Data transformation in stability analysis of dynamic-model. 2.Effects on genetic-parameters

Abstract

Considering the case of the residual mean squares of deviation from the linear regression, if they are not homogeneous across all genotypes, thus a power transformation of data was proposed. To test the adequacy and necessity of transformation, the effects of transformation of data on genetic parameters in stability analysis of dynamic models were discussed. The method of Schlesselman (1973) can by adapted to provide as a means of estimating approximate confidence intervals of a power transformation for reducing error heterogeneity. The method involves the regression of log {sample variance} on log {sample mean}. log Y and Y-0.1 transformations were considered in analyzing the fresh data of 9 x 9 diallel crosses of Arabidopsis thaliana, highly significant correlation was found between these two transformation scales and origianl data. However, the use of any data transformation may reduce the correlation between stability index and phenotypic mean because of the removal of heterogeneity of variance. This may help breeders select genotypes having above mean performance and high stability. On the other hand, the magnitudes of interactions and heterogeneity among regressions are also reduced by using data transformation. The effects on genetic parameters in stability analysis of dynamic model were in accordance with those on space-time parameters of dynamic model in previous study.