On the assessment of multivariate measurement systems