Impact on genetic gain from using misspecified statistical models in generating p‐rep designs for early generation plant‐breeding experiments

Abstract

AbstractThis paper is concerned with the generation of designs for early generation, plant‐breeding experiments that use limited experimental resources as efficiently as possible to maximize the realized genetic gain (RGG) resulting from the selection of lines. A number of authors have demonstrated that partially replicated (p‐rep) designs for such experiments, in which the percentage of lines that are duplicated is p, are likely to be more efficient than grid‐plot designs. Therefore, our aim is to obtain the most efficient p‐rep design for an experiment using one of two distinctly different criteria and employing widely or readily available statistical software packages to search for an optimal design. However, this can be difficult because knowledge of the sources of variation and their magnitudes is required and is often unavailable. To overcome this impediment, a comprehensive simulation experiment was conducted to investigate whether designs that are robust to a wide range of experimental situations can be identified. Designs with p set to 20% and for different experimental situations are generated and the performance of each tested for 24 different variation scenarios. We concluded that for large experiments, the RGG obtained with various optimal designs is indeed not affected by the different variation scenarios and that resolved designs for fixed genetic effects should be generated for robustness. On the other hand, the design assumptions affect the RGG for small p‐rep designs. Even so, an overall recommendation is made.