Comparing yield stability of Musa genotypes in Nigeria using four statistical methods

Four statistical approaches were examined for stability analysis for yield in order to determine their congruence in identification of musa genotypes. Thirty-six Musa genotypes grown in four location for two crop cycle were evaluated. The statistical methods used included Gauch’s additive main effect and multiplicative interaction (AMMI) model, Finlay and Wilkinson’s regression coefficient (FW), Francis and Kennenberg’s grouping based on CV and mean yield (FK), and Kang’s statistic for simultaneous selection (YSi) for high and stable yield. Changes in yield of genotypes across the locations were observed, suggesting that there was little homeostatic adaptation of the genotypes. There was a weak rank correlation among the locations and crop cycles, and also among stability methods. However, AMMI was significantly correlated with YSi. Also, FW was significantly correlated with FK. The level of convergence between any two methods ranged from 16 to 44%, while among any three or the four method was between 16 and 18%. Three cooking banana landraces, Bluggoe, Fougamou and pelipita were identified as high yielding and stable by all methods. In most cases, genotypes selected by AMMI and FW were also classified as stable by the other two methods. We conclude that simultaneous use of different stability statistics may protect breeders from wrongly identifying presumably stable genotypes.