(Co-ordinator: Assoc. Prof. Ari Verbyla)
The research activities of BiometricsSA are still undergoing development, with the current focus on statistical modeling. Strong links with researchers in the national and international communities have been established and are reflected in the supervision and collaborative research being conducted. The research programs are listed below; describing the research directions and the common underlying methods such as the mixed model.
1. Statistical Genetics: QTL analysis
Molecular genetics is becoming increasingly important. The Cooperative Research Center for Molecular Plant Breeding involves many populations that are segregating for particular traits. Molecular markers allow the mapping of the plant genome (linkage map) while field and laboratory trials provide information on the many lines generated for each population.
The analysis of the field and laboratory data in conjunction with molecular markers has been carried out without regard to field and laboratory variation, and this includes genotype by environment interaction. There is a need to incorporate the knowledge of field variation and more recently of laboratory variation into the determination of quantitative trait loci (QTL) based on the marker information. Research on these issues is being carried out (Paul Eckermann and Ari Verbyla) in collaboration with Dr Brian Cullis of NSW Agriculture and Professor Robin Thompson of Rothamsted Experimental Station, Harpenden, UK.
2. Mixed models
Mixed models are models involving both fixed and random effects. Staff of BiometricsSA have a record of achievement in research in this area. Indeed much of the research and collaborative work with researchers has an element of these methods.
Current research is focused on stability analysis (Andreas Kiermeier and Ari Verbyla) in collaboration with Dr Richard Jarrett of CSIRO. This research is concerned with shelf-life of products and involves the use of random coefficient models.
Other research activities in this area include branching splines, for modeling non-linear relationships for experiments where treatments are introduced at various times in the experiment.
3. Generalized linear models and extensions
The class of generalized linear models is used in data analysis for many situations. These models include standard analysis of variance and regression, but also apply to non-normal data such as counts. Extensions to allow for different dispersion parameters are on-going (Ari Verbyla) in collaboration with Dr Gordon Smyth of the University of Queensland.
A very important extension of these models is the inclusion of random effects. Theoretical developments are required in this area. In addition, fast and efficient statistical software to implement these developments is required and is currently underway (Julian Taylor and Ari Verbyla) in collaboration with Dr Bill Venables of CSIRO.
Ordinal data require specialized methods. A collaborative effort between Debra Partington, Ari Verbyla and Raul Ponzoni of Livestock Systems in SARDI will examine the analysis of such data in an animal breeding setting.
4. Spatial and Temporal Modeling
Researchers are often interested in changes over time and or space. Temporal and spatial data require special treatment in design and analysis as the observations are correlated. Research in these areas has lead to many advances in statistical methods, resulting in improved outcomes for substantive research in the agricultural, aquatic, biological and environmental sciences.
An important area of activity is the spatial analysis of field trials (Colleen Hunt) and the extension to the analysis multi-environment trials. Links with NSW Agriculture and Rothamsted Experimental Station, UK, and a GRDC project have resulted in substantial improvements in this area. Current research has involved the quantification of site selection and ranking of varieties.