Specific functional forms are often used in economic models of distributions; goodness-of-fit measures are used to assess whether a functional form is appropriate in the light of real-world data. Standard approaches use a distance criterion based on the EDF, an aggregation of differences in observed and theoretical cumulative frequencies. However, an economic approach to the problem should involve a measure of the information loss from using a badly-fitting model. This would involve an aggregation of, for example, individual income discrepancies between model and data. We provide an axiomatisation of an approach and applications to illustrate its importance.