Judith L. Anderson
Choosing a design for an adaptive management experiment presents special challenges. A scientist planning a laboratory study can usually rely on traditional advice for experimental design, secure in the knowledge that major economic decisions are not likely to be based on his study alone, and that any errors will eventually be discovered when his work is replicated. In contrast, an adaptive management experiment may never be repeated, but instead may lead directly to economically significant management decisions. Also, traditional design recommendations alone may not be helpful because manipulating design variables, such as sample size or length of the experiment, often involves prohibitive costs. Finally, various stakeholders may have very different points of view concerning the impact of possible incorrect inferences. For these reasons, we need to look beyond "textbook" experimental design recommendations in adaptive management.
The EDAM models are intended to help scientists and managers ("experimenters") to visualize these complex implications of design choices for adaptive management experiments. Each model represents a management experiment on a forest ecosystem, which can be examined from different stakeholders' points of view. Given a set of design choices, the model demonstrates all possible outcomes of the experiment and their likelihood of occurrence, with a special emphasis on their future economic and ecological impact. The experiments in the EDAM models are thus not primarily statistical exercises; instead, the adaptive management experiment is a 3-stage interaction between people and nature that may stretch far into the future:
Stage 1 -- Design and implementation-- the experimenters plan and carry out a management plan that probes the forest ecosystem experimentally
Stage 2 -- Analysis -- the experimenters learn, i.e., they make an inference (correct or incorrect) about the ecosystem based on data from the experiment
Stage 3 -- Management response -- the experimenters respond to the inference with new management actions, which, as they are projected into the future, will affect both the forest ecosystem and various stakeholders.
There are two versions of the EDAM model: a Before-After-Control-Impact Paired Series (BACIPS) experiment on a landscape scale, and an Analysis of Variance (ANOVA) experiment on a stand scale.
Motivation and scope
The EDAM models were inspired by recent discussion about the importance of statistical power and the costs of different kinds of errors of inference in ecological experimentation and monitoring. They are intended to help clarify the statistical, social, and economic contexts in which planning for adaptive management experiments takes place. I have tried to make the models flexible, so users can adapt them to their own situation and modify various assumptions. However, it is unlikely that the models will represent any particular experimental perfectly. They are intended to supplement, but not to replace, the advice of a statistician.