Sample Size for Experiments on Milk Yield

Open Access
Publisher Website
Google Scholar
Abstract
Graphs are presented that will permit dairy researchers to estimate the number of cows required to have 50 or 80% chance of detecting specified differences in true mean milk yield between two or more treat- ment groups. Relatively small experiments are shown to be almost hopelessly insensi- tive unless experimental errors can be re- duced below levels ordinarily encountered. If an experimenter has available fewer than ten cows per treatment, estimates show he has less than a 50% chance of detecting a true mean difference in daily milk yield of 4 kg in a completely randomized experi- ment, 3.5 kg with a covariance analysis, 3.2 kg in a randomized block design, or 1.2 kg in a crossover trial. Researchers universally are faced with the problem of designing experiments which, with the resources available, will properly and ade- quately test pertinent hypotheses. Too often little or no attention is given to the likelihood that a proposed experiment will be adequate to detect the smallest differences that are im- portant. To design experiments that will make efficient use of resources, one must predeter- mine the following: 1. The magnitude of the experimental error