A predictive model is sometimes evaluated by plotting its predictions directly against the corresponding experimental data in an (x, y) plot: if the model predicts the data adequately, a regression line ?t should be a 45 degree line with slope 1 and intercept 0; the residuals should appear as a random sequence of numbers that are independent, normally distributed, and with zero mean and variance close to measurement error variance. This technique is to be used to evaluate two models of multicomponent transport as follows.

In Kerkhof and Geboers, (2005)7 , the authors presented a new approach to modeling multicomponent transport that is purported to yield more accurate predictions than previously available models. The table below shows experimentally determined viscosity (10-5 P a.s) of 12 different gas mixtures and the corresponding values predicted by two models: (i) the classical Hirschfelder-Curtiss-Bird (HCB) model8 , and (ii) their new (KG) model.