Atlas: A simulation study of unreplicated and replicated designs for early generation plant breeding trials by Alison Kelly

Atlas home || Conferences | Abstracts | about Atlas

Australasian Genstat Conference 2002
December 4-6, 2002

Busselton, Western Australia, Australia

Organizers
Jane Speijers - Convenor Organising Committee, Peter Clarke - Chairman Programme Committee

A simulation study of unreplicated and replicated designs for early generation plant breeding trials
by
Alison Kelly
Qld Department of Primary Industries
Coauthors: Brian Cullis (NSW Agriculture), John Eccleston (University of Qld)

As genotypes progress through the stages of a cereal breeding program, there is generally increasing rigour in the statistical design of field trials from earlier to later stages of testing. In general, seed is limited in the early stages, and because of the magnitude of the genotype by environment interaction, most test lines are sown in single plots across a range of environments. These large unreplicated trials contain some replicated check varieties interspersed throughout the trial. In later stages, advanced lines are tested in replicated experiments across a wider range of environments.

The programs aim to progress material that is high yielding and so the statistical design and analysis should aim to improve the heritability of this trait. Selection for yield has improved through the use of statistical methods for spatial analysis of field trials (Cullis and Gleeson, 1991), and through advanced models for genotype by environment interaction (Smith, 1999).

Our study examines the impact of modelling spatial trend and genotype by environment interaction under differing replication regimes. Three designs are compared for bias in variance parameter estimates of the underlying mixed model, and for increase in heritability and selection response. The replicated design is a neighbour balanced design (Eccleston, 2001), and the unreplicated design has check entries interspersed at a frequency of 1:3. For the third design, we propose an improved layout for early generation trials that contains partial replication of the genotype set at each environment, balanced across environments, using the design package of Coombs (2002). It is found that this design type is superior to unreplicated designs in improving the heritability of the trait and equivalent in the level of bias in the variance parameters estimates from the mixed model. This layout is termed a super-replicated design.

References

Coombes, N. (2002) Developments in experimental design. Biometrics Meeting, NSW Agriculture, GOSFORD .

Cullis, B.R., and Gleeson, A.C. (1991) Spatial analysis of field experiments - an extension to two dimensions. Biometrics 47, 1449-1460.

Eccleston, J.A., Elliot, L.E., Chan, B., Martin, R.J., and Chauhan, N. (2001) Design of large field trials:spatial dependency, optimality and visualisation. In Proceedings of GENSTAT 2001. European Genstat Conference, Oxford, UK.

Smith, A.B. (1999) Multiplicative mixed models for the analysis of multi-environment trial data. PhD Thesis, Department of Statistics, University of Adelaide.

Date received: September 9, 2002