Replication and Treatment Strength in Whole-Lake Experiments

Replication of large-scale experiments is desirable, but the numbers of rep­licates needed are not known. Costs and feasibility of ecosystem experiments depend critically on the numbers of replicates needed because of the high cost per replicate and the scarcity of experimental ecosystems. This paper examines the numbers of replicates and magnitudes of manipulation needed to detect changes in lake primary productivity resulting from piscivore manipulations. Substantial ( IO x changes in the independent variable, piscivore biomass) and sustained (at least 3-5 yr) manipulations using five ref­erence and five experimental ecosystems produced significant t test results in >80% of simulated experiments. The need for substantial and sustained manipulations is consistent with published results of whole-lake experiments on nutrient inputs, chemical contami­nants, and the biota. In many cases, limited numbers of experimental systems or high costs will prevent adequate replication of ecosystem experiments. When large-scale experiments employ in­sufficient (e.g., 2 or 3) replicates and/or modest perturbations of the independent variate, there is great risk of erroneously accepting the hypothesis of no treatment effect. Therefore, unreplicated paired-system experiments (one reference and one experimental system) are often preferable even though classical statistics cannot be used to determine whether ma­nipulation caused a change in the experimental system. A series of unreplicated paired­system experiments, staggered in time and performed in many locations, will provide more ecological insight than a replicated experiment in a single region. Few statistical methods pertain to large-scale ecological experiments; innovations could be very beneficial.