Selection and Fitting of Mixed Models in Sugarcane Yield Trials

Abstract

Mixed models are a useful tool for the analysis of sugarcane field trials in which sugarcane varieties are allocated in different locations and phenotypic traits are evaluated in the same experimental unit (plot) over time. One challenge to analyze these data is how to build a good mixed model when no experimental design is planned, because all sugarcane varieties in the area of influence of a sugar mill are planted in different years due to the age of the crop and there is no spatial information on all plots. The aim of this research was to examine and to determine the most appropriate mixed model for estimating cane stalk yield of sugarcane varieties when previously there was no planned experimental design. Cane stalk yields of 26 sugarcane genotypes harvested in 24 different locations and in different crop cycles (age) were analyzed. The randomized block nested design (plot within block) with ratoon crop as a class variable in the mixed model was the best for the mean comparisons in sugarcane genotype trials (Model 3), allowing a gain in information. The randomized block design approach helps to fit more general random effects, and the covariance structures helps to improve the performance of mixed model repeated measures. This study emphasizes the need to improve the process of finding a good enough mixed model, that is, how to define the mean structure and the best covariance structure for model sugarcane trials that enables more powerful and efficient parameter estimations. The results showed how a more appropriate mixed model might help avoid errors of judgment in sugarcane genotype recommendations for enhancing the productivity of the cane industry.

Keywords

• mixed model
• fit statistics
• covariance structure
• mean square error