Abstract
Routine estimation of the rate of genetic gain (ΔGt) realized by a breeding program has been proposed as a means to monitor its effectiveness. Several methods of realized ΔGt estimation have been utilized in other studies, but none have been objectively evaluated in a plant breeding context. Stochastic simulations of 80 rice (Oryza sativa) breeding programs over 28 years were done to generate data used to evaluate five methods of realized ΔGt estimation in terms of error, precision, efficiency and correlation between true and predicted annual mean breeding values. Two indicators of ΔGt, the expected ΔGt and the average number of equivalent complete generations (EqCg), were described and evaluated. At best, estimates of realized ΔGt were over or underestimated by 15% and 27% when considering all 28 years and the past 15 years of breeding respectively. The best methods were the control population, estimated breeding value, and ERA trial methods. Among these, correlations between true and estimated ΔGt were at best 0.59, indicating that these methods cannot very accurately rank breeding programs in terms of realized ΔGt. The expected ΔGt and the average EqCg were shown to be useful indicators for determining if a non-zero genetic gain is expected. Determining which of the three best realized ΔGt estimation methods evaluated, if any, would be appropriate for any given breeding program should be done with careful consideration of the objectives, resources, seed stocks, and structure of the data available.