters are then used to model the contribution of health and education spending to economic growth in a sample of 12 African countries observed from 1999 to 2013. We find that neither public spending on education nor that on health has a significant impact on the growth rate of per worker real GDP.