Spatial Analysis Improves Precision of Seed Lot Comparisons

Abstract

Spatial analysis, by least squares smoothing, was compared with randomized block analysis of data from field trials of 50 seed lots of spring wheat (Triticum aestivum L. cv. Katepwa) and of 40 seed lots of spring barley (Hordeum vulgare L, cv. Harrington). Spatial analysis reduced standard errors (SED) of differences for grain yield, but not for stand density. For grain yield, least squares smoothing reduced the SED by 11 to 53% in four spring wheat trials and by 24 to 34% in three spring barley trials. Improved precision is expected to affect correlations between trials. For spring wheat, the inter‐trial correlation increased from 0.40 to 0.60 in one trial, from 0.03 to 0.41 in another, and had little effect in the remaining four of six pairs of wheat trials. For barley yield, the inter‐trial correlation increased from 0.25 to 0.36 in one of three pairs of trials, and decreased from 0.28 to −0.06 or from 0.22 to −0.05 in the other two. In the spring wheat trial with largest spatial variability, least squares smoothing, first‐order autoregressive residuals, and the iterated Papadakis' methods gave similar reductions in SED (53–63%) and adjusted means that were highly correlated (r > 0.94). Row‐column analysis gave little reduction in the SED. A multiplicative model reduced the SED, but adjusted means were poorly correlated with those of other methods (r > 0.78). Simulation showed that an incomplete block analysis could have provided nearly the same improvement in precision as spatial analysis. Our results confirm that spatial or incomplete block analyses can improve the efficiency of field trials.