Abstract

When using linear mixed models for analysis of series of variety trials, the statistical inference space is dependent on the design of the series. In this paper, four inference spaces are recognized for one-year crop variety trials: single-year local, long-term local, single-year regional and long-term regional. In addition, different linear mixed models are appropriate for different designs. Five models are defined that are useful for 13 different types of series of variety trials. The standard statistical analysis includes estimation of variance components using the REML method and estimation of variety means by generalised least squares. Although this standard method gives the best linear estimates, provided correct variance components, the estimated differences between the test varieties and the control variety occasionally do not agree with yearly or local results. An alternative method is outlined, termed the control method, which does not have this problem.