How low can you go? Performance of factor analytic models in the analysis of multi-environment trials with small numbers of varieties

Macdonald, Bethany (2018) How low can you go? Performance of factor analytic models in the analysis of multi-environment trials with small numbers of varieties. Honours thesis, University of Southern Queensland. (Unpublished)


Download (616kB) | Preview


Crop breeding programs test large numbers of crop varieties in field trials spanning a range of years and locations, with these groups of trials known as multi-environment trials (MET). In the early stages of crop breeding programs large numbers of new varieties are grown in a small number of field trials. The best varieties in each stage are selected to progress to the next stage so that in the final stages a small number of elite varieties are grown in a large number of field trials across the country. These trials are conducted to determine which varieties perform best in which environments and an appropriate statistical analysis resulting in accurate predictions of the variety by environment (VxE) effects is integral to this.

There have been many statistical approaches to the analysis of MET data, however all methods involve investigating the nature of the VxE effects. The factor analytic (FA) structure for the VxE effects allows heterogeneity of genetic variance for environments and heterogeneity of genetic covariance between pairs of environments, and is currently considered best practice in the analysis of MET data in Australia. The FA model has been shown to be the superior model for large numbers of varieties both in terms of goodness-of-fit and the selection of superior varieties. However, this superiority has not been demonstrated for small numbers of varieties, such as in the late stages of crop breeding programs, despite being regularly used in such scenarios. Five data sets with different underlying VxE patterns and numbers of trials, four numbers of varieties, and two levels of varietal concurrence were used to provide scenarios for a simulation study to investigate the adequacy of an FA variance structure for VxE effects. How the accuracy of the FA model changes as the number of crop varieties decrease, along with the implications the underlying VxE variance structure and level of varietal concurrence have on the accuracy of the FA model when dealing with small numbers of varieties were investigated. The comparisons were based on the mean square error of prediction of the VxE effects.

This study showed that 15 varieties per trial is sufficient in a MET data set to accurately estimate the VxE effects, and that in some cases MET data sets with even as few as 10 varieties could be considered. It was found that the underlying patterns in the variance of the VxE effects impacted on how the accuracy of the FA model compared to the accuracy of other models, especially for very small numbers of varieties. In addition this study demonstrated that the FA model is affected by changes in concurrence more than the other models that were considered, however these changes in accuracy have minimal implications. Finally, this study highlighted the tendency of the log-likelihood ratio test to select overly complicated models in its test for a significant model improvement.

Statistics for USQ ePrint 37210
Statistics for this ePrint Item
Item Type: Thesis (Non-Research) (Honours)
Item Status: Live Archive
Additional Information: Bachelor Of Science (Honours) thesis.
Faculty/School / Institute/Centre: Historic - Faculty of Health, Engineering and Sciences - School of Agricultural, Computational and Environmental Sciences (1 Jul 2013 - 5 Sep 2019)
Supervisors: King, Rachel; Kelly, Alison
Date Deposited: 22 Oct 2019 00:25
Last Modified: 22 Oct 2019 00:25
Uncontrolled Keywords: crop breeding programs; factor analytic models; multi-environment trials
Fields of Research (2008): 01 Mathematical Sciences > 0104 Statistics > 010401 Applied Statistics
Fields of Research (2020): 49 MATHEMATICAL SCIENCES > 4905 Statistics > 490501 Applied statistics

Actions (login required)

View Item Archive Repository Staff Only