Using Multiple Treatment Levels as a Means of Improving Inference in Wildlife Research

Abstract

Under the most common experimental protocol, which uses comparison of control and experimental groups subject to a single level of manipulative treatment, conclusions regarding the influences of manipulation are restricted to the presence or absence of an effect. An alternative approach, in which the experimental variable is invoked across multiple treatment levels, should improve biological inference by further clarifying the relationship between the independent and dependent variables. Mathematical analyses comparing relative precision indicate that the experimental design with greater standard deviation in the independent variable will have greater precision. If the precision of 2 experimental designs is similar, or if the statistical power of a given design is high or low, power is changed minimally by modifying the number of treatment levels. Conversely, if the power of 1 experimental design is intermediate, then the alternative design may have substantially more or less power depending on the relative precision of the 2 models. In a review of 134 food supplementation experiments, we found that 95% of these studies were designed with a single treatment level, implying that most of these studies revealed little about the relationship between food abundance and dependent variables. Researchers should evaluate the relative merits of study designs with single or multiple treatment levels a priori, and apply the protocol that is most likely to provide statistical or inferential advantages.