More generally, methods such as singular value decomposition and principal components analyses have their limitations--they can work fine for balanced data such as in this example but in more complicated problems I would go with item-response or ideal-point models.