SCHANK, Jeffrey: Is Pseudoreplication a Pseudoproblem?

In 1984, Stuart Hurlbert introduced the problem of pseudoreplication as serious and ubiquitous problem in ecological research design. Since then, the view that pseudoreplication is a serious research problem has spread to other fields including animal behavior. Pseudoreplication is best characterized as having two main concerns. First, the relationship of data samples in space and time can lead to their non-independence. Second, the sampling space of an experimental design often does not match the scope of the hypothesis tested. I argue that both concerns derive from misunderstandings of hypothesis testing, experimental controls, randomization, statistical independence, and the mathematics of statistical inference. Indeed, pseudoreplication is an amalgamation of ideas, some of which are problems and others are not, but none of these ideas have much to do with replication. Perhaps not surprisingly, simple rules that have been recommended for avoiding pseudoreplication–such as averaging over samples, systematically interspersed designs, and never pool data–may lead to serious statistical and methodological errors. Flaws in experimental studies are often not easily discovered and we should always avoid simple rules that either obscure flaws (e.g., averaging data samples) or that can lead to flaws (e.g., always intersperse controls and treatments). I conclude that pseudoreplication is so fundamentally flawed as guiding principle of research design that it should be immediately abandoned as a criterion for evaluating experimental studies.