Use and misuse of multiple comparisons in animal experiments.

The objective of many animal experiments is to detect meaningful relationships among treatments and associated responses. Types of comparisons of means include pairwise multiple comparisons, planned orthogonal or nonorthogonal contrasts, and orthogonal polynomials. Some procedures are appropriate only for specific types of treatment designs and specific types of objectives. Pairwise, multiple comparisons are appropriate only for comparing unstructured, qualitative treatments. Planned comparisons partition the overall set of treatment effects into independent or nonindependent subsets, with special application to factorials. Orthogonal polynomial (regression) procedures assess relationships between quantitative treatments and response when a full range of responses or an optimal dose is of interest. Recommendations for appropriate use of each mean comparison procedure are illustrated using data from three Journal of Animal Science articles. Also mentioned are a number of computer graphics packages that provide creative ways to display biological relationships and can be linked to statistical packages for input and to word processors or 35-mm cameras for output.