From the preceding example, it can be seen that if the modeler is not careful when using transformations, he or she can be tricked into selecting a relatively poor model. This realization becomes especially important when comparing alternative models. Very serious errors can be introduced when selecting the best model unless all comparisons are made with the original data. Otherwise, the choice of best model may be determined by a peculiarity of the transformation rather than on the merits of the model and how well it fits the original data. Although the danger of making transformations is evident in this graphically illustration, a modeler may be fooled if he or she is not especially observant because many computer codes fit models by first making a transformation. If the modeler intends to use indicators such as the sum of the absolute deviations to make decisions about the adequacy of a particular submodel or choose among competing submodels, the modeler must first ascertain how those indicators were computed.