Author
 |
Fisher, Dwight |
|
Submitted to: Journal of Animal Science
Publication Type: Peer Reviewed Journal Publication Acceptance Date: 3/3/2000 Publication Date: N/A Citation: N/A Interpretive Summary: Agricultural field experiments must be laid out according to certain constraints in order to produce statistically valid results. As the plants or animals in the experiment are partitioned to treatments there are some fairly common conceptual errors that can be made that make valid statistical comparisons difficult or impossible. For example, the group of animals on a pasture must be replicated on another pasture for comparison because the animals are all handled as a group. Variation among animals on an individual pasture doesn't provide the information needed for comparison of pasture treatments. Grazing trials are very difficult and expensive to conduct so researchers have sought alternatives to traditional replication. However, these alternatives may result in statistical risks that some scientific communities find unacceptable. Each scientific community establishes standards for publication of research. Studies conducted on experimental units that are very large or unique may have no replication but a single replication may be published as a case study for comparison with other similar case studies. Technical Abstract: An experimental unit is the portion of experimental material to which a treatment is applied. Randomization of treatments between experimental units is implied. This simple definition is often overlooked but statistical analyses are predicated on identification of the experimental unit. The experimental unit is the foundation for experimental error, which his the variation among observations of experimental units treated alike. The group of animals on a pasture in a grazing trial generally makes up a single experimental unit. The logistics of conducting a grazing trial can produce a strong desire to redefine the experimental unit when more than one animal grazes each pasture. Regression analysis has been used as an alternative to traditional replication but results in risks that some scientific communities find unacceptable. For example, this issue has resulted in differences of opinion in the analysis and publication of stocking rate studies without replicated experimental units. Each scientific community establishes standards for experimental design and analysis. Studies conducted on experimental units that are very large or unique may have no replication but a single replication may be published as a case study for comparison with other similar case studies. Standards are based in part on the logistics of research in a particular field of science. Less rigorous designs may be needed in experiments that are very difficult to conduct. |
