Data from multilocation trials are important in agriculture because they can be used to: 1) estimate and predict yield differences among genotypes, 2) assess genotype x site interaction and yield stability, and 3) select superior genotypes for planting in future years and at new sites. Data from multilocation trials is imprecise, complex and noisy. To increase the accuracy of genotypic yield estimates and their contrasts, one can therefore: 1) partition the error variance by using incomplete block designs in the form of lattice or row-column designs, 2) partition the genotypic variation by using spatial models of the form of nearest neighbor analysis, and 3) partition the genotype x site interaction variation by using statistical models that will remove noise from the source of variation. Since these strategies are applied to orthogonal sources of variation, they can be used independently. In this study we discus practical concepts related to increasing the precision of the comparison between genotypic means through improved experimental designs (replicated or unreplicated). Data from the CIMMYT Early Drought Experimental Variety Trial (EDEVT), which included eight drought tolerant maize genotypes, two CIMMYT long-term reference checks and two different local checks, were used. The experimental design used in each of the 21 sites was an alpha (0,1) lattice. The efficiency of the lattice design was compared with the conventional randomized complete block design in each site. Changes in the significance of some genotypic mean contrasts when using adjusted vs unadjusted means were assessed. Genotype x site interaction analysis using the AMMI model, and the grouping of sites without genotypic rank changes using the SHMM model, were performed on the adjusted genotype means.

Genetic Resources Program

English

9802|AGRIS 9702|anterior|R97-98PROCE|FINAL9798

Jose Juan Caballero

CCJL01

CIMMYT Publications Collection