The possible pitfalls of using productivity data to evaluate variety performance

The variation and structure of block productivity data affect the interpretation of variety performance for both cane yield and CCS. The confounding factors inherent in block productivity data can bias any analysis and potentially lead to an over or underestimation of a variety'S perfonnance. Exploratory data analysis of Tully Sugar Ltd block productivity data detennined the nature and variation of farm perfonnance, variety distribution across farms, crop class distribution across fanns, and the distribution of crop age between farms for 1997. All data were found to be unbalanced and unevenly distributed. The industry currently compares the means of CCS and cane yield for each variety grown in the district without considering confounding factors like age or crop class and ranks them accordingly. This method was compared to one derived from using a fonnal statistical procedure (referred to as a linear mixed model) that accounted for the effects of crop class, fann and time of harvest. When compared to the industry method of ranking varieties for cane yield, the linear mixed model improved the rank of Q138 by 2 places, and Q124 and Q149 by 1 place. The rank of Ql17 declined by 2 positions, while Q120 and Q127 declined by 1 position. For CCS, the linear mixed model improved the rank of Q132 by 3 places, while the rank of Q124 declined by 3 places. To further demonstrate the significance of the bias and confounding effects inherent in block productivity data, the two methods were also used to evaluate the performance of Q152 and Q117 from 1994, when Q152 accounted for less than 1 % of the cane supply, to 1997, when 0152 accounted for 18% of the cane supply. 0117 accounted for about 20% of the supply annually throughout this period. In 1994, the mean perfonnance of 0152, determined using the industry method, suggested the variety had similar CCS and more yield than 0117. However, the linear mixed model method indicated that the yield was higher but the CCS was 0.42 units lower. By 1997, both methods showed that the yield of Q152 was higher than Q117 and the CCS was 0.95 units lower than Q117. Therefore, the mixed model was better able to characterise the perfonnance of Q152 soon after its adoption by growers. This study highlights the value of using advanced statistical techniques to accommodate problems with productivity data and illustrates how these methods can be used to improve the prediction of a new variety's perfonnance across a mill district.