2003 Volume 53 Issue 1 Pages 15-20
Multi-step yield screening for a population of test varieties can be performed using three different patterns; a pre-assigned upper fraction of test varieties is selected in each step (pattern I), or, test varieties that showed yield scores superior to one or a few check varieties are selected in each step (II), or, test varieties superior to a critical fixed standard are selected in each step (III). Pattern II has been commonly used in actual screening trials. By comparing the efficiencies (probability of success) of these patterns based on the Monte Carlo calculations for finite populations composed of 20, 40 or 60 test varieties, we concluded that pattern I is the most efficient. Supplementary procedures should be used to minimize the loss due to investing for undesirable populations that lack superior varieties, such as totally discarding the population if none of the test varieties showed a promising score in the first step, and continue screening in pattern I if one or more varieties showed a promising score. Screening in three steps will be efficient when the variety × year interaction is absent, but more steps may be used when a large variety × year interaction is anticipated to occur or is observed in the data in earlier steps. We conclude that patterns II and III are inefficient because they are subject to a large error due to the variety × year interaction or yearly effect.
