Chapter 6 Some Multivariate Methods 281 mean and covariance matrix that increases statistical efficiency. A consistency factor is used to make the estimator unbiased at normal distributions. 6.14.5 Method OP Method OP simply removes any outliers detected by the projection approach described in Section 6.4.9. Then the classic PCA is applied to the data that remain and the p-dimensional representation of the data is computed in the usual way. Croux and Ruiz-Gazen (2005) suggest an algorithm that begins with projections based in part on the L 1 median, but it is evident that their approach differs from method OP. Method OP attempts to eliminate outliers in a manner that takes into account the overall structure of the data. The algorithm used by Croux and Ruiz-Gazen does not do this, but rather searches for projections that maximize a robust measure of scatter applied to the marginal distributions of the scores. Also, Croux, Filzmoser, and Oliveira (2007, pp. 6­7) note that the Croux and Ruiz-Gazen (2005) and Hubert et al. (2002) projection algorithms suffer from severe downward bias. It is unknown whether method OP suffers from the same problem.