Bayesian approach, traditional method, and mixed models for multienvironment trials of soybean

Abstract

Abstract: The objective of this work was to compare the Bayesian approach and the frequentist methods to estimate means and genetic parameters in soybean multienvironment trials. Fifty-one soybean lines and four controls were evaluated in a randomized complete block design, in six environments, with three replicates, and soybean grain yield was determined. The half-normal prior and uniform distributions were used in combination with parameters obtained from data of 18 genotypes collected in previous and related experiments. The genotypic values of the genotypes of high- and low-grain yield, clustered by the Bayesian approach, differed from the means obtained by the frequentist inference. Soybean assessed through the Bayesian approach showed genetic parameter values of the mixed model (REML/Blup) close to those of the following variables: mean heritability (h2mg), accuracy of genotype selection (Acgen), coefficient of genetic variation (CVgi%), and coefficient of environmental variation (CVe%). Therefore, the mixed model methodology and the Bayesian approach lead to similar results for genetic parameters in multienvironment trials.

Keywords

• Glycine max
• mathematical modeling
• prior distribution in plant breeding