Using logistic regression models for selection in non-replicated sugarcane breeding populations

Abstract

Increasing selection efficiency in un-replicated populations has remained a challenge to sugarcane breeders due to the effects of genotype by environment interaction and competition among plots. Therefore studies aimed at exploring models to improve selection efficiency are required. At the South African Sugarcane Research Institute, the Stage II selection populations are planted to un-replicated plots. Visual selection, which is known to be subjective, is used to determine the genotypes advanced. Although path coefficient analysis has identified important yield components, there is little use of that knowledge to enhance selection. This study demonstrates the use of logistic regression model as a statistical decision support tool to enhance selection in un-replicated populations. The logistic regression model used stalk number, stalk height, stalk diameter and estimable recoverable crystal (ERC) % cane to predict sugar yield. The data were collected from Stage II populations across regional breeding programs and analysed using the logistic procedure of Statistical Analysis System. The Likelihood Ratio, Score and Wald statistics were highly significant (P < 0.0001) indicating the robustness of the models. Stalk number is the most influential component to sugar yield followed by diameter, height and ERC % cane. The coefficients of the yield components varied across different programs indicating the potential of the analysis to determine yield component combination. Genotypes selected using the models produced significantly higher trait values than those rejected. The logistic regression models should enhance selection efficiency in un-replicated populations as well as provide for the evaluation of breeding populations and trait combinations across regional breeding programs.