Application of Geostatistical and Neighbor Analyses to Data from Plant Breeding Trials
- Tianxia Wu,
- Diane E. Mather and
- Pierre Dutilleul
In plant breeding trials and other experiments with many treatments, there may be spatial effects due to large-scale trends and small-scale autocorrelation. Geostatisticai analysis may be used to investigate the extent of spatially structured variation. When spatial structure is present, neighbor analysis can be superior to classical analysis of variance (ANOVA). The objectives of this study were to assess the validity and efficiency of neighbor analysis and to investigate the extent of spatially structured variation in cereal breeding trials. Three neighbor analysis methods (the Papadakis procedure, the iterated Papadakis method, and the first differences with errors in variables [FD-EV]) were applied to data from uniformity trials and to data from a set of 361 cereal breeding trials. The iterated Papadakis method consistently underestimated the variance, producing tests with highly inflated Type I error rates. Thus, its relative efficiency could not be estimated correctly. Geostatistical analysis indicated that spatially structured variation was frequently present in the cereal breeding trials, and that first differencing was effective in removing it. The FDEV analysis consistently improved the accuracy and precision of the estimation of entry effects compared with classical analysis of variance and Papadakis analysis. Efficiency relative to classical analysis of variance averaged 152% for FD-EV and 116% for the Papadakis procedure. Considering both validity and efficiency, FD-EV was the best method. In the presence of spatially structured variation, FDEV can improve the interpretation of data from field trials. In the absence of spatial structure, FD-EV causes no loss of efficiency.Please view the pdf by using the Full Text (PDF) link under 'View' to the left.
Copyright © . .