Empirical Evaluation of Decision Support Systems: Concepts and an Example for Trumpeter Swan Management

Decision support systems are often not empirically evaluated, especially the underlying modelling components. This can be attributed to such systems necessarily being designed to handle complex and poorly structured problems and decision making. Nonetheless, evaluation is critical and should be focused on empirical testing whenever possible. Verification and validation, in combination, comprise such evaluation. Verification is ensuring that the system is internally complete, coherent, and logical from a modelling and programming perspective. Validation is examining whether the system is realistic and useful to the user or decision maker, and should answer the question: “Was the system successful at addressing its intended purpose?” A rich literature exists on verification and validation of expert systems and other artificial intelligence methods; however, no single evaluation methodology has emerged as preeminent. Under some conditions, modelling researchers can test performance against a preselected gold standard. Often in natural resource issues, such a standard does not exist. This is particularly true with near real-time decision support that is expected to predict and guide future scenarios while those scenarios are, in fact, unfolding. When validation of a complete system is impossible for such reasons, examining major components can be substituted, recognizing the potential pitfalls. I provide an example of evaluation of a decision support system for trumpeter swan (Cygnus buccinator) management that I developed using interacting intelligent agents, expert systems, and a queuing model. Predicted swan distributions over a 13 year period were tested against observed numbers. Finding such data sets is key to empirical evaluation.