Key points are not available for this paper at this time.
Abstract When the effects of two factors are simply additive, choice of estimates can be made entirely on grounds of minimal variance. When interaction exists, estimation of the magnitudes of main effects and interactions requires an arbitrary definition of what constitutes a main effect. Whether the class frequencies be equal, proportionate, or disproportionate, the definition must conform to the use that is to be made of the results and must not be dependent merely on the convenience of an analysis of variance.
D. J. Finney (Wed,) studied this question.