Inference for fixed effects in linear mixed models
by
Sharon Smith
NSW Agriculture
Coauthors: R. Thompson, B.R. Cullis

It is well known that in the case of linear mixed model exact tests can be derived to test hypotheses concerning fixed effects. However, no exact tests are available for unbalanced or non-orthogonal data with more than one random component. The current approach for testing hypotheses concerning fixed effects in ASREML and GENSTAT is the Wald statistic, which has an asymptotic chi-squared distribution. The Wald statistic is known to be anticonservative especially for small samples. Kenward and Roger (1997) proposed a set of adjustments to improve the performance of this statistic.

In this paper, we overview the work of Kenward and Roger and provide computationally effective methods of implementing these adjustments into current mixed model software.

Date received: August 27, 2002