Nutrition science advances by observation (for example, nutritional epidemiologists observe statistical associations between what people eat and their patterns of health and disease) and by experiment, in which subjects are given specifically formulated diets and certain effects are noted. For example, to test the metabolic effects of components of diets, experiments are performed in which a diet containing the test substance (treatment) is compared against a reference diet (control) which, if possible, is the same in all respects other than the test substance and in which the test substance is substituted by a suitable placebo. Subjects should be allocated randomly to diets so that factors other than those tested, and which may otherwise confound the results, are equalised between treatments. Ideally, neither the subjects nor the experimenters should be aware of the identity of treatment and placebo groups (double-blind).
Source: Biotechnology Glossary