Partition of variation for predicting experimental power with a broiler chicken example.

A 1932 editorial in Poultry Science stated that sampling theory, or experimental power, could be useful for "the investigator to know how many … birds to put into each experimental pen." Nevertheless, in the past 90 yr, appropriate experimental power estimates have rarely been applied to research with poultry. To estimate the overall variation and appropriate use of resources with animals in pens, a nested analysis should be conducted. Bird-to-bird and separate pen-to-pen variances were separated for 2 datasets, one from Australia and one from North America. The implications of using variances for birds per pen and pens per treatments are detailed. With 5 pens per treatment, increasing birds per pen from 2 to 4 decreased the SD from 183 to 154, but increasing birds/pen from 100 to 200 only decreased the SD from 70 to 60. With 15 birds per treatment, increasing pens/treatment from 2 to 3 decreased SD from 140 to 126, but increasing pens/treatment from 11 to 12 only decreased the SD from 91 to 89. Choosing the number of birds to include in any study should be based on expectations from historical data and the amount of risk investigators are prepared to accept. Too little replication will not allow relatively small differences to be detected. On the other hand, too much replication is wasteful in terms of birds and resources, and violates the fundamental principles of the ethical use of animals in research. Two general conclusions can be made from this analysis. First, it is very difficult to detect 1% to 3% differences in broiler chicken body weight with only one experiment consistently because of inherent genetic variability. Second, increasing either birds per pen or pens per treatment decreased the SD in a diminishing returns fashion. The example presented here is body weight, of primary importance to production agriculture, but it is applicable whenever a nested design is used (multiple samples from the same bird or tissue, etc.).