Comparison of regression methods for evaluation of phenotypic stability in sugarcane

The purpose of this work was to compare the regression methods used for evaluating phenotypic stability in plants. The methods of Cruz et al, based on a linear model and of Toler & Burrows based on a non-linear model were considered, Data of the sugarcane 1994 characterization trial coordinated by the genetic improvement program of Copersucar were used. Experiments were set up in 13 locations of the State of Sao Paulo, in Brazil for evaluation of 20 sugarcane genotypes. The fitting of the models to the data was obtained for both methods. Measures of environmental quality were numerically different but agreed in general terms for the ranking of locations. However, when genotypes were classified on the bases of yield and pattern of response only ten coincided for both methods. The statistical advantages of the non-linear model makes it recommendable for studies on stability.