Analysis of Factorial Experiments by Estimating Monotone Transformations of the Data

Abstract

Summary Data may sometimes be described by a linear model with fewer terms if a transformation is applied. For example, in a factorial design the need for certain interaction terms may be eliminated by transformation. If attention is restricted to all monotone transformations, it is possible and perhaps useful to find the monotone transformation which minimizes the residual sum of squares after removing the effects explained by the model. Several sets of data from the statistical literature are analysed this way.