Pak. J. Bot., 41(2): 711-730, 2009. NONPARAMETRIC METHODS IN COMBINED HETEROSCEDASTIC EXPERIMENTS FOR ASSESSING STABILITY OF WHEAT GENOTYPES IN PAKISTAN SYED HAIDER SHAH 1 , SYED MUNAWAR SHAH 2 , M. INAYAT KHAN 3 , MAQSOOD AHMED 4 , IMDAD HUSSAIN 5 AND K. M. ESKRIDGE 6 1 Department of Statistics, University of Balochistan, Quetta, Pakistan 2 Department of Economics, Balochistan University of Information Technology, Engineering and Management Sciences, Quetta, Pakistan 3 Department of Mathematics and Statistics, University of Agriculture, Faisalabad, Pakistan 4 Department of Environmental Management and Policy, BUITEMS, Quetta, Pakistan 5 Department of Business administration, Iqra University, Karachi, Pakistan 6 Department of Statistics, University of Nebraska, Lincoln, USA Abstract Genotype performance in multienvironment trials (METs) are usually analyzed by parametric analysis of variance (ANOVA) and stability models. The results of these models can give misleading inferences when some sensitive assumptions are not satisfied. In this paper, assumptions of combined ANOVA are scrutinized in detail to justify the validity of use of 5 nonparametric stability methods (S i (1) , S i (2) , S i (3) , S i (6) and YS i (1) , YS i (2) ) applied to 20 genotypes grown in 40 hetroscedastic and nonnormal environments in Pakistan for the year 2004-05. There is a severe heterogeneity problem in the data because the ratio of the largest estimated mean squares error (MSE) for individual environments randomized complete block design (RCBD) to the smallest MSE is approximately (1.00/0.02=50). Out of 40 environments individual coefficients of determination (R 2 ), 27 are less than 0.70. This leads to violation of linearity assumption in the model. Standardized residual plots vs. individual environments plots and normal probability plot are indicators of the violation of homogeneity, normality assumptions and absence of outliers. No linear relationship was established between the natural logs of the error variance and the natural log of environments’ mean, which again violates coefficient of variation (CV) assumption. Remedial transformations as suggested in literature were not successful to stabilize environments MSEs and could not normalize the data, so as a last resort in this regard nonparametric stability methods seem to justify the analysis of genotype x environment interactions (GEI). The low values of modified rank-sum statistics YS i (1) and YS i (2) were positively and significantly associated with mean yield but the other nonparametric methods were not correlated with mean yield. The results of principal component analysis and correlation analysis of nonparametric stability methods indicate that the use of modified rank-sum method would be justifiable for simultaneous selection for high yield and stability. Using modified rank-sum method, the genotypes G7, G3, G15, G5 and G12 were found to be the most stable with yield, whereas G14 and G19 were the least stable genotypes. Introduction One of the most challenging issues in plant breeding process to accurately analyze genotype x environment interaction (GEI) is based on data from multienvironment trials. GEI is a universal issue that relating to all living organisms, from humans to plants and bacteria (Kang, 1998). Usually GEI is the nonadditive component of two or more experiments with the same genotypes combined over environments. The process for selecting high yield and stable genotypes usually involves three stages of experimentation: At stage-1, genotypes are tested at a single location; at stage-2, the selected genotypes are tested in a multilocation trials (genotype x location); and finally at stage-3, the most promising genotypes with new set of genotypes are tested for several years under a range of locations (genotype x location x year) (Linn & Binns, 1994). Corresponding author Email: mrhaidershah@yahoo.com