Statistical tools to interpret risks that arise from rare events in host specificity testing

Abstract

Host specificity testing to predict host range is one of the key steps to predicting the risk a biological control agent will present to non-target organisms in the new environment. When host specificity testing data contain discrepancies, or unacceptable levels of uncertainty, it can be difficult for decision-makers to adequately address this uncertainty. To better understand the uncertainty in host specificity testing, we used a range of statistical tools to examine a data set associated with the leaf weevil Cleopus japonicus (Curculionidae), a biological control agent for the weed Buddleja davidii (Buddlejaceae) in New Zealand. Significant uncertainty arose during the early stages of host specificity testing when one C. japonicus larva reared to pupation on a culturally important native plant. Further trials were conducted to evaluate the suitability of C. japonicus as a biological control agent, and despite the uncertainty, C. japonicus was released in New Zealand in 2006, and has since established populations at each release site. However, the possibility of larvae completing their life cycle on the native plant initiated this evaluation of the statistics associated with testing biological control agents. We present results from analyses of the C. japonicus survival data using confidence intervals, equivalence testing, power analyses and survival curves to highlight the appropriateness of each of these tools for interpreting host specificity tests in biological control.