SENSITIVITY (STABILITY) ANALYSIS OF MULTIPLE VARIETY TRIALS, WITH SPECIAL REFERENCE TO DATA EXPRESSED AS PROPORTIONS OR PERCENTAGES

The literature on 'stability analysis', which is better called 'sensitivity analysis', leaves much to be desired. This paper attempts to clarify the issues, particularly the effect of the non-linearity of the standard model relating variety yields to mean yields over sites. The model is extended from its original applications in the analysis of crop yields to the analysis of measurements that cannot be considered Normally distributed, such as disease assessments expressed as percentages or proportions. Modern computer software offers a method that is theoretically sound and can be applied routinely to data of many different distributional types. Graphical methods of displaying the results of the analysis of non-Normal data are discussed.