Optimisation of experimental design and analysis for variety trials to maximise genetic gain : SRDC final report UQ023
Abstract
The analysis of field experimental data usually assumes independent observations. This method is often inappropriate if the data contain strong spatial trends. Methods of analysis (commonly used in cereal trials) which account for the spatial variability existing within a field were shown to improve the precision of the estimates of clonal effects in sugarcane variety trials. As inter-plot competition was a major contributing factor biasing clonal estimates, a model which accounted for the competitive interaction between neighbouring clones was developed. Both factors could be incorporated into a more general model that jointly assessed inter-plot interference and fertility gradients. Two different estimation procedures (marginal likelihood and profile likelihood) for the parameters in the underlying models were investigated in a simulation study. Based on the criteria of average bias and mean-squared error, there was a slight preference for the marginal likelihood estimators. An empirical investigation of the relative efficiency of the augmented randomised complete block design versus the spatial unreplicated design in early generation sugarcane breeding programs was also undertaken. This indicated that the spatial design was to be preferred with respect to the bias in the parameter estimates, but using other criteria for design evaluation there were very few advantages in choosing the more complicated spatial design.