The studies should be carried out at least at two different locations.

The experimental design used must include consideration of adequate statistical power and Type 1 and 2 risks. The protocol must be sufficiently sensitive to detect any effects from the additive at the lowest recommended dose (Type 1 α risk, P ≤ 0,05 in general and P ≤ 0,1 for ruminants, minor species, pets and non-food producing animals) and of sufficient statistical power to guarantee that the experimental protocol meets the study objective. The Type 2 β risk shall be lower than or equal to 20 % in general, and 25 % for experiments with ruminants, minor species, pets and non-food producing animals, hence a power (1-β) greater than or equal to 80 % (75 % for ruminants, minor species, pets and non food producing animals).

It is recognised that the nature of some additives make it difficult to define experimental conditions under which optimal results may be achieved. Consequently, the possibility of using meta-analysis shall be considered when the number of trials available is greater than three. For this reason, similar protocol designs shall be used for all trials so that data can eventually be tested for homogeneity and pooled (if tests so indicate) for statistical evaluation at a level of P ≤ 0,05.