Intrablock variance among duplicate treatments for nearest-neighbor analyses

Nearest-neighbor analysis (NNA) adjusts for spatially correlated residuals, with the goal of increasing precision. The magnitude of the block x treatment interaction mean square is commonly used to evaluate the precision of the NNA model. An alternative method of evaluating the precision of the NNA and classical unadjusted (UNADJ) randomized complete block (RCB) analysis would be to use the pooled variance between duplicate treatments within each block. We defined pure error as variation between plots that are treated alike within a block. Within each location, each genotype was randomly assigned to two plots within each block of an RCB design. The pure error of soybean [Glycine max (L.) Merr.] genotypes was evaluated at eight locations. Our objective was to compare the block x treatment and pure error mean squares for yield, physiological maturity, and plant height to determine whether the NNA or UNADJ analysis reduces intrablock variation. The NNA analysis always decreased the magnitude of the block x treatment interaction mean squares, compared with the UNADJ analysis. In some comparisons, the pure error mean square of the NNA analysis was significantly smaller than the pure error of the UNADJ analysis. The magnitude of the block x treatment mean square is not useful for comparing the relative precision of these two analyses. When the pure error mean square was used to measure precision, the NNA was at least as precise as the UNADJ analysis.