Experimental Design and Analysis with Emphasis on Communicating What Has Been Done: (II) Calculating Interaction Contrasts with SAS

Abstract

Appropriate statistical analyses are of primary importance for understanding experimental results. Shim et al. (2014) detailed the influence of different statistical approaches on results of a two factor nutrition experiment with broiler chickens. However, frequently designs with more than two factors are needed because of the complexity of modern broiler and egg production. Statistical analyses need to be clearly communicated so that readers can properly interpret the results of experiments with poultry. Designs with two or more factors are frequent players in the world of experimental design. The computational burden of the attendant analysis of variance is somewhat eased by the presence of statistical packages. Contrary to expectation, it is not clear from texts or the Manual (s) how the package(s) can be used to find components of the interaction effects, whether the factors are qualitative or quantitative factors. We show how SAS can be persuaded to calculate these components (A x Linear B, etc., when A is a qualitative and B is a quantitative factor and Linear A x Linear B, etc., when both A and B are quantitative factors). The procedure can be adapted to fit other packages which have provision for contrast calculations. The results presented here extend and clarify the analyses of Shim et al. (2014) on the advantages and disadvantages of various techniques for analyzing results from experiments on poultry with more than two factors.