Published online by Cambridge University Press: 01 June 2009
1. Data from seven winter feeding experiments with dairy cows, using 3×3 latinsquare designs and from two randomized block experiments on the supplementation of summer grazing, were used in a study of some aspects of experimental technique.
2. In the 3×3 trials the use of observations from 2 weeks instead of the last week of each period reduced minimum significant differences by 0·1 lb. milk/cow/day, and 0·06 and 0·02 mean percentage unit of fat and solids-not-fat, respectively.
3. In three out of four experiments no advantage was obtained from recording live weights on more than 1 day at the end of each period.
4. It was shown that the analysis of variance of milk yields expressed as individual differences between yield in the experimental period, and the mean of pre- and post- experimental control periods was considerably more efficient than that of yields in the experimental period, or covariance with yield in the initial control period.
5. The results were discussed in relation to the sources of error involved, and the relative efficiency of various types of experimental design.