Early generation testing has been used as a method for discriminating objectively among large populations of entries for grain yield. The objective of this thesis was to evaluate the effectiveness of such procedures for the identification of entries for which further testing was justified.
A population of entries was evaluated in unreplicated trials at three sites (Toobeah, Biloela and Pirrinuan) in 1978, and in replicated trials at five sites (Moonie, Tummaville, Billa Billa, Muckadilla and Bongeen) in 1982/83. For individual trials, coefficients of variation ranged from 7 to 17 percent in 1978 and from 7 to 15 percent in 1982/83. Genetic coefficients of variation ranged from 14 to 22 percent in 1978 and from 9 to 21 percent in 1982/83. The variance component estimate for entry x site interactions was 1.3 times greater than that for entries in 1978 and 2.0 times greater in 1982/83. The pairwise correlations between each of the sites ranged from 0.32 to 0.46 in 1978 and from 0.02 to 0.55 in 1982/83. These results indicated that substantial variability existed for the absolute and relative yield performance of entries within and among sites in both series of trials.
Heterogeneity among regressions on an environmental index accounted for 71 percent of the entry x site interaction sum of squares in 1978 but only 27 percent in 1982/83. Greater than 50 percent of the entry x site interaction sum of squares was accounted for by individual regressions for 65 percent of the entries in 1978, but only 16 percent of the entries in 1982/83. The rank correlations between the two series of trials, for regression coefficients (0.08), ecovalence (-0.03), deviations from regression mean square (0.25) and percentage entry x site interaction sum of squares accounted for by regression (-0.13) indicated that these statistics had no predictive value. Furthermore, they did not facilitate the discrimination among entries on the form of their response across sites.
Classification retained 69 and 63 percent of the entry x site interaction sum of squares among groups for 1978 and for 1982/83, respectively. Groups were relatively homogeneous for mean yield and for the form of entry response across sites. The membership of the groups was related to the origin of the genetic material, its level of productivity and to reactions to diseases. There was only a moderate consistency between the membership of groups from classification on performance in 1978 and in 1982/83. This reflected the marked differences between the two series of trials for the absolute and relative performance of entries.
Adjustment of data did not improve the precision of trials in 1978, suggesting that trials were conducted on relatively uniform sites. The performance of entries for mean yield across sites in 1978, and in 1982/83, was more closely related to the performance at those sites with relatively high mean yield and/or standard deviation, and for which the correlation with the performance of entries at other sites was relatively high. Transformation of data as ranks or as standard normal variates resulted in relatively higher associations between the mean yield across sites and those individual sites comprising the mean for which the mean yield and/or standard deviation were relatively low. Transformation of data as relative yields or as logarithms was not effective for this purpose. However, the effect of each transformation was small for 1982/83 where there were relatively large differences in the correlations between pairs of sites for the yield performance of entries. This suggested that the correlations between pairs of sites had a more marked influence on the association between the performance of entries for the mean yield across sites and the individual sites comprising the mean, than did site mean yields and standard deviations.
Discrimination among entries in 1978 was practised using relatively high mean yield across sites, a combination of relatively high mean yield and a regression coefficient close to zero, and relatively high mean yield of groups delineated by classification. These selection strategies retained populations of entries with many common members. This was probably due to the importance of the mean yield across sites for discriminating among entries.
Relatively larger numbers of entries were retained with elite performance at Biloela, than at Pirrinuan or at Toobeah. The responses to selection in 1982/83 for the populations retained using a selected fraction of 20 percent were similar to, or higher than, those for a selected fraction of 10 percent in several instances. Relatively high proportions of the entries with elite performance in 1982/83 were rejected using a selected fraction of 10 percent.
These results suggested that entry x site interactions in 1978 and in 1982/83 had a marked impact on the relative performance of, and discrimination among, entries for yield performance across environments. It was recommended that in the presence of substantial entry x year and entry x site x year interactions multi-stage selection may be the most effective procedure for the identification of those entries in early generation trials for which further testing was justified. Alternatively, a series of managed environments with a specific set of environmental challenges may achieve more effective discrimination among entries.