ABSTRACT
In traditional psychophysics a statistical model of relatively simple linear form, with attendant gaussian error, is fitted to data by some variant of the method of moments. Refinements, such as the classical Müller-Urban weights ( see Woodworth, 1938, chapter XVII ) or modern partly Bayesian treatments on model inexactness ( Box, 1980) are available, but without it being possible to construct likelihood functions a rigorous treatment of goodness-of-fit questions is generally assumed to be lacking.
