|
Abstract:
|
Genomic prediction (GP) is widely used in plant breeding to help find the best genotypes for selection. Here, the accurate estimation of predictive accuracy (PA) and heritability (H) is essential for genomic selection (GS). However, as in many other applications, the models of choice for analyzing field data are regression models which are known to lead to biased parameter estimates when one or more of their underlying assumptions are violated. These and other biases often translate into inaccurate H and PA estimates which in turn may negatively impact GS. Since phenotypic data are prone to contamination and almost never perfectly conform to the normality premise, one way of improving GS involves refining its estimation accuracy. Here, robust statistical methods provide a natural framework, since they are known to overcome some of the handicaps of the likelihood-based classical methodology like departure from normality. Therefore, a robust analogue of a two-stage method from the literature used for H and PA estimation is presented. Both techniques are then compared through simulation.
|
