Quantifying uncertainty in predictions of invasiveness.
Caley, Peter*,1, Lonsdale, Mark 1, 1 CSIRO Entomology, Canberrra, ACT, Australia
ABSTRACT- Screening models for invasive species are rarely evaluated under the conditions in which they are expected to operate. Namely, the prior or "base-rate" probability of an organism being invasive is typically much lower in practice that the data set for which the screening model was developed. For events that have a low prior probability of occurring, predicted probabilities of occurrence based on the results of imperfect screening tests alone tend to substantially overestimate the true probability of the event occurring. If the prior probability is known, such as is the case when screening for medical conditions, it is straightforward to obtain the revised (posterior) probability of an event occurring by combining the prior probability with the likelihood of the screening test result. However, in contrast to medical screening, where the prior probability of a condition can be estimated with very high precision through complete population census, the prior probability of a deliberately introduced organism becoming invasive is typically low and uncertain at the time of screening. Bootstrapping is a resampling technique which is rapidly gaining acceptance as a robust method for estimating variability of parameter estimates. Combining bootstrapping with a Bayesian prior on the base-rate provides a powerful method for estimating posterior probabilities of invasiveness, and characterizing uncertainty around these probabilities. These estimates enable risk management of deliberate introductions to be better informed. We illustrate the approach as applied to screening for unwanted plants.
Key words: invasive screening, Bayesian, uncertainty, bootstrapping
All materials copyright The Ecological Society of America (ESA), and may not be used without written permission.