GENERALIZED PROCRUSTES ANALYSIS WITH ITERATIVE WEIGHTING TO ACHIEVE RESISTANCE

The paper studies the generalized Procrustes problem with the extensio n of variable weights. A distinction is made between configurations an d configuration classes, depending on the type of data used. The main interest is in configuration classes. Two different types of weighting are introduced: the Huber and Tukey weights. The objective is to use the weights to yield a resistant procedure for the generalized Procrus tes problem. An algorithm is described and some attention is paid to t he starting values in order to avoid non-optimal solutions. An example and simulation results are presented which show that the weighted app roach outperforms least squares when there are outliers in the data.