A critical review of some statistical issues implied by the use of multivariate receptor models