Some Aspects of Estimating Statistical Cost Functions

This article analyzes the results of an empirical study for a multistore farm supply cooperative with respect to several criticisms of statistical cost functions. It shows that combining time series and cross section data in an analysis of covariance allows one to avoid the regression fallacy and provides an implicit test for the fallacy, thus avoiding one of the most damaging criticisms of cost curve fitting. Further consideration, however, validates Friedman's criticism that statistical cost curves lack identifiability. It is postulated that Hoch's development of the covariance model for production functions to avoid simultaneous equation bias in certain instances also holds for cost curves. It is argued that this is an empirical question depending on the specific circumstances, and, in general, fitting a statistical cost curve is not a test of those cost curves postulated in the theory of the firm.