Abstract
The low replicability of scientific studies has become an important issue. One possible cause is low representativeness of the experimental design employed. Already in the 1950's, Egon Brunswick pointed out that experimental setups ideally should be based on a random sample of stimuli from the subjects' natural environment or at least include basic features of that environment. Only experimental designs satisfying this criterion, representative designs in Brunswikian terminology, can produce results generalizable beyond the procedure used and to situations outside the laboratory. Such external validity is crucial in preclinical drug studies, for example, and should be important for replicability in general. Popular experimental setups in rodent research on non-human animals, like the tail suspension test or the Geller-Seifter procedure, do not correspond to contexts likely to be encountered in the animals' habitat. Consequently, results obtained in this kind of procedures can be generalized neither to other procedures nor to contexts outside the laboratory. Furthermore, many traditional procedures are incompatible with current notions of animal welfare. An approximation to the natural social and physical context can be provided in the laboratory, in the form of a seminatural environment. In addition to satisfy the basic demands for a representative design, such environments offer a far higher level of animal welfare than the typical small cages. This perspective article will briefly discuss the basic principles of the generalizability of experimental results, the virtues of representative designs and the coincidence of enhanced scientific quality and animal welfare provided by this kind of design.