Skip to main content
SLU publication database (SLUpub)

Conference paper2007Peer reviewed

Analysing and reporting series of variety trials

Forkman, Johannes

Abstract

Series of variety trials have in Sweden been analysed with mixed models since 1998. The analyses are made, in co-operation with external partners, by the Field Research Unit at the Swedish University of Agricultural Sciences. In this paper we discuss experiences with mixed models. We analyse an unbalanced series of variety trials by i) a mixed model, ii) pairwise calculations, and iii) a fixed effects model. We also introduce iv) a hybrid of pairwise calculations and a mixed model. In each analysis the varieties are compared with a control. The objective of making several analyses is to expose advantages and problems connected with different methods for analysing series of variety trials. The mixed model analysis based on REML has, compared with traditional pairwise t-tests, many advantages. In the mixed model analysis all varieties are investigated simultaneously, which adds degrees of freedom to the estimate of the residual variance. Pairwise t-tests are made under the assumption that all differences are independent. This assumption is not fulfilled when some but not all differences belong to the same year. In the mixed model analysis this problem is solved by estimating several variance components. Nevertheless, the mixed model analysis sometimes produces apparently strange results. The analysis is also more difficult to perform and understand than the traditional pairwise t-tests. Many researchers and partners still prefer to compare average yields directly. To overcome the reluctance to mixed models the concepts of variance components need to be communicated. The traditional ANOVA, with a single random error term, simply cannot describe complex field experiments. Estimated differences between varieties should if possible be reported together with a measure of the precision in the estimates. Imprecise estimates are otherwise easily assumed to be reliable. A mixed model proved suitable for routine analysis of series of variety trials is presented

Published in

Title: Proceedings of the International Symposium 'Agricultural Field Trials - Today and Tomorrow'
ISBN: 978-3-86186-541-4

Conference

International Symposium 'Agricultural Field Trials - Today and Tomorrow'