Under the stimulus of some of the newer methods of experimentation there is a decided tendency toward the grouping of classes of data into smaller and more homogeneous sub-classes. The weights of swine, for example, may be simultaneously classified according to the sex as well as the litter of the individual animals. Corn yields may be entered in a three-way table by applying the criteria of variety, treatment and soil type. From the resulting tables of multiple classification can be derived information not only of the main effects, such as sex and litter, but also of the interactions between them. Analysis of variance is the most convenient and effective method of reducing such classified data to summary form and testing the significance of the various effects.
Snedecor, George W. and Cox, Gertrude M.
"Disproportionate subclass numbers in tables of multiple classification,"
Research Bulletin (Iowa Agriculture and Home Economics Experiment Station): Vol. 15
, Article 1.
Available at: https://lib.dr.iastate.edu/researchbulletin/vol15/iss180/1