Accounting for Spatial Yield Variability in Field Experiments Increases Statistical Power

Parametric statistical techniques evaluate treatment significance in field experiments by comparing variability attributed to treatments to variability attributed to random error. In many experiments, a considerable amount of the variability attributed to random error is actually due to large‐scale soil variability that cannot be accounted for by blocking. This variability can, in part, be accounted for by a technique called nearest neighbor analysis, thus reducing the amount of variability attributed to random error; variability attributed to treatments is then larger in comparison, and the statistical significance of treatment effects is increased. Our objective was to evaluate the utility of nearest neighbor analysis in the statistical analysis of a set of field experiments. Four experiments with fall N treatments on winter wheat ( Triticum aestivum L.) were analyzed using analysis of variance (ANOVA). According to this analysis, treatment had no significant effect on yield in any of the four experiments. After nearest neighbor analysis was used to remove spatial yield variability from the random error term, ANOVA revealed statistically significant treatment effects in two of the four experiments. Accounting for spatial variability is a practical way to increase the power of ANOVA and accompanying means‐separation techniques when analyzing data from replicated field plot experiments.