Prediction in linear mixed models
by
Arthur R Gilmour
NSW Agriculture
Coauthors: Sue Welham, IACR, Rothamsted, Brian R Cullis, NSW Agriculture, Beverly Gogel, NSW Agriculture, Robin Thompson, IACR, Rothamsted

The development of ASREML for fitting a comprehensive range of linear mixed models prompted the development of prediction procedures which accommodated these models. Previous work on prediction had not addressed all the issues involved in predicting from unbalanced mixed models. Prediction problems include calculation of adjusted means, fitted curves and response surfaces. There needs to be a robust procedure for handling unbalance in the design, identifying non-estimable prediction functions, distinguishing between random effects which are error terms to be ignored and random terms to be averaged or used in prediction. The procedures must be able to handle large models and prediction of many values without excessive use of computing resources. The underlying principles of the procedures developed for use in ASREML are reported in Welham et al. (2002). The paper will highlight two interesting examples from that paper. References Welham, S., Gilmour, A.R., Cullis, B.R. Gogel, B.G. and Thompson, R., (2002) Prediction in linear mixed models. In preparation.

Date received: September 3, 2001

Copyright © 2001 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cahg-85.