Premium
Methods for Combining Data from Gemrplasm Evaluation Trials 1
Author(s) -
Hill R. R.,
Rosenberger J. L.
Publication year - 1985
Publication title -
crop science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.76
H-Index - 147
eISSN - 1435-0653
pISSN - 0011-183X
DOI - 10.2135/cropsci1985.0011183x002500030009x
Subject(s) - best linear unbiased prediction , statistics , mathematics , variance (accounting) , econometrics , biology , selection (genetic algorithm) , computer science , accounting , artificial intelligence , business
Seven different methods of combining total season yields from a series of alfalfa ( Medicago sativa L.) germplasme valuation trials at a single location were evaluated. Our objective was to find the “best” method for estimating mean yields for the cultivars and experimental ines included in a series of trials that did not contain all entries in equal numbers. The methods of estimation included percent of checks, summation of differences between entries and checks, an unweighted two‐way analysis with trials and entries as factors, a weighted two‐way analysis, and three versions of best linear unbiased prediction (BLUP). The versions of BLUP involved different estimates of the entry and environmental variance components. The trial series included 150 different entries (cultivars ande xperimentasl ynthetics) that had been evaluated in one to 14 trials between 1975 and 1982. Eleven of these trials had data for total yield for each of three growing seasons. Each of the methods of estimation were conducted with each of the 11 complete trials omitted from the data set, and estimates were used to predict means for the omitted trial. The method with smallest prediction error was judged “best.” The smallest prediction errors were obtained with the version of BLUP in which the entry and environmental variances were estimated from an unbalanced two‐way (trials and entries) analysis of variance. The second best method was the version of BLUP in which environmental and entry variances were estimated from the pooled sums of squares over trials. The version of BLUP in which variances were estimated for each trial ranked third and was only slightly superior to the summation of differences, unweighted, and the weighted analyses. The percent of checks had the largest prediction errors of all the methods we evaluated.