Best Linear Unbiased Prediction

!BLUP

!BLUP n
restricts ASReml to performing only part of the first iteration on the assumption that variance parameters are known and provided in the job. The estimation routine is aborted after
     n = 1: forming the estimates of the vector of fixed and random effects,
     n = 2: forming the estimates of the vector of fixed and random effects, REML log-likelihood and residuals (this is the default),
     n = 3: forming the estimates of the vector of fixed and random effects, REML log-likelihood, residuals and inverse coefficient matrix.
     n = 4: forming the estimates of the vector of fixed and random effects by solving the mixed model equations using the preconditioned gradient method,
     n = 10:19 forming the estimates of the vector of fixed and random effects by solving the mixed model equations by Gauss-Seidal iteration with relaxation factor n/10.
     Use the !MAXIT i where i > 100 to set the maximum number of iterations used when $4 <= i <= 19.

ASReml prints its standard reports as if it had completed the iteration normally, but since it has not completed it, some of the information printed will be incorrect. In particular, variance information on the variance parameters will always be wrong. Standard errors on the estimates will be wrong unless n=3. Residuals are not available if n=1. Use of n=3 or n=2 will halve the processing time when compared to the alternative of using !MAXIT 1. However, !MAXIT 1 does result in a complete and correct output report.

Back to general qualifiers

Return to start