PRACTICAL USE OF NEIGHBOR METHODS AND ALTERNATIVES IN THE ANALYSIS OF FIELD TRIALS

Abstract

Neighbor methods for the analysis of field experiments are described with the minimum of mathematical detail. Using actual field data, two popular methods are compared with standard blocking, row and column elimination, and covariance analysis. It is shown that simple randomized block analysis is likely to give biased treatment effects when there is a fertility trend. For large experiments, row and column analysis is likely to be inferior to covariance analysis using well-chosen covariates. Neighbor analysis can be more precise than standard methods based on well-designed experiments with appropriate blocking. Situations where neighbor methods may not work well are suggested, but since the methods are easy to use they are recommended when fertility trends are suspected and a simple detective tool is required.Key words: Neighbor analysis, covariate analysis, lattice design, fertility trends