Fitting Response Surfaces with Power Transformations

Summary The role of the likelihood function in statistical inference is briefly discussed, with particular reference to problems where exact probability (structural or fiducial) distributions for the parameters are unknown. The usefulness of non‐linear models in response surface methodology and the difficulties in making exact inferences about these models are briefly reviewed. Likelihood inference is suggested as a method of overcoming these problems, and the procedures for this are developed. An example using biological data demonstrates the procedures.