On Hypothesis Testing in Ecology and Evolution

Theories of causality in ecology and evolution rarely lend themselves to analysis by the formal method of "hypothesis testing" envisioned by champions of a "strong inference" model of scientific method. The objective of biological research typically is to assess the relative contributions of a number of potential causal agents operating simultaneously. Sensibly stated hypotheses in the methodology of most field investigations are similar to hypotheses of applied statistics. They are not intended to be mutually exclusive, in any sense exhaustive, or global in their application. It is not possible in principle to perform a "critical test" or experiment to distinguish between the truth of "alternative hypotheses" if the proposed causal processes they caricature occur simultaneously. We consider several examples in which a rigid hypothetico-deductive methodology applied to nonalternative ecological "hypotheses" could lead to fallacious conclusions. It has been proposed that processes of ecological succession may be separated into alternative modes of "facilitation," "inhibition," and "tolerance." Yet attempts to experimentally reject one or more of the supposedly distinct hypotheses cannot, in principle, distinguish between them in a variety of biologically interesting cases. In studies of the limits of distributions of intertidal organisms, reasonable univariate experimental tests of possible causes would lead to rejection of "biological enemy" hypotheses when a "keystone predator effect" occurs because the interaction between competition and predation reverses the direction of the effect on some prey populations expected from either process in isolation. Particular problems arise when "null models" in ecology are treated as hypotheses of "strong inference." Models of ecological or evolutionary causality rarely have single or easily stated "null" converses. Tractable null models have no probability of being strictly true, and thus may be rejected a priori as hypothetico-deductive constructs. In practice, their role is as a reference point for measurement of departures. Their usefulness in this regard depends upon the reliability with which the characteristics of biology without interaction can be estimated. Applied to studies of interspecific competition through the use of species distributions, purported null hypotheses make different biological assumptions than those of the interactive models. They seem neither especially more reliable nor in any way more fundamental. We see no reason to accept the recent claims that "null hypotheses," as applied in ecology and evolution, have any logical primacy or greater parsimony than other approaches to partitioning the variation observed in natural communities among the contributions of many observable causes. Careful consideration of possible explanations and controlled experimentation contribute a great deal to ecological and evolutionary knowledge. However, we believe that the hypothetico-deductive model of scientific method can provide misleading prescriptions for efficient investigation and acceptance of evidence in phenomena with multiple causes, and should be applied with appropriate skepticism.