Logistic Regression Models to Aid Selection in Early Stages of Sugarcane Breeding

In sugarcane breeding, visual appraisal is used to select for cane yield among seedlings (Stage I) and non-replicated clonal plots (Stage II). Genotype by environment interaction and interplot competition reduces selection efficiency. Although path coefficient analysis studies identified important yield components for indirect selection, currently there is little integration of that knowledge in models to improve early stage selection. This study demonstrates the use of logistic regression models as a statistical decision support tool that uses simultaneous selection for yield components (stalk number, stalk height, stalk diameter, estimable recoverable crystal (ERC) % cane) to help identify genotypes that produce high yield in non-replicated stages of sugarcane breeding. Data were collected from two Stage I populations grown at the USDA-ARS, Sugarcane Research Station at Houma and LSU AgCenter, Sugar Research Station, St. Gabriel, Louisiana, USA and two Stage II populations grown at Gingindlovu and Empangeni research stations, South African Sugarcane Research Institute, in South Africa. The data were analyzed using the logistic procedure of the statistical analysis system. Genotypes selected using the logistic regression models produced higher yield and larger values for yield components than those not selected. Stalk number was the most influential yield component while ERC% cane was the least. The logistic regression models identified the yield components that contributed significantly to yield for each population. The differences in variability within and among the populations provided guidance as to the precision of selection and therefore could indicate to the breeder the need for a higher or lower selection rate. Higher discriminating ability between selected and rejected genotypes indicated greater precision of selection.