Getting the story straight: laying the foundations for statistical evaluation of the performance of surveillance.

This chapter describes the foundations for statistical evaluation of the performance of surveillance. A 'story', about a conversation between biosecurity and quantitative participants, helps weave together these concepts and make them less abstract. The chapter begins with an overview of the biosecurity questions applicable to quantitative analysis, by defining the types of response variables. This provides a basis for introducing the different statistical modelling paradigms that might be adopted for analysis, such as classical or frequentist hypothesis testing, Bayesian approaches and deterministic modelling. Regardless of paradigm, various objectives of the surveillance programme can be identified, and characterized, as 'seek and destroy', 'maintaining the status quo' or hybrids. The chapter proceeds by addressing the elements of statistical design, requiring a more detailed view of the spatio-temporal context of surveillance: identifying the unit of surveillance, the role of randomization, and issues of extent, scale and sampling effort. With all of this preparation, it is now possible to come to the main purpose of the chapter, to evaluate surveillance. This involves deciding whether diagnostic and/or predictive ability are paramount when quantifying surveillance efficiency and efficacy. To facilitate this, the roles of observation versus the reality of the pest incursion are separated and explained, taking advantage of Bayes' theorem. Finally the chapter and the accompanying story end by focusing on interpretation of surveillance design parameters: How can we describe what it is that we wish to learn from surveillance?