Abstract
The results of a large numerical study of the classical and of robust estimators of the model with fixed and random effects are presented together with the graph of hausman test of specificity. It reveals that for a low or mild level of contamination the robustified versions of the classical within-groups and generalized least squares estimators can work well.
For higher level of contamination, say over 5\%, it is better to rely on the smoothly robustified OLS, as they give the results which are closer to the true value that the robustified classical estimators. One reason is that the robust estimation of location - which w need for robustified within-group estimator - need not be easy task to estimate because the ''global'' level and character of contamination can significantly differ from the local one within the groups.