Two methods of robust estimation of a covariance matrix - A practice study for some liver data

We consider two methods of computing robust covariance matrices: the hybrid method proposed by Rocke and Woodruff (1996) and the Fast-MCD method proposed by Rousseeux and van Driessen (1999). We compare the obtained robust covariance matrices both analytically and graphically. The evaluations are done using some medical data. The comparison of the obtained matrices for these data shows, that the two robust methods give systematically slightly differing results, in particular: The MCD method points to more outliers than the Hybrid method proposed by Rocke and Woodruff.