Improving the power of pasture cultivar trials to discriminate cultivars on the basis of differences in herbage yield

Abstract

A review of 7 recently published perennial ryegrass cultivar trials (from 6 contrasting environments) with data expressed as an aggregate of seasonal harvests (autumn, winter, spring, and summer) revealed that the l.s.d. (P = 0.05) varied between 4 and 255% of the mean herbage yield of the trial in a given season, with 56 of 72 data points having an l.s.d. (P = 0.05) >10% of the trial mean. Power analysis of a perennial ryegrass trial that was conducted at Heywood (Vic.) from 1997 to 1998, with a 16% apparent difference in the total yield of a new synthetic and commercial cultivars, demonstrated that this difference would have only been detected 45% of the time. However, if the number of replications in the trial was increased from 4 to 8, then it was predicted that this difference would have been detected 70% of the time. In response to the data from this experiment, a trial was sown in 1999 that compared 4, 6, and 8 replicates to detect differences in the herbage yield of perennial ryegrass cultivars. In this trial, differences that were detected (P < 0.05) with 8 replicates would have routinely gone undetected when 4 or 6 replicate combinations were used. The use of a row–column design on the 8 replicates of the trial reduced the error variance of the trial by 5–12%, depending on the harvest. It was concluded that current pasture cultivar trials are routinely failing to detect differences between cultivars at an adequate level, given the rates of genetic gain in pasture species. In order to increase this precision, the number of replicates in a trial should be optimised on the basis of past data and the likely difference between control and test cultivars. Once the number of replications in a trial has been optimised then the use of row–column, or nearest neighbour designs, and analysis, will further increase precision for little extra cost.