Summary

Statisticians strongly recommend balanced, orthogonal designs because their analysis lacks several complexities and ambiguities (marginality and order dependence) that arise when these desirable properties are absent (as in most observational data sets). When these complexities are present, some careful forethought can save a lot of work and help to lay out an efficient route through the analysis that’s consistent with the original goals and minimizes ambiguity. This includes when the focus is on prediction or on hypothesis testing, explanation, and understanding.