Introduction

Ecological experimentation has some characteristics in common with mathematical modeling in ecology, which has been stated by Levins (1968) to face the impossibility of simultaneously maximizing precision, realism, and generality [see a discussion of this problem by Hunt & Doyle (1984)]. Ecologists conducting experiments face analogous choices, in that providing confidence in a relevant ecological process may preclude its general application, or the requirements of a sophisticated experimental design may severely decrease realism, or the use of an elaborate design in the field may put the necessary amount of replication beyond the resources of the investigator. In each case, the ecologist faces a difficult choice. All of the choices should be made deliberately, because to let them be made by default can lead to a misinterpretation or to such an unfortunate effect as replication insufficient to yield a convincing (that is, statistically significant) result.

Generality versus confidence

Conclusions reached from ecological studies increase in value as they can be extended to more and more situations. Therefore, any factors reducing this generality should also incorporate sufficient benefits to offset the loss. Experience over a wide range of habitats might be offset by a thorough knowledge of the ecological processes taking place in one or a few. Anyone undertaking ecological experiments is faced with this dilemma. The experiments are needed to test hypotheses about the processes, but they necessarily limit the results of the tests to one or a few localities.