[Study on the calculation methods of confidence intervals of genetic parameter estimates by method R and number of...

The objective of the study was to probe the calculation methods of confidence intervals of genetic parameter estimates by Method R and influences of number of repeating estimations (NORE). Four models were used to generate the datasets by simulation. The datasets originated from 200 sires and 2 000 dams and simulation progressed by BLUP( Best Linear Unbiased Prediction) selection for five overlapping generations. Variance components were estimated by using multivariate multiplicative iteration (MMI), combined with preconditioned conjugate gradient (PCG) to solve the mixed model equations. Parameter estimates and their standard errors and confidence intervals were computed by classical method, classical method after Box-Cox transformation and bootstrapping. The results showed that when NORE was larger, all three methods were feasible, but when NORE was fewer, bootstrapping was recommended. Fewer NORE was feasible under simple models, but more NORE was needed for complex models. Direct heritabilities were overestimated with the increase of number of random effects, but underestimation was found for many parameters with the increase of iteration numbers of PCG and MMI.