How many choices can your test animal compare effectively? Evaluating a critical assumption of behavioral preference tests

Behavioral choice tests comprise one of the most commonly used experimental designs in ecology. However a critical assumption of these assays, that the outcome is independent of the number of choices, has not been tested explicitly. We developed a methodology for testing this assumption, and discuss how it can be incorporated into experimental design. The model with which we performed this test consisted of an insect herbivore, the gypsy moth Lymantria dispar L., feeding on a clonal host plant, Populus. We established a dose-response feeding gradient by amending leaves of a single age class with defined concentrations of a diterpene, isopimaric acid, that exhibits feeding deterrent properties. We selected various concentrations that elicited different levels of feeding for subsequent tests in which we modified the number of choices. A sample size of 30 assay units per test generated statistically significant separations in two-way choice tests, yielded statistically significant but somewhat inconsistent results when four concentrations were offered, and failed to provide complete separation when five concentrations were offered. Other factors associated with the number of choices that affected results included specific combinations of doses, physical arrangement of choices, and total consumption per assay unit. We used our results to develop procedures for estimating the sample sizes needed to compare a specified number of choices. We based these methods on power considerations, the requirements for data transformation and inclusion of covariates. We develop a general approach for estimating the number of replicates needed to support a particular number of choices for a test organism, and discuss factors to be considered when relating this approach to various types of behavioral choice assays.