From Design to Analysis: Effective Statistical Approaches for Host Range Testing

The major goal of host range testing in biological control is to minimize the probability that released biological control agents have unwanted effects on populations of non-target hosts. This leads to a non-trivial problem in statistical hypothesis testing, since the standard approach in statistical tests is to ask whether or not an effect – in this case acceptance of a nontarget host – exists and to attribute a precise probability to err only with rejecting the null hypothesis that assumes no effect. The problem is that it is difficult to assign a probability with accepting the null hypothesis of no effect, i.e., that the biological control agent does not include a given non-target insect into its host range. Yet, this piece of information is exactly what we need for high precision and confidence. Confidence in this respect increases with sample size and the statistical effect size, i.e., the difference from the null hypothesis that is considered biologically meaningful. However, sample size is often limited due to limitations in test subjects, research money, and space for testing arenas. Consequently, there is a high premium on using a very good experimental design and employing the most powerful statistical approach available. This paper discusses common problems with experimental designs, emphasizes the necessity to decide on the statistical effect size that is biologically meaningful, points towards the need to determine the statistical power of the host range test employed, and provides an overview about powerful statistical approaches for analyzing experiments on the host range of potential biological control agents.