ITERATED NEAREST-NEIGHBOR ANALYSIS OF FIELD EXPERIMENTS

Four iterated methods of analysis of agronomic trials were tested by simulation. Complete blocks and totally random analysis were also performed. The experimental trials were simulated with 4 different maps for soil fertility trend, added to controlled genetic values randomly placed within the trial and random values for error. Iterated spatial analysis proved more efficient than block analysis when the natural trend was irregular. Estimated residual variances were far lower than block analysis estimations but at no time did they they underestimate the introduced value as previously stated by some authors. It was found that unbiased estimations of genetic variances were obtained. Estimations from spatial analysis were less sensitive to reduction of block number. Among the 4 spatial methods tested, the iterated Papadakis method, using the average of nearest neighbour residuals as covariate, provided more information regarding the heterogeneity trend of the trial.