Correlation between residuals and chemical class

During inspection of the residuals from linear regression, it was noticed that there was some correlation between residual value and the chemical class of solute. Each observation can be categorised as pertaining to a sugar, acid, small molecule or other molecule. These classifications reflect the composition of the source data. 

Therefore, the difference between chemical classes appears to be important. The results show that for sugars, observed values are smaller than predicted this applies to all models given herein. Before considering this further, we should consider whether this problem (or a similar one) exists within the Wilke-Chang model [3]. 

There are several similarities to the results of this work. In log form, the Wilke-Chang equation is 

This relationship is clearly related to equation (15.18) in particular. The term in log is of note. In the original reference [3] very limited justification for its inclusion was made. Furthermore, the values of quoted in Reference [3] were obtained from then unpublished work, and effectively have the nature of an empirical correction, which in itself is not a major problem in the utility of the model. However, when the work was published the values of the parameter were quite different [12], yet this error in the Wilke-Chang equation has not been corrected in the literature, to the authors knowledge. 

The importance of the factor is of interest this is an adjustable parameter for each solvent, considered by Wilke and Chang to account for the aggregation of the solvent molecules. In fact, it is likely that the effect of aggregation (mutual association of solvent molecules) has, in fact, already been accounted for in the viscosity term r B. So what is the significance of the value In their data set, for the classes sugar , acid and small , water is usually the solvent, which has greater than 1. Therefore the value artificially increases the predicted D, which would otherwise be predicted too small (and appears to make physical sense only if it is considered as a correction to the viscosity term). 

---